21.02.2013 Views

Noise in AlGaN/GaN HEMTs and Oscillators - Microwave Electronics ...

Noise in AlGaN/GaN HEMTs and Oscillators - Microwave Electronics ...

Noise in AlGaN/GaN HEMTs and Oscillators - Microwave Electronics ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Committee <strong>in</strong> charge:<br />

UNIVERSITY of CALIFORNIA<br />

Santa Barbara<br />

<strong>Noise</strong> of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> <strong>Oscillators</strong><br />

A dissertation submitted <strong>in</strong> partial satisfaction of the<br />

requirements for the degree of<br />

Professor Robert A. York, Chair<br />

Professor Umesh K. Mishra<br />

Professor Mark J. Rodwell<br />

Dr. Yifeng Wu<br />

Doctor of Philosophy<br />

<strong>in</strong><br />

Electrical <strong>and</strong> Computer Eng<strong>in</strong>eer<strong>in</strong>g<br />

by<br />

Christopher Sanabria<br />

June 2006


The dissertation of Christopher Sanabria is approved:<br />

Chair<br />

June 2006


<strong>Noise</strong> of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> <strong>Oscillators</strong><br />

Copyright c○ 2006<br />

by<br />

Christopher Sanabria<br />

iii


EDUCATION<br />

Curriculum Vitæ<br />

Christopher Sanabria<br />

Bachelor of Science <strong>in</strong> Electrical Eng<strong>in</strong>eer<strong>in</strong>g, Magna Cum Laude, University of<br />

Notre Dame, May 2001.<br />

Master of Science <strong>in</strong> Electrical <strong>and</strong> Computer Eng<strong>in</strong>eer<strong>in</strong>g, University of California,<br />

Santa Barbara, December 2002.<br />

Doctor of Philosophy <strong>in</strong> Electrical Eng<strong>in</strong>eer<strong>in</strong>g, University of California, Santa Barbara,<br />

June 2006.<br />

PROFESSIONAL EMPLOYMENT<br />

May 1998 - August 1998, Intern, Delphi-Delco <strong>Electronics</strong>, Kokomo, IN.<br />

May 1999 - August 1999, Intern, Delphi-Delco <strong>Electronics</strong>, Kokomo, IN.<br />

May 2000 - August 2000, Intern, Texas Instruments, Houston, TX.<br />

September 2001 - March 2002, Teach<strong>in</strong>g Assistant, Department of Electrical <strong>and</strong><br />

Computer Eng<strong>in</strong>eer<strong>in</strong>g, University of California, Santa Barbara.<br />

June 2003 - September 2003, Intern, Agilent Labs, Agilent Technologies, Palo Alto,<br />

CA.<br />

March 2002 - May 2006, Research assistant, Department of Electrical <strong>and</strong> Computer<br />

Eng<strong>in</strong>eer<strong>in</strong>g, University of California, Santa Barbara.<br />

iv


PUBLICATIONS<br />

S. Gao, C. Sanabria, H. Xu, S. Heikman, U. K. Mishra <strong>and</strong> R. A. York, MMIC Class-<br />

F Power Amplifiers us<strong>in</strong>g Field-Plated <strong>GaN</strong> <strong>HEMTs</strong>, accepted IEEE Proceed<strong>in</strong>gs on<br />

<strong>Microwave</strong>, Antennas <strong>and</strong> Propagation, 2006.<br />

C. Sanabria, A. Chakraborty, H. Xu, M. J. Rodwell, U. K. Mishra, <strong>and</strong> R. A. York,<br />

The Effect of Gate Leakage on the <strong>Noise</strong> Figure of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>, IEEE Electron<br />

Device Letters, January 2006, pp. 19-21.<br />

C. Sanabria, H. Xu, A. Chakraborty, M. J. Rodwell, U. K. Mishra, <strong>and</strong> R. A. York,<br />

<strong>Noise</strong> Figure Measurements <strong>and</strong> Model<strong>in</strong>g of Field-Plated <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>, International<br />

Conference on Nitride Semiconductors, Bremen, Germany, August 2005.<br />

C. Sanabria, H. Xu, S. Heikman, U. K. Mishra, R. A. York, A <strong>GaN</strong> Differential Oscillator<br />

with Improved Harmonic Performance, IEEE <strong>Microwave</strong> <strong>and</strong> Wireless Components<br />

Letters, July 2005, pp. 463-465.<br />

H. Xu, C. Sanabria, S. Heikman, S. Keller, U. K. Mishra, <strong>and</strong> R. A. York, High<br />

Power <strong>GaN</strong> <strong>Oscillators</strong> us<strong>in</strong>g Field-Plated HEMT Structure, IEEE <strong>Microwave</strong> Theory<br />

<strong>and</strong> Technique International <strong>Microwave</strong> Symposium, June 2005.<br />

C. Sanabria, H. Xu, T. Palacios, A. Chakraborty, S. Heikman, U. K. Mishra, R. A.<br />

York, Influence of Epitaxial Structure <strong>in</strong> the <strong>Noise</strong> Figure of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,<br />

IEEE <strong>Microwave</strong> Theory <strong>and</strong> Technique Transactions, Vol. 53, February 2005, pp.<br />

762-769.<br />

H. Xu, C. Sanabria, Y. Wei, S. Heikman, S. Keller, U. K. Mishra, <strong>and</strong> R. A. York,<br />

Characterization of two field-plated <strong>GaN</strong> HEMT structures, IEEE Topical Workshop<br />

on Power Amplifiers for Wireless Communications, September 2004.<br />

H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, Y. Wei, S. Heikman, S. Keller, U. K. Mishra <strong>and</strong> R. A.<br />

York, A new field-plated <strong>GaN</strong> HEMT structure with improved power <strong>and</strong> noise performance,<br />

IEEE Lester Eastman Conference on High Performance Devices, August<br />

2004.<br />

H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, S. Keller, U. K. Mishra, <strong>and</strong> R. A. York, A C-b<strong>and</strong><br />

high-dynamic range <strong>GaN</strong> HEMT low-noise amplifier, IEEE <strong>Microwave</strong> <strong>and</strong> Wireless<br />

Components Letters, Vol. 14, June 2004, pp. 262 264.<br />

v


C. Sanabria, H. Xu, T. Palacios, A. Chakraborty, S. Heikman, U. K. Mishra, R. A.<br />

York, Influence of the Heterostructure Design on <strong>Noise</strong> Figure of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,<br />

Device Research Conference, June 2004.<br />

H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, S. Keller, U. K. Mishra, <strong>and</strong> R. A. York, Robust Cb<strong>and</strong><br />

MMIC Low <strong>Noise</strong> Amplifier us<strong>in</strong>g <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT Power Devices, 8th Wide-<br />

B<strong>and</strong>gap III-Nitride Workshop, September 2003.<br />

vi


Abstract<br />

<strong>Noise</strong> of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> <strong>Oscillators</strong><br />

by<br />

Christopher Sanabria<br />

<strong>GaN</strong> <strong>HEMTs</strong> will likely become the solid-state device of choice for power <strong>in</strong> mi-<br />

crowave <strong>and</strong> millimeter-wave circuits. These products, such as base stations <strong>and</strong> other<br />

communication systems, tend to be space-constra<strong>in</strong>ed. Hence solutions cont<strong>in</strong>uously<br />

move from a hybrid (circuit board plus components) approach to a microwave mono-<br />

lithic <strong>in</strong>tegrated circuit (MMIC). To be successful <strong>in</strong> a MMIC design, <strong>GaN</strong> will have<br />

to perform well <strong>in</strong> other areas besides power. One of the most crucial metrics of a<br />

system is its noise. The noise of <strong>GaN</strong> devices <strong>and</strong> circuits has only been critically<br />

exam<strong>in</strong>ed <strong>in</strong> the last five years.<br />

This work will <strong>in</strong>vestigate several aspects of the noise performance of <strong>GaN</strong> <strong>HEMTs</strong>.<br />

Measurements of noise figure (NF) <strong>and</strong> low-frequency noise (LFN) are used to char-<br />

acterize devices. Model<strong>in</strong>g useful for calculations <strong>and</strong> circuit simulation are applied,<br />

with some <strong>in</strong>troduced. Several studies of NF <strong>and</strong> LFN are presented. Some confirm<br />

or challenge previous publications while others are new observations. Two differen-<br />

tial oscillators were built to characterize the phase noise. As it is believed that <strong>GaN</strong><br />

vii


<strong>HEMTs</strong> will replace GaAs <strong>HEMTs</strong> <strong>in</strong> various applications, the NF, LFN, <strong>and</strong> phase<br />

noise of the two are compared.<br />

viii


Contents<br />

List of Figures xii<br />

List of Tables xvi<br />

1 Introduction 1<br />

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1<br />

1.2 Literature Review of <strong>Noise</strong> <strong>in</strong> <strong>GaN</strong> <strong>HEMTs</strong> . . . . . . . . . . . . . . 5<br />

1.3 Thesis Outl<strong>in</strong>e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8<br />

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9<br />

2 <strong>Noise</strong> Figure Model<strong>in</strong>g of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> 12<br />

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12<br />

2.2 <strong>Noise</strong> Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13<br />

2.2.1 Thermal <strong>Noise</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 13<br />

2.2.2 Shot<strong>Noise</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . 15<br />

2.2.3 Other Sources of <strong>Noise</strong> . . . . . . . . . . . . . . . . . . . . . 16<br />

2.3 Equivalent Circuit Model . . . . . . . . . . . . . . . . . . . . . . . . 17<br />

2.4 <strong>Noise</strong> Figure <strong>and</strong> <strong>Noise</strong> Parameters . . . . . . . . . . . . . . . . . . . 21<br />

2.5 HEMT <strong>Noise</strong> Figure Models . . . . . . . . . . . . . . . . . . . . . . 25<br />

2.5.1 van der Ziel <strong>and</strong> Pucel Models . . . . . . . . . . . . . . . . . 25<br />

2.5.2 Fukui Model . . . . . . . . . . . . . . . . . . . . . . . . . . 28<br />

2.5.3 Pospieszalski Model . . . . . . . . . . . . . . . . . . . . . . 29<br />

2.5.4 Pospieszalski <strong>and</strong> Correlated <strong>Noise</strong> Models Applied to Al-<br />

<strong>GaN</strong>/<strong>GaN</strong> <strong>HEMTs</strong> . . . . . . . . . . . . . . . . . . . . . . . 31<br />

2.6 A Proposed <strong>Noise</strong> Figure Model . . . . . . . . . . . . . . . . . . . . 35<br />

2.6.1 Setup Details . . . . . . . . . . . . . . . . . . . . . . . . . . 36<br />

2.6.2 Derivation of <strong>Noise</strong> Parameters . . . . . . . . . . . . . . . . 39<br />

2.6.3 Derivation of Dra<strong>in</strong> <strong>Noise</strong> Source . . . . . . . . . . . . . . . 44<br />

2.6.4 <strong>Noise</strong> Parameter Scal<strong>in</strong>g . . . . . . . . . . . . . . . . . . . . 49<br />

2.6.5 Discussion of the Model . . . . . . . . . . . . . . . . . . . . 53<br />

2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57<br />

ix


References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57<br />

3 <strong>Noise</strong> Figure Measurements <strong>and</strong> Studies 60<br />

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60<br />

3.2 Device Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61<br />

3.3 <strong>Noise</strong> Figure Measurement Setup <strong>and</strong> Method . . . . . . . . . . . . . 62<br />

3.4 Bias Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68<br />

3.5 <strong>GaN</strong> HEMT <strong>Noise</strong> Figure Studies . . . . . . . . . . . . . . . . . . . 75<br />

3.5.1 Substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75<br />

3.5.2 Al Composition <strong>in</strong> the Barrier . . . . . . . . . . . . . . . . . 76<br />

3.5.3 AlN Interlayer . . . . . . . . . . . . . . . . . . . . . . . . . 78<br />

3.5.4 Gate Leakage . . . . . . . . . . . . . . . . . . . . . . . . . . 81<br />

3.5.5 Field-Plated Devices . . . . . . . . . . . . . . . . . . . . . . 84<br />

3.5.6 Thick-Epitaxial Cap Devices . . . . . . . . . . . . . . . . . . 91<br />

3.6 Comparison of High-Performance <strong>GaN</strong> <strong>HEMTs</strong> to Other Material Systems<br />

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />

3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95<br />

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96<br />

4 Low-Frequency <strong>Noise</strong> of <strong>GaN</strong> <strong>HEMTs</strong> 101<br />

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101<br />

4.2 Review of Low-Frequency <strong>Noise</strong> . . . . . . . . . . . . . . . . . . . . 102<br />

4.3 Low-Frequency <strong>Noise</strong> Setup . . . . . . . . . . . . . . . . . . . . . . 107<br />

4.4 <strong>GaN</strong> HEMT Low-Frequency <strong>Noise</strong> Model<strong>in</strong>g . . . . . . . . . . . . . 113<br />

4.5 <strong>GaN</strong> HEMT Low-Frequency <strong>Noise</strong> Studies . . . . . . . . . . . . . . 119<br />

4.5.1 Substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119<br />

4.5.2 Passivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 119<br />

4.5.3 Thick-Epitaxial Cap Devices . . . . . . . . . . . . . . . . . . 121<br />

4.5.4 Field-Plated Devices . . . . . . . . . . . . . . . . . . . . . . 122<br />

4.6 Comparison to GaAs <strong>HEMTs</strong> . . . . . . . . . . . . . . . . . . . . . . 123<br />

4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125<br />

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126<br />

5 <strong>GaN</strong> HEMT Based <strong>Oscillators</strong> 128<br />

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128<br />

5.2 Concern<strong>in</strong>g Phase <strong>Noise</strong> . . . . . . . . . . . . . . . . . . . . . . . . 129<br />

5.3 MMIC Process Description . . . . . . . . . . . . . . . . . . . . . . . 134<br />

5.4 Differential <strong>Oscillators</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 135<br />

5.4.1 High L<strong>in</strong>earity Oscillator . . . . . . . . . . . . . . . . . . . . 135<br />

5.4.2 Low-Phase <strong>Noise</strong> Oscillator . . . . . . . . . . . . . . . . . . 140<br />

5.5 Comparison to Other <strong>Oscillators</strong> . . . . . . . . . . . . . . . . . . . . 143<br />

x


5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146<br />

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147<br />

6 Summary, Conclusions, <strong>and</strong> Future Directions 149<br />

6.1 Summary <strong>and</strong> Conclusions . . . . . . . . . . . . . . . . . . . . . . . 149<br />

6.2 FuturePaths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151<br />

A ADS Files 154<br />

A.1 Small-Signal Extraction . . . . . . . . . . . . . . . . . . . . . . . . . 155<br />

A.2 <strong>Noise</strong> Figure Simulation . . . . . . . . . . . . . . . . . . . . . . . . 158<br />

A.3 Correlated <strong>Noise</strong> Model Extraction . . . . . . . . . . . . . . . . . . . 160<br />

B Matlab Code for <strong>Noise</strong> Parameter Model<strong>in</strong>g 163<br />

C Matlab Code for Pospieszalski <strong>Noise</strong> Parameter Model<strong>in</strong>g 166<br />

xi


List of Figures<br />

1.1 Cartoon of a very simple transmitter. . . . . . . . . . . . . . . . . . . 2<br />

1.2 Cartoon of a CDMA-like spectrum with four channels. . . . . . . . . 3<br />

2.1 Cartoon show<strong>in</strong>g the device small-signal model on a cross section of<br />

a HEMT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />

2.2 Equivalent model of a transistor driven by a noisy source of impedance<br />

Zsource. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22<br />

2.3 <strong>Noise</strong> <strong>and</strong> ga<strong>in</strong> circles on a Smith Chart. . . . . . . . . . . . . . . . . 24<br />

2.4 Pucel noise model <strong>in</strong> a small-signal circuit. . . . . . . . . . . . . . . 27<br />

2.5 Pospieszalski noise model <strong>in</strong> a small-signal circuit. . . . . . . . . . . 29<br />

2.6 Comparison of Correlated <strong>Noise</strong> <strong>and</strong> Pospieszalski models to measured<br />

noise parameters versus frequency. . . . . . . . . . . . . . . . . 34<br />

2.7 A simplified HEMT circuit model <strong>in</strong>clud<strong>in</strong>g noise sources. . . . . . . 36<br />

2.8 Cartoon show<strong>in</strong>g the effect of source degeneration. . . . . . . . . . . 38<br />

2.9 Circuit model used for deriv<strong>in</strong>g noise figure. . . . . . . . . . . . . . . 39<br />

2.10 Cartoon used for deriv<strong>in</strong>g the channel noise. . . . . . . . . . . . . . . 47<br />

2.11 Variation <strong>in</strong> Γ for different dra<strong>in</strong> <strong>and</strong> gate voltages. . . . . . . . . . . 49<br />

2.12 <strong>Noise</strong> parameters versus total gate width. . . . . . . . . . . . . . . . 51<br />

2.13 <strong>Noise</strong> parameters versus number of gate f<strong>in</strong>gers for a constant total<br />

gatewidth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52<br />

2.14 <strong>Noise</strong> parameters predicted with the proposed model <strong>and</strong> compared to<br />

measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53<br />

2.15 Relative contributions of different noise sources to the overall noise<br />

figure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />

3.1 Typical epitaxial structures for the devices <strong>in</strong> this work. . . . . . . . . 61<br />

3.2 Schematic of the source-pull noise figure setup. . . . . . . . . . . . . 63<br />

3.3 Coplanar waveguide attenuator. . . . . . . . . . . . . . . . . . . . . . 65<br />

3.4 <strong>Noise</strong> factor, fτ <strong>and</strong> fmax for devices from different samples versus<br />

current. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65<br />

xii


3.5 Variation <strong>in</strong> expected m<strong>in</strong>imum noise figure with changes <strong>in</strong> three<br />

small-signal parameters. . . . . . . . . . . . . . . . . . . . . . . . . 68<br />

3.6 Change <strong>in</strong> noise parameters with dra<strong>in</strong>-source voltage. . . . . . . . . 69<br />

3.7 Typical plots of the noise parameters versus dra<strong>in</strong> source current with<br />

Correlated <strong>Noise</strong> <strong>and</strong> Pospieszalski models. . . . . . . . . . . . . . . 71<br />

3.8 <strong>Noise</strong> variables for the Pospieszalski <strong>and</strong> a CN noise model versus<br />

dra<strong>in</strong>-source current. . . . . . . . . . . . . . . . . . . . . . . . . . . 73<br />

3.9 M<strong>in</strong>imum noise figure, small signal associated <strong>and</strong> maximum ga<strong>in</strong> for<br />

devices on sapphire <strong>and</strong> SiC substrates. . . . . . . . . . . . . . . . . 75<br />

3.10 <strong>Noise</strong> parameters versus frequency for devices of different alum<strong>in</strong>um<br />

composition <strong>in</strong> the barrier. . . . . . . . . . . . . . . . . . . . . . . . 77<br />

3.11 M<strong>in</strong>imum noise figure of samples with different alum<strong>in</strong>um composition<br />

<strong>in</strong> the barrier at vary<strong>in</strong>g dra<strong>in</strong>-source current. . . . . . . . . . . . 78<br />

3.12 M<strong>in</strong>imum noise figure, fτ, <strong>and</strong> fmax versus dra<strong>in</strong>-source current for a<br />

sample with <strong>and</strong> without an AlN <strong>in</strong>terlayer. . . . . . . . . . . . . . . 79<br />

3.13 (a) Associated <strong>and</strong> maximum ga<strong>in</strong> <strong>and</strong> (b) source resistance for devices<br />

with <strong>and</strong> without an AlN <strong>in</strong>terlayer at different applied currents. 80<br />

3.14 (a) M<strong>in</strong>imum noise figure, (b) device associated ga<strong>in</strong>, <strong>and</strong> maximum<br />

ga<strong>in</strong> versus frequency for devices with different gate leakage currents 82<br />

3.15 Simulated (l<strong>in</strong>e) <strong>and</strong> measured (crosses) noise parameters for devices<br />

with different gate leakage currents. . . . . . . . . . . . . . . . . . . 83<br />

3.16 fτ <strong>and</strong> fmax of devices with different field-plate lengths. . . . . . . . . 84<br />

3.17 <strong>Noise</strong> parameters versus frequency for devices with field plates of differentlength.<br />

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85<br />

3.18 Typical change <strong>in</strong> gate leakage for devices of <strong>in</strong>creas<strong>in</strong>g field-plate<br />

length. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87<br />

3.19 Electric field profile for a device with <strong>and</strong> without a field plate. . . . . 87<br />

3.20 Small-signal parameters that change with a field plate. . . . . . . . . 89<br />

3.21 M<strong>in</strong>imum noise figure versus gate width for devices with <strong>and</strong> without<br />

a longfieldplate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90<br />

3.22 M<strong>in</strong>imum noise figure of the field-plated devices at different measurement<br />

frequencies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90<br />

3.23 <strong>Noise</strong> parameters versus dra<strong>in</strong>-source current of a thick cap device<br />

(triangles) <strong>and</strong> a st<strong>and</strong>ard HEMT (squares). . . . . . . . . . . . . . . 92<br />

3.24 M<strong>in</strong>imum noise figure of two 0.15 µm gate length transistors provided<br />

by Tomás Palacios. . . . . . . . . . . . . . . . . . . . . . . . . . . . 95<br />

4.1 Sketch of the key features of low-frequency noise. . . . . . . . . . . . 103<br />

4.2 Variation of α with (a) dra<strong>in</strong>-source voltage bias <strong>and</strong> (b) frequency of<br />

extraction for two devices on the same sample. . . . . . . . . . . . . 106<br />

xiii


4.3 Measured noise floor of the HP 3561A DSA only <strong>and</strong> with the SRS<br />

SR560 LNA (short-circuited <strong>in</strong>put). . . . . . . . . . . . . . . . . . . 109<br />

4.4 Schematic of the setup used for device dra<strong>in</strong>-side low-frequency noise<br />

measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110<br />

4.5 A typical low-frequency plot. . . . . . . . . . . . . . . . . . . . . . . 112<br />

4.6 Plots of the measured dra<strong>in</strong> low-frequency noise with (a) change <strong>in</strong><br />

dra<strong>in</strong>-source current <strong>and</strong> (b) voltage. . . . . . . . . . . . . . . . . . . 114<br />

4.7 Measured gate low-frequency noise versus gate-source voltage (<strong>and</strong><br />

current). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115<br />

4.8 Change <strong>in</strong> low-frequency noise with gate width at various decade frequencies<br />

for three devices. . . . . . . . . . . . . . . . . . . . . . . . 116<br />

4.9 Change <strong>in</strong> low-frequency noise with gate length. . . . . . . . . . . . . 116<br />

4.10 Proposed low-frequency noise model<strong>in</strong>g of the HEMT with a gate <strong>and</strong><br />

dra<strong>in</strong> noise source. . . . . . . . . . . . . . . . . . . . . . . . . . . . 117<br />

4.11 Measurement of devices on a sapphire <strong>and</strong> SiC substrate at a bias of<br />

Vds 5V,Ids 30mA . . . . . . . . . . . . . . . . . . . . . . . . . . . 120<br />

4.12 (a) Low-frequency noise of a device before <strong>and</strong> after passivation. (b)<br />

Low-frequency noise at 10 Hz <strong>and</strong> 1 kHz of a device before <strong>and</strong> after<br />

passivation at different Vgs . . . . . . . . . . . . . . . . . . . . . . . . 120<br />

4.13 Comparison of st<strong>and</strong>ard passivated <strong>HEMTs</strong> to an unpassivated thick<br />

cap <strong>HEMTs</strong>. Bias is Vds =5V<strong>and</strong>Ids = 30 mA. . . . . . . . . . . . . 122<br />

4.14 Low-frequency noise of field-plated devices. . . . . . . . . . . . . . . 123<br />

4.15 Low-frequency noise comparison of <strong>GaN</strong> <strong>and</strong> GaAs <strong>HEMTs</strong>. . . . . . 124<br />

5.1 Examples of typical phase noise plots. . . . . . . . . . . . . . . . . . 131<br />

5.2 Circuit schematic of the oscillator (bias<strong>in</strong>g not shown). . . . . . . . . 136<br />

5.3 Photograph of the high l<strong>in</strong>earity oscillator. . . . . . . . . . . . . . . . 136<br />

5.4 Measurements of the oscillator: (a) power spectrum (b) frequency<br />

pull<strong>in</strong>g. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138<br />

5.5 Output power, second harmonic power, <strong>and</strong> efficiency of the highl<strong>in</strong>earity<br />

oscillator. . . . . . . . . . . . . . . . . . . . . . . . . . . . 138<br />

5.6 Circuit schematic of the low-phase noise oscillator (bias<strong>in</strong>g not shown). 140<br />

5.7 Photograph of the low-phase noise oscillator. . . . . . . . . . . . . . 141<br />

5.8 Measured phase noise of the oscillator. . . . . . . . . . . . . . . . . . 142<br />

5.9 Phase noise at 100 kHz <strong>and</strong> 1 MHz offsets versus dra<strong>in</strong>-source bias<br />

for a few oscillators. . . . . . . . . . . . . . . . . . . . . . . . . . . . 142<br />

5.10 Relative comparison of <strong>GaN</strong> oscillator to a typical oscillator with low<br />

phase noise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146<br />

A.1 First page of template small signal parameter extraction.dds. . . . . 156<br />

xiv


A.2 Schematic used for simulat<strong>in</strong>g S-parameters of the small-signal circuit<br />

<strong>and</strong> for optimization. . . . . . . . . . . . . . . . . . . . . . . . . . . 158<br />

A.3 Schematic used for simulat<strong>in</strong>g noise parameters. . . . . . . . . . . . . 159<br />

A.4 Schematic used for extract<strong>in</strong>g correlated noise model noise variables. 162<br />

A.5 Data display used for extract<strong>in</strong>g correlated noise model noise variables. 162<br />

xv


List of Tables<br />

2.1 Extracted small-signal parameters for various samples. Devices have<br />

a gate geometry of 0.7 × 150 µm. . . . . . . . . . . . . . . . . . . . 21<br />

2.2 Comparison of Pospieszalski <strong>and</strong> Correlated <strong>Noise</strong> models to measured<br />

data. Devices have a gate geometry of 0.7 × 150 µm . . . . . . 33<br />

2.3 Comparison of the various noise models’ <strong>in</strong>put <strong>and</strong> output noise currents. 56<br />

3.1 M<strong>in</strong>imum noise figure for devices <strong>in</strong> many technologies. . . . . . . . 94<br />

5.1 Comparison of <strong>GaN</strong> oscillators from this <strong>and</strong> other works to oscillators<br />

<strong>in</strong> other materials (FET, MMICs, only). . . . . . . . . . . . . . . 144<br />

xvi


Acknowledgements<br />

BEHIND the c<strong>and</strong>y Santa Barbara exterior of sun, perfect weather, surf<strong>in</strong>g, <strong>and</strong><br />

tourists is a university that is a powerhouse of research that I didn’t have the<br />

foggiest idea existed until I decided to come to graduate school. I’ve been impressed<br />

with the faculty, facilities, labs, peers, <strong>and</strong> the graduate sciences <strong>and</strong> eng<strong>in</strong>eer<strong>in</strong>g programs<br />

at UCSB. It has made for a great research experience <strong>and</strong> I want to thank <strong>and</strong><br />

acknowledge all those who have helped me along the way toward gett<strong>in</strong>g my Ph.D.<br />

I would first like to thank my advisor Professor Robert York. Not only did he<br />

take me <strong>in</strong>, he has provided (along with my co-advisor) five years of constant, un<strong>in</strong>terrupted,<br />

f<strong>in</strong>ancial support. Watch<strong>in</strong>g peers struggle with fund<strong>in</strong>g shortfalls <strong>and</strong><br />

chang<strong>in</strong>g projects has made me realize what a bless<strong>in</strong>g fund<strong>in</strong>g is. I would also like<br />

to thank him for giv<strong>in</strong>g me a great deal of freedom. This was frighten<strong>in</strong>g for me at<br />

first as I thought I would become lost, but later I realized what research I wanted to do<br />

<strong>and</strong> was able to do it without restra<strong>in</strong>t. Professor York has always steered me <strong>in</strong> the<br />

right direction on important matters of publications, research, <strong>and</strong> th<strong>in</strong>gs <strong>in</strong> general.<br />

We had many good, frank, talks that I enjoyed.<br />

I thank my co-advisor Professor Umesh Mishra for many th<strong>in</strong>gs. The first is the<br />

amusement of watch<strong>in</strong>g someone juggle the work of five men like some god with 10<br />

arms. Yet, when I needed to discuss work, we always got a few m<strong>in</strong>utes <strong>in</strong> <strong>and</strong> I never<br />

felt pushed away. I also salute the <strong>GaN</strong> program he has put together at UCSB. Umesh<br />

is one of the smartest, <strong>and</strong> slyest, <strong>in</strong>dividuals I have met <strong>and</strong> it was fun to be around<br />

for the show. I also really appreciate the happy hours he would host.<br />

My other committee members, Professor Mark Rodwell <strong>and</strong> Dr. Yifeng Wu, have<br />

been a great help to me not only <strong>in</strong> prepar<strong>in</strong>g the thesis but <strong>in</strong> important research<br />

discussions over the years. Professor Rodwell’s class notes on noise <strong>and</strong> exchanges<br />

have steered me through to the basic truths about noise figure. With Dr. Wu I’ve had<br />

many conversations: load-pull, noise, circuits, <strong>and</strong> even high-end audio equipment. I<br />

have enjoyed it all. I also thank Professor Long for the conversations that we have<br />

had.<br />

My peers <strong>in</strong> the trenches have been of tremendous help toward this work. Hongtao<br />

Xu taught me his fabrication techniques for MMICs <strong>and</strong> offered very useful suggestions<br />

for my microwave circuits. We worked collectively on several projects <strong>and</strong> became<br />

friends. Tomás Palacios, the wonder Spaniard, has been of tremendous help. His<br />

enormous curiosity, knowledge, <strong>and</strong> keen eye helped me out several times by po<strong>in</strong>t<strong>in</strong>g<br />

me <strong>in</strong> the right direction or giv<strong>in</strong>g me <strong>in</strong>spiration that lead to useful research. My<br />

group mates (past <strong>and</strong> present), Nadia, Val, Raj, Jaehoon, Jiwei, Just<strong>in</strong>, Paolo (e-lo?),<br />

Vicki, Conrad<strong>in</strong>, Jim, <strong>and</strong> Pengcheng have helped me out many a time (or helped provide<br />

a refresh<strong>in</strong>g work break). The small army known as Professor Mishra’s group<br />

have also helped a great deal: Rob, Siddarth, Likun (Mona), Mike, Dario, Arpan, Ale,<br />

Naiqian, Sten, Lee M., Ilan, Yuvaraj, Felix, Pei Yi, Eric, Chang, Chris, Jeff, Karl, <strong>and</strong><br />

xvii


Man Hoi. I would especially like to thank those who grew my material for devices <strong>and</strong><br />

circuits, Arpan, Sten, Stacia, <strong>and</strong> Nick, without whom I would have had no project.<br />

I also had useful discussions with students <strong>in</strong> the groups of both Professors Rodwell<br />

<strong>and</strong> Long. I had many conversations, <strong>in</strong>teractions, <strong>and</strong> chats with Vikas <strong>and</strong> Joe,<br />

whose projects focused heavily on phase noise. Vikas also provided the GaAs devices<br />

that were measured <strong>in</strong> this work <strong>and</strong> both of them helped with equipment <strong>and</strong> setups.<br />

Nav<strong>in</strong>, Zack, <strong>and</strong> Paidi all gave me some useful <strong>in</strong>formation at one po<strong>in</strong>t or another.<br />

Also, I had discussions with Adam about low-frequency noise <strong>and</strong> Professor Elliot<br />

Brown let me borrow some equipment from his personal collection for a measurement.<br />

I thank Agilent <strong>and</strong> Maury for their technical support on various pieces of equipment<br />

(I called a lot). I am greatful to the cleanroom staff (Jack, Bob, Neil, Tom,<br />

Brian T., Bill, Don, Mike, Luis <strong>and</strong> N<strong>in</strong>g) who kept our excellent facilities go<strong>in</strong>g 24x7<br />

<strong>and</strong> for be<strong>in</strong>g available for the occasional even<strong>in</strong>g crisis. The assistants over the year<br />

(Masika, Pam, Emeka, Lee B., <strong>and</strong> Laura) have been a big help with red tape. I thank<br />

Val de Veyra <strong>and</strong> the other assistants <strong>in</strong> the ECE graduate student office, with whom I<br />

had a good laugh because of a non-unique nickname <strong>in</strong> my email address book.<br />

I very much thank Dr. Harry Dietrich <strong>and</strong> the Office of Naval Research (ONR) who<br />

funded <strong>and</strong> oversaw the Center for Advance Nitride <strong>Electronics</strong> (CANE) program that<br />

I was on.<br />

I would also like to thank Dr. Joy Vann-Hamilton. Joy, who was director of the<br />

M<strong>in</strong>ority Eng<strong>in</strong>eer<strong>in</strong>g Program (MEP) while I was an undergraduate at the University<br />

of Notre Dame, listened to all my unhappy college compla<strong>in</strong>ts, po<strong>in</strong>ted me toward<br />

resources that got me through undergraduate, <strong>and</strong> is solely responsible for my go<strong>in</strong>g<br />

to graduate school. Grad school wasn’t even on my radar <strong>and</strong> I didn’t th<strong>in</strong>k I had the<br />

talent. I can’t thank her enough.<br />

As if I hadn’t thanked enough people <strong>and</strong> entities, I would also like to thank all my<br />

friends who made my time here enjoyable (even if it was jokes at my expense, Perk<br />

<strong>and</strong> Dave). Also, I don’t know what my parents did that I turned out the way I did,<br />

but I want to give them applause for all that they have done. Don’t worry, I won’t do<br />

anyth<strong>in</strong>g to embarrass you like I did at high school graduation.<br />

F<strong>in</strong>ally, no stay <strong>in</strong> Santa Barbara is complete without pick<strong>in</strong>g up a wife. Sarah, who<br />

would have guessed we would meet here <strong>in</strong>stead of at Notre Dame? I thank you for<br />

h<strong>and</strong>l<strong>in</strong>g all my rants with grace, giv<strong>in</strong>g me hope when I had none, distract<strong>in</strong>g me so I<br />

would realize there are a whole lot of <strong>in</strong>terest<strong>in</strong>g th<strong>in</strong>gs to do <strong>in</strong> the world, <strong>and</strong> be<strong>in</strong>g<br />

some one worth liv<strong>in</strong>g for.<br />

“Thanks y’all”...<br />

xviii


For those who never quit.<br />

xix


1.1 Motivation<br />

1<br />

Introduction<br />

GALLIUM nitride (<strong>GaN</strong>) <strong>and</strong> its related compound materials, <strong>in</strong>dium nitride<br />

(InN) <strong>and</strong> alum<strong>in</strong>um nitride (AlN), have already been commercialized for<br />

several applications. This is an impressive feat for an immature technology where each<br />

year research cont<strong>in</strong>ues to br<strong>in</strong>g improvements. The reason for this rush to market is<br />

the needs the nitrides are address<strong>in</strong>g. For optical applications, <strong>GaN</strong>-based solutions<br />

<strong>in</strong>clude LEDs, lasers, <strong>and</strong> detectors <strong>in</strong> the UV <strong>and</strong> blue wavelengths. For electronic<br />

applications, <strong>GaN</strong> high electron mobility transistors (<strong>HEMTs</strong>) provide some of the<br />

highest microwave power performance to be found from solid state devices.<br />

<strong>GaN</strong>-based electronics have two major setbacks. The first is the cost. This re-<br />

sults largely from the expensive substrates needed, with the better substrates currently<br />

cost<strong>in</strong>g nearly an order of magnitude more than lower-performance options. The other<br />

hurdle is growth immaturity, which causes reliability to be mediocre. Because of these<br />

disadvantages, if another material system (Si, GaAs, InP, etc.) can meet the needs of an<br />

1


CHAPTER 1. INTRODUCTION<br />

application, <strong>in</strong>dustry will choose it over <strong>GaN</strong>. This means <strong>GaN</strong> <strong>HEMTs</strong> are be<strong>in</strong>g con-<br />

sidered ma<strong>in</strong>ly for power amplifiers <strong>in</strong> microwave products such as base stations [1,2].<br />

Industry has a great need for these products to be compact. Designers cont<strong>in</strong>uously<br />

move from hybrid (a circuit board with discrete components) to <strong>in</strong>tegrated (a s<strong>in</strong>gle,<br />

small chip) solutions. Instead of a chip with just a power amplifier, it is preferred to<br />

have the amplifier, <strong>and</strong> complete transmit <strong>and</strong> receive paths, <strong>in</strong> a s<strong>in</strong>gle chip called a<br />

front-end module (FEM). Figure 1.1 shows an example of a simple transmitter, <strong>and</strong> the<br />

preferred <strong>in</strong>tegration boundary of power amplifier plus other components. It may be<br />

possible that <strong>GaN</strong> monolithic microwave <strong>in</strong>tegrated circuits (MMICs) provide better<br />

performance <strong>in</strong> terms of power, radiation hardness, <strong>and</strong> operation temperature, lead<strong>in</strong>g<br />

to future products.<br />

From<br />

DSP<br />

Modulator<br />

Oscillator<br />

Mixer Filter Power<br />

Amp.<br />

Ideally implemented as a s<strong>in</strong>gle transmitter MMIC<br />

Figure 1.1: Cartoon of a very simple transmitter.<br />

2<br />

Antenna


CHAPTER 1. INTRODUCTION<br />

Power<br />

Channel<br />

Width<br />

Channel<br />

Space<br />

<strong>Noise</strong><br />

Side<br />

B<strong>and</strong><br />

Frequency<br />

<strong>Noise</strong><br />

Floor<br />

Figure 1.2: Cartoon of a CDMA-like spectrum with four channels.<br />

Basic <strong>GaN</strong> MMIC build<strong>in</strong>g blocks are now appear<strong>in</strong>g <strong>in</strong> the literature [3–7], <strong>and</strong><br />

it is only a matter of time before <strong>GaN</strong> transmit <strong>and</strong> receive MMICs are made. An<br />

important metric of such circuits will be their noise performance. In particular, phase<br />

noise <strong>and</strong> noise figure (NF) are common figures of merit for characteriz<strong>in</strong>g noise. A<br />

ma<strong>in</strong> focus of this work is exam<strong>in</strong><strong>in</strong>g these aspects of <strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> oscillators.<br />

Communication channel spac<strong>in</strong>g is set based on specifications of the maximum<br />

noise a channel will produce out-of-b<strong>and</strong>. This is shown <strong>in</strong> figure 1.2. Sett<strong>in</strong>g the<br />

space between channels as small as possible is extremely important due to regulated<br />

b<strong>and</strong>width restrictions. 1 Typically the component that sets the m<strong>in</strong>imum on this noise<br />

1In fact, whenever the Federal Communications Commission (FCC) auctions b<strong>and</strong>width the sell<strong>in</strong>g<br />

price is <strong>in</strong> the hundreds of millions [8].<br />

3


CHAPTER 1. INTRODUCTION<br />

performance is an oscillator <strong>in</strong> the circuit, as <strong>in</strong> figure 1.1. An oscillator provides<br />

a signal source at a (usually variable) reference frequency. In addition, oscillators<br />

produce a large amount of noise at frequencies close to the reference frequency. This<br />

is called phase noise.<br />

It is understood that various device noise sources contribute to the phase noise.<br />

These <strong>in</strong>clude thermal, shot, <strong>and</strong> low-frequency (also called flicker, or 1/f) sources.<br />

The quantitative analysis of how these sources contribute to the phase noise is difficult<br />

even for the simplest of cases. A qualitative description that embodies many important<br />

po<strong>in</strong>ts was presented by Leeson [9], <strong>and</strong> is described as<br />

⎡<br />

L (∆ω) = 10 log ⎣ 2FkT<br />

Psig<br />

⎧<br />

⎨<br />

⎩ 1+<br />

� ω0<br />

2Q∆ω<br />

� ⎫<br />

2 �<br />

�<br />

⎬ ∆ω1/f 3<br />

1+<br />

⎭ |∆ω|<br />

⎤<br />

⎦ (1.1.1)<br />

where the script L is the phase noise <strong>in</strong> a 1 Hz b<strong>and</strong>width at an offset angular fre-<br />

quency, ∆w, from the carrier angular frequency, ω0, Q is the quality of the resonator,<br />

Psig the signal power, <strong>and</strong> F the effective noise figure. 2 Shot <strong>and</strong> thermal noise sources<br />

contribute to F, a measure of the background noise of the device <strong>and</strong> circuit. The low-<br />

frequency noise (LFN) contributes through ∆ω1/f 3, a corner frequency for the phase<br />

noise that presumably relates to the corner frequency of the LFN. From equation 1.1.1,<br />

we can discern that knowledge of the noise sources <strong>in</strong> the device <strong>and</strong> circuit will help<br />

<strong>in</strong> underst<strong>and</strong><strong>in</strong>g the phase noise, <strong>and</strong> if we can reduce them the phase noise will be<br />

2 The parameter F here is not the device noise figure or factor but an “Effective noise figure” which<br />

some see as just a fitt<strong>in</strong>g parameter. F <strong>in</strong> this chapter should not be confused with noise factor <strong>in</strong> the<br />

other chapters.<br />

4


CHAPTER 1. INTRODUCTION<br />

improved. Leeson’s equation also tells us that if Psig can be <strong>in</strong>creased, we should<br />

see an improvement <strong>in</strong> phase noise. Because <strong>GaN</strong> devices are capable of provid<strong>in</strong>g<br />

an order of magnitude more power than gallium arsenide (GaAs) devices, the next<br />

best commercial material, it is of great <strong>in</strong>terest to know if the phase noise of <strong>GaN</strong><br />

circuits can be better than GaAs circuits. One of the goals of this work is to answer<br />

this question.<br />

This dissertation will look at several aspects of the noise performance of <strong>GaN</strong><br />

<strong>HEMTs</strong>. NF <strong>and</strong> LFN measurements are used to evaluate the noise performance of<br />

the <strong>GaN</strong> <strong>HEMTs</strong>. Comparisons are made to measurements of GaAs <strong>HEMTs</strong>, the most<br />

similar commercial device, which <strong>GaN</strong> is try<strong>in</strong>g to displace. Models for NF <strong>and</strong> LFN<br />

noise, some old <strong>and</strong> some new, are presented. To study the phase noise, differen-<br />

tial oscillators were constructed <strong>and</strong> measured. Their performance is expla<strong>in</strong>ed <strong>in</strong> the<br />

context of LFN.<br />

A quick review of previous noise measurements <strong>and</strong> studies of <strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong><br />

oscillators is now presented, followed by a sketch of this work.<br />

1.2 Literature Review of <strong>Noise</strong> <strong>in</strong> <strong>GaN</strong> <strong>HEMTs</strong><br />

This section gives a short background of the <strong>GaN</strong> HEMT literature for NF, LFN,<br />

<strong>and</strong> phase noise prior to this work (late 2002). Most publications after this po<strong>in</strong>t <strong>in</strong><br />

time are already referenced throughout the work. More background <strong>in</strong>formation on<br />

5


CHAPTER 1. INTRODUCTION<br />

each noise subject can also be found at the beg<strong>in</strong>n<strong>in</strong>g of its respective chapter.<br />

Of the three types of noise measurements performed on <strong>GaN</strong> <strong>HEMTs</strong>, LFN has<br />

the longest history. Reports of measurements first started to appear <strong>in</strong> 1998 [10, 11].<br />

Because LFN measurements can be used as a way of monitor<strong>in</strong>g crystal quality, some<br />

reports came from materials scientists [12]. Physicists have measured <strong>GaN</strong> LFN <strong>and</strong><br />

are us<strong>in</strong>g some of the follow<strong>in</strong>g arguments to expla<strong>in</strong> it: tail states near the b<strong>and</strong> gap<br />

edge [11], mobility fluctuations [13], <strong>and</strong> tunnel<strong>in</strong>g of electrons from the channel to<br />

traps <strong>in</strong> surround<strong>in</strong>g layers [12]. No theory is yet accepted as the best explanation.<br />

Various other LFN papers have appeared [12,14–16]. Most measurements are at very<br />

low bias<strong>in</strong>gs such that the device is <strong>in</strong> the l<strong>in</strong>ear region. The only publication that stud-<br />

ies gate <strong>and</strong> dra<strong>in</strong> LFN for the full bias<strong>in</strong>g range of a <strong>GaN</strong> HEMT is Hsu et al. [17].<br />

This paper will be returned to <strong>in</strong> chapter 4. Due to the use of the Hooge parame-<br />

ter, the devices not be<strong>in</strong>g optimized <strong>HEMTs</strong> (some are doped channel HFETs), <strong>and</strong><br />

measurements only <strong>in</strong> the l<strong>in</strong>ear region, many of the above results are not of use to a<br />

circuit designer or device eng<strong>in</strong>eer who is try<strong>in</strong>g to characterize <strong>and</strong> optimize LFN.<br />

Accurate measurements of LFN can also be difficult. Therefore, this work presents<br />

measurements <strong>and</strong> model<strong>in</strong>g of LFN that will be of use to the eng<strong>in</strong>eer<strong>in</strong>g community.<br />

The first published NF report for a <strong>GaN</strong> HEMT was performed by P<strong>in</strong>g <strong>in</strong> 2000 [18].<br />

For a 0.25 x 100 µm gate device, a m<strong>in</strong>imum noise figure (NFm<strong>in</strong>) of 1.06 dB <strong>and</strong> a<br />

ga<strong>in</strong> of 12 dB at a frequency of 10 GHz was demonstrated. This was followed shortly<br />

6


CHAPTER 1. INTRODUCTION<br />

by Nguyen [19], present<strong>in</strong>g a NFm<strong>in</strong> of 0.6 dB at 10 GHz <strong>and</strong> 13.5 dB ga<strong>in</strong> with a<br />

0.15 x 100 µm gate. Other results [20, 21], with improved performance, followed<br />

the next year. After optimiz<strong>in</strong>g the device geometry, Moon showed <strong>in</strong> 2002 that a<br />

<strong>GaN</strong> HEMT could present similar NF at a similar bias<strong>in</strong>g to a GaAs HEMT [22]. The<br />

next notable result <strong>in</strong> the field was when Lu showed that <strong>GaN</strong> <strong>HEMTs</strong> on a sapphire<br />

substrate can have similar NFm<strong>in</strong> performance to <strong>HEMTs</strong> on a SiC substrate [23]. All<br />

previously published results had been on SiC substrates. This work seeks to extend<br />

on the noise figure literature <strong>and</strong> to confirm or deny some of the previously published<br />

results, <strong>in</strong> addition to add<strong>in</strong>g new studies. Table 3.1 lists many <strong>GaN</strong> (as well as GaAs,<br />

SiGe, <strong>and</strong> InP) HEMT NF results.<br />

Prior to this work, <strong>GaN</strong> oscillator references were scarce. The first <strong>GaN</strong> oscillator<br />

was reported by Shealy <strong>in</strong> 2001 [24]. The phase noise was respectable: -92 dBc at<br />

100 kHz for a 6 GHz carrier frequency. While Shealy’s work was a hybrid design, the<br />

first MMIC oscillator was presented by Kaper <strong>in</strong> 2002 [25]. He estimated (but did not<br />

show measurements) the phase noise to be -87 dBc at 100 kHz. These works left room<br />

for further <strong>in</strong>vestigation. A summary of all published <strong>GaN</strong> oscillators appears <strong>in</strong> table<br />

5.1.<br />

The exist<strong>in</strong>g literature was th<strong>in</strong> at the beg<strong>in</strong>n<strong>in</strong>g of this project. The noise research<br />

of <strong>GaN</strong> has s<strong>in</strong>ce approximately doubled to tripled. Because noise is difficult to mea-<br />

sure, second <strong>and</strong> third op<strong>in</strong>ions are valuable. This work will do this, <strong>in</strong> addition to its<br />

7


CHAPTER 1. INTRODUCTION<br />

own unique studies <strong>and</strong> model<strong>in</strong>g.<br />

1.3 Thesis Outl<strong>in</strong>e<br />

Chapter 2 builds up to a NF model. This model is an extension of previous mod-<br />

els <strong>and</strong> fills the need of underst<strong>and</strong><strong>in</strong>g how various noise sources <strong>and</strong> small-signal<br />

parameters <strong>in</strong>fluence NF <strong>and</strong> other noise parameters without the need for prior NF<br />

measurements. The chapter starts with reviews of noise sources, the equivalent circuit<br />

model, NF <strong>and</strong> parameters, <strong>and</strong> previous models. The proposed model is then derived<br />

<strong>and</strong> discussed <strong>in</strong> detail.<br />

With the model<strong>in</strong>g <strong>in</strong> place, several NF studies are presented <strong>in</strong> chapter 3. Effects<br />

of bias <strong>and</strong> gate leakage are studied. Different epitaxial structures are compared, as<br />

is the addition of a field plate. High-frequency <strong>GaN</strong> HEMT devices were borrowed,<br />

measured, <strong>and</strong> compared to <strong>HEMTs</strong> <strong>in</strong> other technologies (<strong>in</strong> particular, GaAs).<br />

A LFN setup was constructed <strong>and</strong> measurements performed to help underst<strong>and</strong> the<br />

phase noise performance. Chapter 4 covers these results. It first shows through mea-<br />

surements why the Hooge parameter should not be used for compar<strong>in</strong>g devices (only<br />

materials). LFN versus bias shows the need for an improved model for the dra<strong>in</strong> noise,<br />

which was then added. LFN studies are then presented <strong>and</strong> comparisons to measured<br />

GaAs <strong>HEMTs</strong> are made.<br />

Differential oscillators are constructed <strong>in</strong> chapter 5 <strong>and</strong> phase noise is measured<br />

8


CHAPTER 1. INTRODUCTION<br />

<strong>and</strong> evaluated. The first version had excellent l<strong>in</strong>earity but poor phase noise. A second<br />

version had good phase noise. The oscillators are compared to other published <strong>GaN</strong><br />

oscillators <strong>and</strong> to differential oscillators <strong>in</strong> Si <strong>and</strong> GaAs technologies.<br />

References<br />

[1] T. Kikkawa, T. Maniwa, H. Hayashi, M. Kanamura, S. Yokokama, M. Nishi,<br />

N. Adachi, M. Yokoyama, Y. Tateno, <strong>and</strong> K. Josh<strong>in</strong>, “An Over 200-W Output<br />

Power <strong>GaN</strong> HEMT Push-Pull Amplifier with High Reliability,” IEEE <strong>Microwave</strong><br />

Theory <strong>and</strong> Tech. Symp., pp. 1347–1350, 2004.<br />

[2] Y. Okamoto, Y. Ando, T. Nakayama, K. Hataya, H. Miyamoto, T. Inoue, M. Senda,<br />

K. Hirata, M. Kosaki, N. Shibata, <strong>and</strong> M. Kuzuhara, “High-Power Recessed-Gate<br />

<strong>Al<strong>GaN</strong></strong><strong>GaN</strong> HFET With a Field-Modulat<strong>in</strong>g Plate,” IEEE Trans. Electron Devices,<br />

vol. 51, no. 12, pp. 2217–2222, Dec. 2004.<br />

[3] H. Ishida, Y. Hirose, T. Murata, Y. Ikeda, T. Matsuno, K. Inoue, Y. Uemoto,<br />

T. Tanaka, T. Egawa, <strong>and</strong> D. Ueda, “A High-Power RF Switch IC Us<strong>in</strong>g Al-<br />

<strong>GaN</strong>/<strong>GaN</strong> HFETs with S<strong>in</strong>gle-Stage Configuration,” Electron Devices, IEEE<br />

Transactions on, vol. 52, no. 8, pp. 1893–1899, 2005.<br />

[4] H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, S. Keller, U. Mishra, <strong>and</strong> R. A. York, “Low Phase-<br />

<strong>Noise</strong> 5 GHz <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT Oscillator Integrated with BaxSr1−xTiO3 Th<strong>in</strong><br />

Films,” IEEE <strong>Microwave</strong> Theory <strong>and</strong> Tech. Symp., pp. 1509–1512, 2004.<br />

[5] V. Kaper, R. Thompson, T. Prunty, <strong>and</strong> J. Shealy, “Signal Generation, Control,<br />

<strong>and</strong> Frequency Conversion <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT MMICs,” <strong>Microwave</strong> Theory <strong>and</strong><br />

Techniques, IEEE Transactions on, vol. 53, no. 1, pp. 55–65, 2005.<br />

[6] G. Ellis, J.-S. Moon, D. Wong, M. Micovic, A. Kurdoghlian, P. Hashimoto, <strong>and</strong><br />

M. Hu, “Wideb<strong>and</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT MMIC Low <strong>Noise</strong> Amplifier,” <strong>in</strong> <strong>Microwave</strong><br />

Symposium Digest, 2004 IEEE MTT-S International, vol. 1, 2004, pp.<br />

153–156 Vol.1.<br />

[7] H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, S. Keller, U. Mishra, <strong>and</strong> R. York, “A C-B<strong>and</strong> High-<br />

Dynamic Range <strong>GaN</strong> HEMT Low-<strong>Noise</strong> Amplifier,” IEEE <strong>Microwave</strong> Components<br />

Lett., vol. 14, no. 6, pp. 262–264, Jun. 2004.<br />

[8] R. Rast, “The Dawn of Digital TV,” Spectrum, IEEE, vol. 42, no. 10, pp. 26–31,<br />

2005.<br />

[9] D. Leeson, “A Simple Model of Feedback Oscillator <strong>Noise</strong> Spectrum,” Proc. IEEE,<br />

vol. 54, no. 2, pp. 329–330, 1966.<br />

9


CHAPTER 1. INTRODUCTION<br />

[10] D. V. Kuksenkov, H. Temk<strong>in</strong>, R. Gaska, <strong>and</strong> J. W. Yang, “Low-Frequency <strong>Noise</strong><br />

<strong>in</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> Heterostructure Field Effect Transistors,” IEEE Electron Devices<br />

Lett., vol. 19, no. 7, pp. 222–224, July 1998.<br />

[11] M. E. Lev<strong>in</strong>shte<strong>in</strong>, F. Pascal, S. Contreras, W. Knap, S. L. Rumyantsev, R. Gaska,<br />

J. W. Yang, <strong>and</strong> M. S. Shur, “Low-Frequency <strong>Noise</strong> <strong>in</strong> <strong>GaN</strong>/GaAlN Heterojunctions.”<br />

Applied Physics Letters, vol. 72, no. 23, pp. 3053–5, June 1998.<br />

[12] A. Bal<strong>and</strong><strong>in</strong>, Ed., <strong>Noise</strong> <strong>and</strong> Fluctuations Control <strong>in</strong> Electronic Devices. Stevenson<br />

Ranch, CA: American Scientific Publishers, 2002.<br />

[13] J. A. Garrido, B. E. Foutz, J. A. Smart, J. R. Shealy, M. J. Murphy, W. J. Schaff,<br />

L. F. Eastman, <strong>and</strong> E. Munoz, “Low-Frequency <strong>Noise</strong> <strong>and</strong> Mobility Fluctuations<br />

<strong>in</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> Heterostructure Field-Effect Transistors.” Applied Physics Letters,<br />

vol. 76, no. 23, pp. 3442–4, June 2000.<br />

[14] M. Lev<strong>in</strong>shte<strong>in</strong>, S. Rumyantsev, M. Shur, R. Gaska, <strong>and</strong> M. Khan, “Low Frequency<br />

<strong>and</strong> 1/f <strong>Noise</strong> <strong>in</strong> Wide-Gap Semiconductors: Silicon Carbide <strong>and</strong> Gallium Nitride,”<br />

Circuits, Devices <strong>and</strong> Systems, IEE Proceed<strong>in</strong>gs [see also IEE Proceed<strong>in</strong>gs G-<br />

Circuits, Devices <strong>and</strong> Systems], vol. 149, no. 1, pp. 32–39, 2002.<br />

[15] A. Bal<strong>and</strong><strong>in</strong>, S. Morozov, S. Cai, R. Li, K. Wang, G. Wijeratne, <strong>and</strong><br />

C. Viswanathan, “Low Flicker-<strong>Noise</strong> <strong>GaN</strong>/<strong>Al<strong>GaN</strong></strong> Heterostructure Field-Effect<br />

Transistors for <strong>Microwave</strong> Communications,” <strong>Microwave</strong> Theory <strong>and</strong> Techniques,<br />

IEEE Transactions on, vol. 47, no. 8, pp. 1413–1417, 1999.<br />

[16] S. Rumyantsev, N. Pala, M. Shur, M. Lev<strong>in</strong>shte<strong>in</strong>, R. Gaska, X. Hu, J. Yang,<br />

G. Sim<strong>in</strong>, <strong>and</strong> M. Khan, “Low Frequency <strong>Noise</strong> <strong>in</strong> <strong>GaN</strong>-Based Transistors,” <strong>in</strong><br />

High Performance Devices, 2000. Proceed<strong>in</strong>gs. 2000 IEEE/Cornell Conference<br />

on, 2000, pp. 257–264.<br />

[17] S. Hsu, P. Valizadeh, D. Pavlidis, J. Moon, M. Micovic, D. Wong, <strong>and</strong> T. Hussa<strong>in</strong>,<br />

“Characterization <strong>and</strong> Analysis of Gate <strong>and</strong> Dra<strong>in</strong> Low-Frequency <strong>Noise</strong> <strong>in</strong><br />

<strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” <strong>in</strong> High Performance Devices, 2002. Proceed<strong>in</strong>gs. IEEE<br />

Lester Eastman Conference on, 2002, pp. 453–460.<br />

[18] A. P<strong>in</strong>g, E. P<strong>in</strong>er, J. Redw<strong>in</strong>g, M. Khan, <strong>and</strong> I. Adesida, “<strong>Microwave</strong> <strong>Noise</strong> Performance<br />

of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” Electron. Lett., vol. 36, no. 2, pp. 175–176, Jan.<br />

2000.<br />

[19] N. Nguyen, M. Micovic, W.-S. Wong, P. Hashimoto, P. Janke, D. Harvey, <strong>and</strong><br />

C. Nguyen, “Robust Low <strong>Microwave</strong> <strong>Noise</strong> <strong>GaN</strong> MODFETs with 0.6 dB <strong>Noise</strong><br />

Figure at 10 GHz,” IEEE Electron. Lett., vol. 36, pp. 469–471, March 2000.<br />

[20] S. Hsu <strong>and</strong> D. Pavlidis, “Low <strong>Noise</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> MODFETs with High Breakdown<br />

<strong>and</strong> Power Characteristics,” Gallium Arsenide Integrated Circuit (GaAs IC)<br />

Symposium, 2001. 23rd Annual Technical Digest, pp. 229–232, Oct. 2001.<br />

10


CHAPTER 1. INTRODUCTION<br />

[21] W. Lu, J. Yang, M. Khan, <strong>and</strong> I. Adesida, “<strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> on SiC with over<br />

100 GHz fT <strong>and</strong> Low <strong>Microwave</strong> <strong>Noise</strong>,” IEEE Trans. Electron Devices, vol. 48,<br />

no. 3, pp. 581–585, Mar. 2001.<br />

[22] J. Moon, M. Micovic, A. Kurdoghlian, P. Janke, P. Hashimoto, W.-S. Wong, L. Mc-<br />

Cray, <strong>and</strong> C. Nguyen, “<strong>Microwave</strong> <strong>Noise</strong> Performance of <strong>Al<strong>GaN</strong></strong><strong>GaN</strong> <strong>HEMTs</strong><br />

With Small DC Power Dissipation,” IEEE Electron Devices Lett., vol. 23, no. 11,<br />

pp. 637–639, Nov. 2002.<br />

[23] W. Lu, V. Kumar, R. Schw<strong>in</strong>dt, E. P<strong>in</strong>er, <strong>and</strong> I. Adesida, “DC, RF, <strong>and</strong> <strong>Microwave</strong><br />

<strong>Noise</strong> Performance of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> on Sapphire Substrates,” IEEE Trans.<br />

<strong>Microwave</strong> Theory Tech., vol. 50, pp. 2499–2503, Nov. 2002.<br />

[24] J. B. Shealy, J. A. Smart, <strong>and</strong> J. R. Shealy, “Low-Phase <strong>Noise</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> FET-<br />

Based Voltage Controlled <strong>Oscillators</strong> (VCOs),” IEEE <strong>Microwave</strong> Components<br />

Lett., vol. 11, no. 6, pp. 244–245, Jun. 2001.<br />

[25] V. Kaper, V. Tilak, H. Kim, R. Thompson, T. Prunty, J. Smart, L. F. Eastman, <strong>and</strong><br />

J. Shealy, “High Power Monolithic <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT Oscillator,” IEEE GaAs<br />

Digest, pp. 251–254, 2002.<br />

11


2<br />

<strong>Noise</strong> Figure Model<strong>in</strong>g of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong><br />

<strong>HEMTs</strong><br />

2.1 Introduction<br />

TO characterize, compare, <strong>and</strong> improve the noise performance of devices, a<br />

theoretical framework is needed that identifies the noise sources, how these<br />

sources contribute to the overall noise, <strong>and</strong> how the noise changes with other param-<br />

eters, such as bias <strong>and</strong> match<strong>in</strong>g conditions. A common approach is to add discrete<br />

noise sources to a small-signal model [1–3]. Depend<strong>in</strong>g on the model, the sources may<br />

be correlated, add<strong>in</strong>g complexity to the derivation <strong>and</strong> <strong>in</strong>terpretation of the particular<br />

model.<br />

This chapter first fills <strong>in</strong> background for the model. <strong>Noise</strong> sources of <strong>in</strong>terest to<br />

noise figure (NF) are reviewed, as is a full small-signal model. NF is then def<strong>in</strong>ed<br />

<strong>in</strong> § 2.4, along with the noise parameters NFm<strong>in</strong> , Γopt <strong>and</strong> rn. Those familiar with<br />

noise sources, NF, <strong>and</strong> small-signal model<strong>in</strong>g could skip these sections <strong>and</strong> cont<strong>in</strong>ue<br />

on to § 2.5. The more widely-used models are presented along with their strengths<br />

12


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

<strong>and</strong> weaknesses <strong>in</strong> § 2.5. <strong>Noise</strong> model<strong>in</strong>g derived from these previous models is then<br />

<strong>in</strong>troduced. Its strengths <strong>and</strong> weaknesses are also discussed. F<strong>in</strong>ally, the NF model<br />

is used for simulation of device noise figure performance of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,<br />

compared to other models, <strong>and</strong> discussed <strong>in</strong> depth.<br />

2.2 <strong>Noise</strong> Sources<br />

The two most common types of noise are shot <strong>and</strong> thermal noise. Both are referred<br />

to as “flat” or “white,” mean<strong>in</strong>g the noise power versus frequency is constant. Each<br />

will be reviewed <strong>in</strong> this section. The relevance of other noise sources to NF is also<br />

discussed.<br />

2.2.1 Thermal <strong>Noise</strong><br />

Thermal noise was first studied <strong>in</strong> detail by Johnson <strong>in</strong> 1927 [4], <strong>and</strong> expla<strong>in</strong>ed<br />

mathematically by Nyquist [5]. Its physical orig<strong>in</strong> is the agitation of electrons <strong>in</strong> a<br />

conductor. The r<strong>and</strong>om scatter<strong>in</strong>g of electrons by atoms followed by their relaxation<br />

back to a ground state leads to fluctuations that can be measured as a voltage or current.<br />

The use of statistical analysis <strong>and</strong> thermal physics [6] leads to the follow<strong>in</strong>g expression<br />

that represents the available noise power from a resistance <strong>in</strong>to a matched load<br />

PT hermal =<br />

⎡<br />

⎣ hf<br />

2 +<br />

13<br />

exp � hf<br />

kT<br />

⎤<br />

hf<br />

� ⎦ ∆f (2.2.1)<br />

− 1


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

where h is Planck’s constant, k is Boltzmann’s constant, T is temperature, <strong>and</strong> f is the<br />

frequency.<br />

The term ∆f is the b<strong>and</strong>width <strong>in</strong> which the noise is be<strong>in</strong>g measured. To determ<strong>in</strong>e<br />

the total noise com<strong>in</strong>g from, for example, an amplifier, this expression would be <strong>in</strong>te-<br />

grated over the b<strong>and</strong>width of the amplifier. A noise measurement might be performed<br />

over a small b<strong>and</strong>width, less than one Hertz for some low-frequency noise measure-<br />

ments, show<strong>in</strong>g the need to keep this fact <strong>in</strong> m<strong>in</strong>d. When values are quoted <strong>and</strong> a<br />

b<strong>and</strong>width is not specified, it is assumed to be a 1 Hz b<strong>and</strong>width. This convention is<br />

followed <strong>in</strong> this work.<br />

Equation 2.2.1 is a general expression, <strong>and</strong> is needed if operat<strong>in</strong>g at cryogenic tem-<br />

peratures or extreme frequencies (such as the THz range). For most eng<strong>in</strong>eer<strong>in</strong>g pur-<br />

poses, the simpler expression<br />

PT hermal ≈ kT∆f (2.2.2)<br />

can be used. A h<strong>and</strong>y value from this expression is that the available noise at room<br />

temperature <strong>in</strong> a 1 Hz b<strong>and</strong>width is -174 dBm (4 × 10 −21 W).<br />

When the load is not matched, or the noise needs to be expressed as a voltage or<br />

current (described by a variance), the follow<strong>in</strong>g forms are useful:<br />

�<br />

v 2 �<br />

n =4kTR∆f (2.2.3)<br />

�<br />

i 2 �<br />

n = 4kT<br />

∆f (2.2.4)<br />

R<br />

14


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

R is the resistance of the noise generat<strong>in</strong>g material or medium. A 1kΩ resistor gener-<br />

ates 4 nV/ √ Hz of noise. Not every resistor represented <strong>in</strong> a circuit schematic generates<br />

noise. For example, rπ <strong>in</strong> a bipolar transistor is a manifestation of the transistor’s I-V<br />

characteristics, <strong>and</strong> is not a real resistance.<br />

Ideal reactive components do not generate noise. However, because real <strong>in</strong>ductors<br />

<strong>and</strong> capacitors are lossy, the circuit-modeled parasitic resistances of these components<br />

will generate thermal noise. Also, these components shape the b<strong>and</strong>width of the noise<br />

<strong>in</strong> ways similar to shap<strong>in</strong>g a signal.<br />

F<strong>in</strong>ally, the above equations only apply at thermal equilibrium. A biased transistor<br />

is not at thermal equilibrium. However, a small-signal model of a HEMT, which does<br />

not <strong>in</strong>clude bias, can have its parasitic resistances considered at thermal equilibrium.<br />

Later, <strong>in</strong> § 2.6.3, the channel noise, which is a form of non-equilibrium thermal noise,<br />

are derived. Further <strong>in</strong>formation on thermal noise can be found <strong>in</strong> [6–9].<br />

2.2.2 Shot <strong>Noise</strong><br />

Shot noise was reported <strong>and</strong> expla<strong>in</strong>ed by W. Schottky <strong>in</strong> 1918. The analogy of this<br />

noise to buck shot be<strong>in</strong>g dropped <strong>in</strong>to a bucket is the basis for its name [8]. Shot noise<br />

exists when two conditions are met: (1) a DC current is flow<strong>in</strong>g <strong>and</strong> (2) the charge<br />

carriers compos<strong>in</strong>g the DC current cross a potential barrier. This second condition<br />

is why resistors <strong>and</strong> the channel of a HEMT do not generate shot noise. The noise<br />

15


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

exists because charge is discrete <strong>and</strong> will cross the potential barrier at r<strong>and</strong>om times.<br />

If the charge crossed at the same time or equal time <strong>in</strong>tervals, the spectrum of the shot<br />

noise would not be white <strong>and</strong> could be better dealt with <strong>in</strong> circuit design (possibly by<br />

filter<strong>in</strong>g).<br />

As with thermal noise, shot noise is not constant at all frequencies. The noise de-<br />

creases at frequencies above which the transit time across the barrier is short compared<br />

to the <strong>in</strong>verse of this frequency [7]. Because devices are operated at frequencies well<br />

below this, the noise can be considered flat [7].<br />

Us<strong>in</strong>g math from r<strong>and</strong>om processes [9], shot noise is describe by the follow<strong>in</strong>g<br />

equation for current fluctuations:<br />

�<br />

i 2 �<br />

n =2qIDC∆f (2.2.5)<br />

with q be<strong>in</strong>g electronic charge (1.6 × 10 −19 C), IDC the DC current (Amps), <strong>and</strong> ∆f<br />

aga<strong>in</strong> the b<strong>and</strong>width of <strong>in</strong>terest.<br />

2.2.3 Other Sources of <strong>Noise</strong><br />

Shot <strong>and</strong> thermal are the important noise sources for NF model<strong>in</strong>g. Others exist, <strong>and</strong><br />

at frequencies lower than RF should be considered. They will be briefly mentioned<br />

here. Flicker <strong>and</strong> generation-recomb<strong>in</strong>ation noise (covered <strong>in</strong> detail <strong>in</strong> chapter 4) are<br />

important for oscillator phase noise, but are not of concern for NF measurements <strong>in</strong><br />

the RF frequency range. Burst noise is <strong>in</strong>terpreted as a type of low-frequency noise<br />

16


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

with a 1/f 2 spectrum. It also would not affect NF measurements <strong>in</strong> the GHz range, nor<br />

has it been reported <strong>in</strong> <strong>GaN</strong>-based <strong>HEMTs</strong>. Avalanche noise is a form of shot noise,<br />

<strong>and</strong> usually applies to a semiconductor junction. With a high enough field across a<br />

junction, avalanche multiplication can occur [10]. This <strong>in</strong>crease <strong>in</strong> carriers leads to<br />

an <strong>in</strong>crease <strong>in</strong> the shot noise proportional to the cube of the multiplication factor [9].<br />

While the fields can be very high, no evidence has been shown that this type of noise<br />

appears <strong>in</strong> <strong>GaN</strong>-based <strong>HEMTs</strong>. As will be seen <strong>in</strong> model<strong>in</strong>g later, shot <strong>and</strong> thermal<br />

types of noise are sufficient to predict the noise figure of <strong>GaN</strong> <strong>HEMTs</strong>.<br />

2.3 Equivalent Circuit Model<br />

A small-signal model will be used <strong>in</strong> determ<strong>in</strong><strong>in</strong>g the device noise figure. Device<br />

small-signal model<strong>in</strong>g has been covered extensively <strong>in</strong> the literature [11–17]. Here<br />

a short summary will be given. The model used <strong>in</strong> this work is superimposed on<br />

the cross section of a HEMT <strong>in</strong> figure 2.1. The parameters are bias-dependent but<br />

frequency <strong>in</strong>dependent. In this chapter <strong>and</strong> chapter 3, the parameters are extracted at<br />

the bias of best device m<strong>in</strong>imum noise figure performance.<br />

At the heart of the model is the gate-source capacitance, Cgs, <strong>and</strong> the transconduc-<br />

tance, gm. The phase associated with gm, ωτ, is a necessary delay that accounts for<br />

channel charge to redistribute after a change <strong>in</strong> gate voltage. The dra<strong>in</strong>-source resis-<br />

tance, Rds, is a measure of how effectively a signal can be extracted from the device.<br />

17


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Ls<br />

Rs<br />

Vc +<br />

-<br />

Source Gate<br />

Cpgs<br />

Cgs<br />

Ri<br />

g me jwτ v c<br />

R ds<br />

Cds<br />

Rgd<br />

C gd<br />

Cpds<br />

Lg<br />

R g<br />

Rd<br />

Cpgd<br />

L d<br />

Dra<strong>in</strong><br />

Figure 2.1: Cartoon show<strong>in</strong>g the device small-signal model on a cross section of a<br />

HEMT.<br />

This is because it will reduce the effective load of the device. There are two primary<br />

causes for its degradation: conduction <strong>in</strong> the buffer due to poor growth (traps) <strong>and</strong><br />

spread of electrons from the 2DEG when the device is under extreme bias<strong>in</strong>g (very<br />

high electric fields).<br />

The channel <strong>and</strong> gate-dra<strong>in</strong> resistances, Ri <strong>and</strong> Rgd, have vague physical <strong>in</strong>terpre-<br />

tations. Ri is regarded as either a physical channel resistance or charg<strong>in</strong>g resistance<br />

for Cgs. Because it is hard to extract separately from the gate resistance, Rg, the two<br />

are sometimes lumped together. The gate-dra<strong>in</strong> capacitance, Cgd, is setup by the space<br />

charge region between the gate <strong>and</strong> dra<strong>in</strong>, similar to Cgs, but typically an order of mag-<br />

nitude smaller. It <strong>in</strong>troduces a bothersome feedback which reduces the high-frequency<br />

18


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

performance of the device. The gate-dra<strong>in</strong> resistance, Rgd, is also argued as a charg<strong>in</strong>g<br />

resistance for Cgd. Any capacitance between the dra<strong>in</strong> <strong>and</strong> source, which will be very<br />

small, is accounted for with Cds.<br />

The parameters Cgs, gm, τ, Ri, Rds, Rgd, Cgd, <strong>and</strong> Cds taken together are referred to<br />

as the <strong>in</strong>tr<strong>in</strong>sic device. The other elements of the model are unfortunate side-effects of<br />

hav<strong>in</strong>g to provide physical connections to the device <strong>and</strong> the parasitic <strong>and</strong> distributed<br />

effects that occur when operat<strong>in</strong>g at frequencies <strong>in</strong> the GHz range. These parasitic<br />

elements are called the extr<strong>in</strong>sic elements of the device.<br />

The dra<strong>in</strong> <strong>and</strong> source resistances, Rd <strong>and</strong> Rs, come from two physical processes.<br />

The first is the f<strong>in</strong>ite resistance of the 2DEG, <strong>and</strong> so the access regions between gate<br />

source <strong>and</strong> gate dra<strong>in</strong> contribute to these parasitics. The second source is the non-<br />

ideal ohmic contact behavior between the metal pads <strong>and</strong> the semiconductor. These<br />

resistances scale directly with the device width as used <strong>in</strong> § 2.6.5. Because the gate<br />

length is short (0.7 µm here), the resistances of the gate metals contribute to Rg.<br />

From an AC viewpo<strong>in</strong>t, the signal decays as it propagates along the length of the gate<br />

resistance. A distributed model argument shows this resistance to be approximately<br />

1/3 of its DC value [14]. All of these resistances generate thermal noise. Rs <strong>and</strong><br />

Rg will be important for model<strong>in</strong>g of noise figure as seen later. Because Rd is at<br />

the output, <strong>and</strong> its magnitude of noise is far smaller than the channel noise, it can be<br />

ignored.<br />

19


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

The dra<strong>in</strong>, gate, <strong>and</strong> source <strong>in</strong>ductances, Ld Lg, <strong>and</strong> Ls, respectively, account for<br />

signal delay along the various contact pads. Lg is usually largest, <strong>and</strong> will affect<br />

any <strong>in</strong>put match (noise or power) to the device. F<strong>in</strong>ally there are the various pad<br />

capacitances between their respective device term<strong>in</strong>als: Cpgs, Cpgd, <strong>and</strong> Cpds. Cpgs<br />

<strong>and</strong> Cpgd are small, while Cpds for devices used <strong>in</strong> this work is of the same order as<br />

Cgd.<br />

S-parameter measurements of devices along with open <strong>and</strong> shorted test structures<br />

were used to f<strong>in</strong>d the small-signal parameters. In addition, DC measurements of TLM<br />

structures <strong>and</strong> some basic h<strong>and</strong> calculations [14] were used to determ<strong>in</strong>e Rg, Rs, <strong>and</strong><br />

Rd. For a better fit, optimization was performed us<strong>in</strong>g Advanced Design System<br />

(ADS). It is important to have accurate small-signal parameters; any discrepancies<br />

directly translate <strong>in</strong>to <strong>in</strong>correct noise figure parameter prediction. Some ADS files<br />

useful for extract<strong>in</strong>g <strong>and</strong> optimiz<strong>in</strong>g the small-signal parameters can be found <strong>in</strong> ap-<br />

pendix A.<br />

Extracted small-signal parameters of devices from various samples are displayed <strong>in</strong><br />

table 2.1. The top half of the table is the <strong>in</strong>tr<strong>in</strong>sic parameters, while the bottom half is<br />

the extr<strong>in</strong>sic parameters. Many of the extr<strong>in</strong>sic parasitics come from the geometry of<br />

the device or the pad parasitics <strong>and</strong> will be the same for the different samples. These<br />

parameters will be used <strong>in</strong> this chapter <strong>and</strong> <strong>in</strong> chapter 3.<br />

20


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Sample: 15% 25% 25% 27% 35% 35%<br />

on SiC w/AlN<br />

Intr<strong>in</strong>sic Ri (Ω) 6.2 8.24 8.86 9.23 8.23 4.07<br />

Parameters Rds (Ω) 577 588 622 562 833 627<br />

Rgd (Ω) 75.7 287 238 96 96.2 33.4<br />

Cgd (fF) 19 19.1 13.5 20.6 18.7 29.9<br />

Cds (fF) 2.97 2.33 1.34 2.34 1.66 1.13<br />

Cgs (pF) 0.23 0.265 0.22 0.191 0.191 0.215<br />

gm (mS) 34.1 37 33.2 33 30.5 34.9<br />

τ (ps) 1.2 2.25 2.6 2 2.25 2.03<br />

Extr<strong>in</strong>sic Rs (Ω) 10.2 6.18 5.19 4.04 4.31 5.30<br />

Parameters Rd (Ω) 17.9 10.24 8.99 7.41 7.20 7.54<br />

Rg (Ω) ←− 3.03 −→<br />

Ls (pH) ←− 12 −→<br />

Ld (pH) ←− 22.3 −→<br />

Lg (pH) ←− 45.6 −→<br />

Cpgs (fF) ←− 1.38 −→<br />

Cpds (fF) ←− 29.6 −→<br />

Cpgd (fF) ←− 5.4 −→<br />

Table 2.1: Extracted small-signal parameters for various samples. Devices have a gate<br />

geometry of 0.7 × 150 µm.<br />

2.4 <strong>Noise</strong> Figure <strong>and</strong> <strong>Noise</strong> Parameters<br />

An amplifier provides ga<strong>in</strong> to both the <strong>in</strong>put signal <strong>and</strong> the noise. The amplifier will<br />

also add noise from its <strong>in</strong>tr<strong>in</strong>sic noise sources. This makes the signal to noise ratio at<br />

the output worse than at the <strong>in</strong>put. Express<strong>in</strong>g a ratio of these two ratios at a reference<br />

temperature leads to the def<strong>in</strong>ition of noise figure.<br />

F ≡ (S/N)<strong>in</strong><br />

�<br />

�<br />

�<br />

�<br />

(S/N)out<br />

� T =Treference=290K<br />

(2.4.1)<br />

It can be applied to any two port [3] (by extension even to mixers [7]). Here, it will<br />

21


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

be applied to a s<strong>in</strong>gle HEMT. Sometimes NF is called noise factor, F. Traditionally,<br />

the names were <strong>in</strong>terchangeable, but now it is common that noise figure is the noise<br />

factor expressed <strong>in</strong> decibels (NF = 10 log 10(F)). This is the convention followed <strong>in</strong><br />

this work. Sometimes it is expressed as a temperature (Tnoise =(F − 1)Treference).<br />

<strong>Noise</strong> figure by itself gives no details of the noise sources of the device or amplifier.<br />

However, noise sources can be added as <strong>in</strong> Figure 2.2. Inoise is the equivalent short-<br />

circuited noise current source <strong>and</strong> Enoise the equivalent open-circuited noise voltage<br />

source. Together, they account for all noise sources of the device which can now be<br />

modeled as a noiseless two port. These two sources will likely be correlated because<br />

of the various physical noise sources of the device that contribute to them [8]. Models<br />

that describe these sources of noise are discussed <strong>in</strong> the next section. An <strong>in</strong>put is<br />

connected to the device, represented as a source impedance <strong>in</strong> figure 2.2. This <strong>in</strong>put<br />

will generate its own noise, represented as Esource. This could be thermal noise from<br />

E source<br />

+ -<br />

Z source<br />

E noise<br />

+ -<br />

I noise<br />

<strong>Noise</strong>less<br />

2-port<br />

network<br />

Figure 2.2: Equivalent model of a transistor driven by a noisy source of impedance<br />

Zsource.<br />

22


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

a passive network <strong>and</strong>/or active device noise. From this, F (<strong>and</strong> hence, NF) can be<br />

written as [8]<br />

F =1+<br />

�<br />

|Enoise + InoiseZsource| 2�<br />

〈Esource〉 2<br />

(2.4.2)<br />

As can be seen, the value of the source impedance actually affects the noise figure.<br />

Once Enoise <strong>and</strong> Inoise have been determ<strong>in</strong>ed through whichever model is applied, the<br />

noise figure can be predicted for chang<strong>in</strong>g source impedance. F can also be expressed<br />

as [18,19]:<br />

F = Fm<strong>in</strong> + 4Rn<br />

Zo<br />

|Γsource − Γopt| 2<br />

(1 −|Γsource| 2 ) |1+Γopt| 2<br />

(2.4.3)<br />

Γsource is the reflection coefficient of the source impedance. Equation 2.4.3 conta<strong>in</strong>s<br />

four parameters that taken together are called the noise parameters. They are:<br />

Fm<strong>in</strong> The best achievable noise figure. It occurs only when the source impedance is<br />

set to Zopt (conversely Γopt).<br />

|Γopt| The magnitude of the source reflection coefficient that provides the m<strong>in</strong>imum<br />

noise figure, Fm<strong>in</strong>.<br />

� Γopt The angle of the source reflection coefficient that provides Fm<strong>in</strong>.<br />

Rn An effective “slope.” The larger its value, the quicker the noise figure <strong>in</strong>creases<br />

as Γsource is changed from its optimum value. 1<br />

These parameters are usually measured as described <strong>in</strong> § 3.3. As can be reasoned<br />

from equation 2.4.3, there are circles of constant noise on an impedance or Smith<br />

1Rn has units of ohms. It can also be normalized to the reference impedance, Zo. Then the variable<br />

is usually written as rn <strong>in</strong>stead.<br />

23


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Gmax<br />

Gmax-1dB<br />

Gmax-2dB<br />

NFm<strong>in</strong><br />

NFm<strong>in</strong>+1dB<br />

NFm<strong>in</strong>+2dB<br />

NFm<strong>in</strong>+3dB<br />

Figure 2.3: Smith Chart show<strong>in</strong>g m<strong>in</strong>imum noise figure <strong>and</strong> circles of constant noise<br />

figure (solid circles) along with maximum ga<strong>in</strong> <strong>and</strong> circles of constant ga<strong>in</strong> (dashed<br />

circles).<br />

Chart plane. This is shown <strong>in</strong> Figure 2.3. It should not be surpris<strong>in</strong>g to f<strong>in</strong>d the small-<br />

signal ga<strong>in</strong> circles <strong>and</strong> noise circles do not overlap because Γopt is an <strong>in</strong>tentional ga<strong>in</strong><br />

mismatch to m<strong>in</strong>imize noise [20].<br />

That the noise performance changes with the source match is extremely important<br />

for LNA design. In the next section, noise models for <strong>HEMTs</strong> that are more <strong>in</strong> depth<br />

than <strong>in</strong> figure 2.2 will be reviewed. Once these noise sources are determ<strong>in</strong>ed, they can<br />

be used elsewhere, such as assist<strong>in</strong>g <strong>in</strong> oscillator phase noise prediction.<br />

24


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

2.5 HEMT <strong>Noise</strong> Figure Models<br />

The major noise figure models used for <strong>HEMTs</strong> are now presented. The work by<br />

van der Ziel, Pucel, Fukui, <strong>and</strong> Pospieszalski is the basis for most other <strong>in</strong>vestigations<br />

<strong>and</strong> model<strong>in</strong>g of noise figure. The key ideas, equations, strengths <strong>and</strong> weaknesses of<br />

each will be reviewed. Shortcom<strong>in</strong>gs found <strong>in</strong> each will show the need for further<br />

work <strong>and</strong> provide motivation for the next section.<br />

2.5.1 van der Ziel <strong>and</strong> Pucel Models<br />

The theoretical work for noise sources <strong>in</strong> FETs at microwave frequencies was <strong>in</strong>-<br />

troduced by van der Ziel <strong>in</strong> the early 1960s [3, 21, 22]. He derived noise sources for<br />

the channel <strong>and</strong> “<strong>in</strong>duced gate noise.” This gate noise is expla<strong>in</strong>ed as fluctuat<strong>in</strong>g noise<br />

<strong>in</strong> the channel capacitively coupl<strong>in</strong>g to the gate through Cgs <strong>and</strong> Cgd, caus<strong>in</strong>g an ef-<br />

fective, <strong>and</strong> correlated, noise source at the gate. The gate noise, channel (dra<strong>in</strong>) noise,<br />

correlation (C), <strong>and</strong> cross-term 〈igi∗ d 〉 can be written as [3]<br />

�<br />

i 2 �<br />

g<br />

=4kTaδ ω2C 2 gs<br />

5gd0<br />

∆f (2.5.1)<br />

�<br />

i 2 �<br />

d =4kTaΓgd0∆f (2.5.2)<br />

C = 〈igi∗ d 〉<br />

�� �<br />

i 2 g<br />

〈i 2 d 〉<br />

(2.5.3)<br />

〈igi ∗ d〉 = 2<br />

3 jωCgskTa∆f (2.5.4)<br />

25


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

ω is the angular frequency, Ta is the ambient temperature, gd0 is the dra<strong>in</strong>-source con-<br />

ductance when Vds is zero, δ is a parameter van der Ziel gives as 4/3, <strong>and</strong> Γ will be<br />

derived <strong>in</strong> § 2.6.2 (for now, it can be considered a constant of 2/3). Van der Ziel orig-<br />

<strong>in</strong>ally found a correlation of 0.395j for JFETs [3]. For aggressively-scaled <strong>HEMTs</strong>,<br />

the correlation is experimentally found to be ∼0.7j [23,24]. Van der Ziel does give ex-<br />

pressions for noise figure, but they are often <strong>in</strong> terms of more complicated parameters,<br />

under specific conditions, or for different devices, mak<strong>in</strong>g them of limited use [3].<br />

While the gate <strong>and</strong> dra<strong>in</strong> noise expressions above allow for noise prediction, for accu-<br />

rate results the correlation must be measured. This is because small changes <strong>in</strong> cor-<br />

relation can have a large effect <strong>in</strong> predicted noise parameters. Also, the noise sources<br />

are bias-<strong>and</strong> frequency-dependent.<br />

In 1975, Pucel <strong>and</strong> his co-workers took what was learned from van der Ziel <strong>and</strong><br />

extended it <strong>in</strong> a very detailed publication [2]. Their work was explicitly for a GaAs<br />

MESFET (<strong>and</strong> thus more applicable to a HEMT), whereas van der Ziel’s work was<br />

ma<strong>in</strong>ly for JFETs. The Pucel model<strong>in</strong>g takes <strong>in</strong>to account velocity saturation <strong>and</strong> how<br />

the noise changes with bias based on changes <strong>in</strong> the small-signal parameters <strong>and</strong> their<br />

noise variables P, R, <strong>and</strong> C. The model also uses a gate <strong>and</strong> dra<strong>in</strong> noise source that<br />

are correlated, shown <strong>in</strong> a small-signal model <strong>in</strong> figure 2.4. In this figure, the gate<br />

<strong>and</strong> dra<strong>in</strong> noise sources account for all noise generated by the <strong>in</strong>tr<strong>in</strong>sic device (<strong>in</strong>side<br />

the dashed “<strong>Noise</strong>less” box). The parasitic resistances, Rg, Rs, <strong>and</strong> Rd, still generate<br />

26


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Gate<br />

Lg<br />

Rg<br />

Ri gme Rds Cgs<br />

jwτ Ig v Cds<br />

+<br />

c<br />

Id Vc Cpgs<br />

-<br />

Rgd Cgd<br />

Cpgd<br />

Rs<br />

Ls<br />

Source<br />

<strong>Noise</strong>less<br />

Figure 2.4: Pucel noise model <strong>in</strong> a small-signal circuit.<br />

Rd<br />

Ld<br />

Cpds<br />

thermal noise. The gate <strong>and</strong> dra<strong>in</strong> noise sources are related to the noise variables by<br />

P =<br />

R =<br />

〈i 2 d 〉<br />

4kTa |Y21| 2 ∆f =<br />

|Y21| 2<br />

4kTa |Y11| 2 ∆f<br />

〈i 2 d 〉<br />

4kTagm∆f<br />

�<br />

i 2 �<br />

g =<br />

gm<br />

4kTaω2C 2 �<br />

i<br />

gs∆f<br />

2 �<br />

g<br />

Dra<strong>in</strong><br />

(2.5.5)<br />

(2.5.6)<br />

<strong>and</strong> the same correlation as <strong>in</strong> equation 2.5.3. Although Pucel determ<strong>in</strong>ed very de-<br />

tailed expressions for these noise variables, for accurate results the model<strong>in</strong>g had to<br />

be fitted to data. Today, if this model is used, the parameters are determ<strong>in</strong>ed empiri-<br />

cally [7], fitt<strong>in</strong>g to noise figure measurements. An excellent paper us<strong>in</strong>g this model on<br />

<strong>GaN</strong> <strong>HEMTs</strong> that shows directly how to extract these parameters from measurements<br />

is found <strong>in</strong> [25]. This method also treats both parts of the correlation, magnitude <strong>and</strong><br />

phase, as variables to be determ<strong>in</strong>ed <strong>in</strong>stead of restrict<strong>in</strong>g the phase to 90 ◦ (strictly<br />

imag<strong>in</strong>ary). This formulation is used <strong>in</strong> this work. Information about an ADS project<br />

27


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

that helps <strong>in</strong> determ<strong>in</strong><strong>in</strong>g these parameters, conta<strong>in</strong>s a small-signal circuit with this<br />

type of noise model<strong>in</strong>g, <strong>and</strong> calculates noise figure can be found <strong>in</strong> appendix A. Be-<br />

cause it uses the same correlation as van der Ziel’s model<strong>in</strong>g, the Pucel model suffers<br />

the same drawbacks. As mentioned, the parameters usually must be determ<strong>in</strong>ed from<br />

measurements, limit<strong>in</strong>g its predictive power. Because the noise sources are similar for<br />

both models, they will be referred to as correlated noise (CN) models <strong>in</strong> this work.<br />

2.5.2 Fukui Model<br />

Fukui garnered much attention <strong>in</strong> the late 1970s with the <strong>in</strong>troduction of his empir-<br />

ical model [26]. Although it <strong>in</strong>volved empirical parameters, it was far simpler than<br />

the previously reported noise models, was expressed directly <strong>in</strong> terms of the noise<br />

parameters, <strong>and</strong> made clear how key small-signal parameters contributed to the noise<br />

performance. The model is also convenient because the noise at different frequencies<br />

<strong>and</strong> device scal<strong>in</strong>g can be easily determ<strong>in</strong>ed. His expressions for the noise parameters<br />

are:<br />

�<br />

Rg + Rs<br />

Fm<strong>in</strong> = 1+k1fCgs<br />

gm<br />

� �<br />

1<br />

Ropt = k3 + Rg + Rs<br />

4gm<br />

Xopt = k4<br />

fCgs<br />

Rn = k2<br />

g2 m<br />

28<br />

(2.5.7)<br />

(2.5.8)<br />

(2.5.9)<br />

(2.5.10)


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

The variables k1−4, are the fitt<strong>in</strong>g parameters, <strong>and</strong> will change with the device tech-<br />

nology <strong>and</strong> bias. While this model can be useful for h<strong>and</strong> calculations, the Pucel <strong>and</strong><br />

Pospieszalski models are more complete <strong>and</strong> directly usable <strong>in</strong> a simulator such as<br />

ADS.<br />

2.5.3 Pospieszalski Model<br />

In the late 1980’s, Pospieszalski <strong>in</strong>troduced a new noise figure model that took a<br />

different approach than the previous methods by remov<strong>in</strong>g correlation between the<br />

noise sources [1]. There are only two noise sources for the entire transistor: thermal<br />

noise of Ri <strong>and</strong> Rds. Instead of these resistors be<strong>in</strong>g at ambient temperature, Ta, they<br />

are at higher effective temperatures Tg <strong>and</strong> Td, shown <strong>in</strong> figure 2.5. Tg is usually<br />

(but not always) close to room temperature, while Td can be several thous<strong>and</strong> degrees<br />

Kelv<strong>in</strong>. These elevated temperatures are not l<strong>in</strong>ked to a physical temperature. There<br />

Gate<br />

Lg<br />

Rg<br />

Ri<br />

T = Tg<br />

+<br />

Vc<br />

-<br />

Cpgs<br />

Cgs<br />

4kTg/Ri∆f<br />

Rgd Cgd<br />

Cpgd<br />

gme jwτ vc<br />

R s<br />

Ls<br />

Source<br />

Rds<br />

T = Td<br />

4kTd/Rds∆f<br />

Figure 2.5: Pospieszalski noise model <strong>in</strong> a small-signal circuit. Resistances (<strong>and</strong> their<br />

thermal noise sources) Ri <strong>and</strong> Rds are at elevated temperatures Tg <strong>and</strong> Td respectively.<br />

29<br />

Rd<br />

Cds<br />

Ld<br />

Cpds<br />

Dra<strong>in</strong>


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

have been attempts to l<strong>in</strong>k these noise temperatures to the device physics, but none<br />

were viewed as successful. The noise can be described by the noise temperatures Tg<br />

<strong>and</strong> Td as:<br />

�<br />

f<br />

Fm<strong>in</strong> = 1+2<br />

Ropt =<br />

+ 2f<br />

� 2 RiTd<br />

fT RdsTa<br />

�<br />

�<br />

�<br />

�RiTgTd<br />

+<br />

fTTa Rds<br />

�<br />

��<br />

�<br />

� 2<br />

� fT TgRiRds<br />

1<br />

f<br />

Xopt =<br />

ωCgs<br />

Rn = TgRi<br />

Ta<br />

+<br />

Td<br />

� f<br />

fT<br />

+ R 2 i<br />

� 2 R 2 i T 2 d<br />

R 2 ds<br />

(2.5.11)<br />

(2.5.12)<br />

(2.5.13)<br />

Td<br />

TaRdsg2 (1 + ω<br />

m<br />

2 C 2 gsR 2 i ) (2.5.14)<br />

After measur<strong>in</strong>g S-parameters <strong>and</strong> noise parameters, the noise temperatures can be<br />

extracted by solv<strong>in</strong>g the above equations. A Matlab script us<strong>in</strong>g equations 2.5.11<br />

<strong>and</strong> 2.5.12 to perform this calculation can be found <strong>in</strong> appendix C. A criteria that<br />

Pospieszalski mentions for the model to work is<br />

1 ≤ 4 Ropt<br />

Rn(F − 1)<br />

< 2 (2.5.15)<br />

The lower limit is fundamental <strong>and</strong> is because the noise of a 2-port modeled by a pair<br />

of noise sources must be Hermitian <strong>and</strong> non-negative def<strong>in</strong>ite as shown by Pospieszal-<br />

ski [27]. The upper limit comes from the model itself [1]. As with the other models, it<br />

can be used to determ<strong>in</strong>e the match for noise <strong>and</strong> expected noise at other frequencies.<br />

The model is criticized for its dependence on Ri as a thermal noise source, which is<br />

30


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

hard to determ<strong>in</strong>e precisely, <strong>and</strong> for the noise temperatures not hav<strong>in</strong>g a physical basis.<br />

As with the other models, the noise parameters must be measured prior to model<strong>in</strong>g.<br />

Thus the model does not “predict” noise. In addition, the noise temperatures change<br />

with device bias [28]. How they change with bias is reported <strong>in</strong> the literature, but not<br />

modeled or predicted. Information for an ADS project that conta<strong>in</strong>s a small-signal<br />

circuit with this type of noise model<strong>in</strong>g <strong>and</strong> that calculates noise figure can be found<br />

<strong>in</strong> appendix A.<br />

2.5.4 Pospieszalski <strong>and</strong> Correlated <strong>Noise</strong> Models Applied to Al-<br />

<strong>GaN</strong>/<strong>GaN</strong> <strong>HEMTs</strong><br />

Based on the above discussion, the Pospieszalski <strong>and</strong> CN models are useful for:<br />

1. Model<strong>in</strong>g device noise figure <strong>in</strong> a circuit simulator versus frequency <strong>and</strong> <strong>in</strong>put<br />

match<br />

2. Devices that are stable <strong>and</strong> well-characterized, such that the noise variables of<br />

the model (Tg, Td, R, P, C) do not change<br />

3. Predict<strong>in</strong>g noise at frequencies outside the range of available measurement equipment.<br />

4. When the bias <strong>in</strong> a design will not be much different from what gives the best<br />

NF performance<br />

Thus, they are useful for basic LNA designs as shown <strong>in</strong> [17, 29]. The application of<br />

these models to <strong>GaN</strong>-based <strong>HEMTs</strong> is relatively new, with very few publications [25,<br />

28,30].<br />

In table 2.2, the Pospieszalski <strong>and</strong> CN model have been applied to devices from six<br />

samples. The fτ <strong>and</strong> fmax of the different devices at the bias<strong>in</strong>g for best NFm<strong>in</strong> are<br />

31


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

listed along with the bias at the top of the table. The next section conta<strong>in</strong>s the mea-<br />

sured noise parameters, obta<strong>in</strong>ed as described <strong>in</strong> § 3.3, along with the ga<strong>in</strong> that could<br />

be expected from the device hav<strong>in</strong>g an <strong>in</strong>put match of Γopt <strong>and</strong> an output match for<br />

small-signal ga<strong>in</strong>. These noise measurements, along with the small-signal parameters<br />

from table 2.1, were used to calculate the noise variables for both the Pospieszalski<br />

<strong>and</strong> a CN model. These models’ variables are listed <strong>in</strong> table 2.2 with the predicted<br />

noise parameters us<strong>in</strong>g each model at 5 GHz. The fit is generally very good. Both<br />

models predict NFm<strong>in</strong> <strong>and</strong> |Γopt| well. The Pospieszalski model does not predict � Γopt<br />

as well as the other parameters, <strong>and</strong> the CN model has trouble with predict<strong>in</strong>g rn.<br />

From equation 2.5.13, it is seen that the Pospieszalski model’s prediction of Xopt only<br />

depends on Cgs. A better prediction agreement would be expected. An explanation for<br />

this discrepancy will be offered <strong>in</strong> § 3.5.4. Both these models fit the noise parameters<br />

versus frequency, shown <strong>in</strong> figure 2.6 for one of the sets of measurements <strong>in</strong> table 2.2.<br />

Variations <strong>in</strong> the predicted Tg of Pospieszalski’s model is large, a factor of 4. This<br />

is likely because of the differences <strong>in</strong> Ri <strong>in</strong> the measurements of table 2.1, but they<br />

change by only a factor of 2. This high sensitivity to Ri might make the model difficult<br />

to use <strong>in</strong> predict<strong>in</strong>g noise. However, it should be stressed that the model works well<br />

without the need for correlation.<br />

32


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

�<br />

�<br />

�<br />

Sample: 15% 25% 25% 27% 35% 35%<br />

on SiC w/AlN<br />

Ids @NFm<strong>in</strong> 20 13 10 19 11 10<br />

Vds @NFm<strong>in</strong> 4 4 7 4 5 4<br />

fτ @NFm<strong>in</strong> 19.7 19.4 21.4 23.3 22.3 21.3<br />

fmax @NFm<strong>in</strong> 30.2 33.6 39 42.9 43.7 40.4<br />

Measure <strong>Noise</strong> Parameters <strong>and</strong> Associated Ga<strong>in</strong><br />

NFm<strong>in</strong> (dB) 1.14 1.13 1.1 1.16 1.15 1.18<br />

rn 0.772 0.707 0.648 0.83 0.796 0.723<br />

Γopt 0.727 0.716 0.746 0.774 0.76 0.733<br />

Γopt 20.5 21.3 23.2 19.8 19.5 23.9<br />

Ga<strong>in</strong>assoc. (dB) 14 13.7 14.6 14.3 14 12.7<br />

Us<strong>in</strong>g Pospieszalski Model<br />

NFm<strong>in</strong> (dB) 1.18 1.1 1.1 1.04 1.07 1.03<br />

rn 0.749 0.597 0.729 0.81 0.839 0.721<br />

Γopt 0.702 0.666 0.715 0.744 0.744 0.72<br />

Γopt 25.7 27.3 26.3 22.7 22.3 26.9<br />

Tg (K) 607 292 250 358 462 988<br />

Td (K) 3445 3018 3893 3469 4454 3966<br />

Us<strong>in</strong>g a Correlated <strong>Noise</strong> Model<br />

NFm<strong>in</strong> (dB) 1.23 1.3 1.21 1.28 1.25 1.33<br />

rn 1.21 0.989 0.831 1.04 0.977 0.95<br />

Γopt 0.806 0.758 0.778 0.797 0.783 0.766<br />

Γopt �<br />

i<br />

19.2 21 22.5 19.6 19.2 23.6<br />

2 �� �<br />

−24 A2<br />

g 10 Hz<br />

〈i<br />

7.05 12.00 3.55 7.09 7.23 6.74<br />

2 d〉 � �<br />

−22 A2<br />

10 Hz 4.56 5.60 4.14 5.93 4.79 5.36<br />

|C| 0.802 0.799 0.821 0.803 0.759 0.739<br />

� C (degrees) 103 101 129 114 114 118<br />

Table 2.2: Comparison of Pospieszalski <strong>and</strong> Correlated <strong>Noise</strong> models to measured<br />

data. Devices have a gate geometry of 0.7 × 150 µm .<br />

33


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

NF m<strong>in</strong> (dB)<br />

r n<br />

|Γ opt |<br />

2.6<br />

2.4<br />

2.2<br />

2.0<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

1.0<br />

0.8<br />

1.8<br />

1.5<br />

1.2<br />

0.9<br />

0.6<br />

0.3<br />

0.0<br />

1.0<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0.0<br />

Measured<br />

Pospieszalski Model<br />

CN Model<br />

4 6 8 10 12<br />

4 6 8 10 12<br />

Measured<br />

Pospieszalski<br />

CN Model<br />

4 6 8 10 12<br />

Frequency (GHz)<br />

Figure 2.6: Comparison of Correlated <strong>Noise</strong> <strong>and</strong> Pospieszalski models to measured<br />

noise parameters versus frequency.<br />

34<br />

180<br />

150<br />

120<br />

90<br />

60<br />

30<br />

0<br />

Angle Γopt (deg.)


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Lee’s study of the CN models to <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> experimentally f<strong>in</strong>ds that the<br />

correlation tends to a constant of 0.7 [25]. The average value for |C| found <strong>in</strong> table<br />

2.2 is 0.718, <strong>in</strong> agreement with Lee. The average phase was found to be 110 ◦ , which<br />

is higher than the expected 90 ◦ . However, when Lee de-embeds contributions from<br />

extr<strong>in</strong>sic shot <strong>and</strong> thermal noise sources he f<strong>in</strong>ds the phase to be very close to 90 ◦ [25].<br />

The average gate noise from table 2.2 is 7.3 pA 2 /Hz <strong>and</strong> the average dra<strong>in</strong> noise is 506<br />

pA 2 /Hz.<br />

Average values for the Pospieszalski noise variables are Tg of 493 K <strong>and</strong> Td of<br />

3708 K. These numbers are only approximately close to those found <strong>in</strong> [28]; however,<br />

<strong>in</strong> that reference the devices which had a gate length of 0.35 µm (compared to 0.7 µm<br />

here) had an fτ of only 30 GHz, <strong>and</strong> thus were excessively noisy. Despite the accu-<br />

racy <strong>in</strong> the Pospieszalski model, the condition <strong>in</strong> equation 2.5.15 was not met for the<br />

model<strong>in</strong>g <strong>in</strong> table 2.2. The upper limit was broken; values of 8 were common.<br />

2.6 A Proposed <strong>Noise</strong> Figure Model<br />

A model that was developed for some of the noise studies <strong>in</strong> chapter 3 is now pre-<br />

sented. It uses some of the formulation of the CN models, but without correlation <strong>and</strong><br />

replacement of the <strong>in</strong>duced gate noise with a shot noise source. In addition, this model<br />

can give accurate noise parameter results without prior noise parameter measurements.<br />

35


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Gate<br />

R g +R i +R s<br />

I s<br />

+<br />

Vc -<br />

E r<br />

C' gs<br />

Source<br />

g' m v c<br />

R ds<br />

Dra<strong>in</strong><br />

Figure 2.7: A simplified HEMT circuit model <strong>in</strong>clud<strong>in</strong>g noise sources.<br />

2.6.1 Setup Details<br />

A circuit schematic of the model is represented <strong>in</strong> figure 2.7 with six small-signal<br />

parameters <strong>and</strong> three noise sources. A full small-signal model is still used for small-<br />

signal parameter extraction from a HEMT, but only Rg, Rs, Ri, Cgs, gm, <strong>and</strong> Rds are<br />

used. The noise sources are Er, Is, <strong>and</strong> Ic. Er is thermal noise from the two parasitic<br />

resistances <strong>and</strong> channel resistance. Its voltage spectral density is written as<br />

�<br />

E 2 �<br />

r =4kT (Rs + Rg + Ri)∆f (2.6.1)<br />

Is is a shot noise source. The DC current used for it is from a three-term<strong>in</strong>al measure-<br />

ment at the bias of S-parameter extraction. This source can be represented between<br />

gate-<strong>and</strong>-source or gate-<strong>and</strong>-dra<strong>in</strong> with the result<strong>in</strong>g derivations be<strong>in</strong>g the same. The<br />

current spectral density of this shot noise term is<br />

�<br />

I 2 �<br />

s =2qIgs∆f (2.6.2)<br />

36<br />

I c


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Igs is the measured gate leakage current.<br />

The f<strong>in</strong>al noise source is Ic. It represents distributed thermal noise <strong>in</strong> the channel,<br />

<strong>and</strong> is the same as found by van der Ziel [3]<br />

�<br />

I 2 �<br />

c =4kTΓgm∆f (2.6.3)<br />

Γ is a bias dependent quantity. For a saturated <strong>GaN</strong> HEMT <strong>and</strong> most other FETs, it<br />

is a constant value of 2/3. It will be derived <strong>in</strong> detail <strong>in</strong> § 2.6.3. All three mentioned<br />

noise sources will be assumed to be uncorrelated, greatly simplify<strong>in</strong>g the math.<br />

The derivation of the noise parameters is also simplified if the source resistance<br />

can be transformed up from the source node to Ri. This is often done <strong>in</strong> device<br />

small-signal circuit calculations through the use of source degeneration, 2 illustrated<br />

<strong>in</strong> figure 2.8. Source degeneration modifies the values of Cgs <strong>and</strong> gm as follows [31]<br />

C ′ gs =<br />

Cgs<br />

1+gmRs<br />

gm<br />

g ′ m =<br />

1+gmRs<br />

(2.6.4)<br />

(2.6.5)<br />

with Rs, Cgs, <strong>and</strong> gm be<strong>in</strong>g values extracted from the device <strong>and</strong> C ′ gs <strong>and</strong> g ′ m the<br />

degenerate values that appear <strong>in</strong> figure 2.7 <strong>and</strong> the math that follows. These equations<br />

hold for ωC ′ gs Rs ≪ 1. The thermal noise of Rs, ERs, can be moved to the <strong>in</strong>put with<br />

the same condition that ωC ′ gsRs ≪ 1. As is seen, the effect of this degeneration is a<br />

reduction <strong>in</strong> the effective Cgs, <strong>and</strong> gm. Because the gate-dra<strong>in</strong> capacitance, Cgd, is not<br />

2The technique is also called emitter degeneration, depend<strong>in</strong>g on if the device <strong>in</strong> question is a FET<br />

or BJT type of device.<br />

37


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Gate<br />

+<br />

Vc -<br />

R i<br />

Cgs<br />

Source<br />

R s<br />

E Rs<br />

Dra<strong>in</strong><br />

g mv c<br />

Gate<br />

E Rs<br />

+<br />

Vc -<br />

R i+R s<br />

C'gs<br />

Source<br />

Figure 2.8: Cartoon show<strong>in</strong>g the effect of source degeneration.<br />

Dra<strong>in</strong><br />

g' mv c<br />

used <strong>in</strong> this model, Ri, Rg, <strong>and</strong> Rs can be lumped together as one resistance, further<br />

simplify<strong>in</strong>g the derivation.<br />

Several assumptions have been made. The most worrisome is neglect<strong>in</strong>g the <strong>in</strong>-<br />

tr<strong>in</strong>sic feedback capacitance, Cgd. Pospieszalski has shown [1] at typical noise mea-<br />

surement frequencies remov<strong>in</strong>g Cgd from the model<strong>in</strong>g <strong>in</strong>troduces only a small error<br />

(less than the typical precision of a noise measurement setup). So long as the mea-<br />

surements are made far enough from the device fτ, it should not cause alarm. As an<br />

example, a device with an fτ of 50 GHz will probably not be used for LNA applica-<br />

tions greater than half the device unity current ga<strong>in</strong> frequency. A derivation <strong>in</strong>clud<strong>in</strong>g<br />

Cgd was carried out but the expressions were too complicated for <strong>in</strong>sight to be ga<strong>in</strong>ed<br />

from them.<br />

Another important assumption is that the noise sources are uncorrelated. It is not<br />

38


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

be<strong>in</strong>g implied that physically all the noise processes are not <strong>in</strong>fluenc<strong>in</strong>g one another.<br />

However, to have a simple enough <strong>and</strong> <strong>in</strong>sightful model, it is a necessary assumption.<br />

As mentioned earlier, it has already been shown that not us<strong>in</strong>g correlation can give<br />

accurate results [1,32].<br />

F<strong>in</strong>ally, neglect<strong>in</strong>g many of the extr<strong>in</strong>sic parasitics is a m<strong>in</strong>or assumption. The<br />

<strong>in</strong>ductances (Lg, Ls, Ld) <strong>and</strong> capacitances (Cpgs, Cpgd, Cpds) do not affect the noise<br />

parameters as much as the <strong>in</strong>tr<strong>in</strong>sic parameters. This will be seen <strong>in</strong> § 2.6.5. Also, the<br />

thermal noise of Rd pales <strong>in</strong> comparison to the noise generated by the channel.<br />

2.6.2 Derivation of <strong>Noise</strong> Parameters<br />

A load, RL, signal generator (not a noise source), Vgen, <strong>and</strong> generator impedance,<br />

Zgen = Rgen + Xgen, are added to the model, redrawn <strong>in</strong> figure 2.9. Also, for conve-<br />

nience, the collected resistances are renamed to<br />

X gen<br />

V gen<br />

R gen<br />

E gen<br />

Z <strong>in</strong><br />

R <strong>in</strong><br />

I s<br />

+<br />

Vc -<br />

E r<br />

C' gs<br />

g' m v c<br />

Figure 2.9: Circuit model used for deriv<strong>in</strong>g noise figure.<br />

39<br />

R ds<br />

I c<br />

R L<br />

+<br />

Vout -


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

R<strong>in</strong> = Rg + Ri + Rs<br />

(2.6.6)<br />

Assume for the moment that the <strong>in</strong>put <strong>and</strong> output are matched for power: RL = Rds<br />

<strong>and</strong> Zgen = Z∗ <strong>in</strong> . equation 2.4.1 can then be rewritten<br />

F = (S/N)<strong>in</strong><br />

(S/N)out<br />

=<br />

|Vgen| 2<br />

4Zgen /〈v<strong>in</strong>,noisev∗ <strong>in</strong>,noise〉<br />

4Zgen<br />

|AvVgen| 2<br />

4Rds /〈vout,noisev∗ out,noise〉<br />

4Rds<br />

= 1<br />

|Av| 2<br />

�<br />

vout,noisev ∗ out,noise<br />

�<br />

v<strong>in</strong>,noisev ∗ <strong>in</strong>,noise<br />

�<br />

� (2.6.7)<br />

with Av def<strong>in</strong>ed as the voltage ga<strong>in</strong>, vsignal,out = Avvgen. An <strong>in</strong>terest<strong>in</strong>g po<strong>in</strong>t is<br />

that because noise figure is a ratio of powers, the output <strong>and</strong> <strong>in</strong>put match cancel <strong>in</strong><br />

equation 2.6.7. This does not mean that the source impedance has no effect on noise<br />

figure nor that a matched load has no effect on the output power. It means that the ratio<br />

of signal-to-noise at the <strong>in</strong>put only or the output only is not affected by the match<strong>in</strong>g<br />

conditions.<br />

We need to def<strong>in</strong>e the <strong>in</strong>put noise voltage. This may be from an active device, or<br />

it may be unknown, but the derivation depends only on signal-to-noise ratio <strong>and</strong> the<br />

source impedance. Here thermal noise of Zgen, represented as Egen <strong>in</strong> figure 2.9, is<br />

used. The <strong>in</strong>put noise can be written as<br />

�<br />

v 2 �<br />

<strong>in</strong> =4kTRgen<br />

(2.6.8)<br />

The b<strong>and</strong>width (∆ f, assumed to be 1 Hz) will no longer be <strong>in</strong>cluded. The noise figure<br />

is now<br />

F = 1<br />

|Av| 2<br />

�<br />

vout,noisev∗ �<br />

out,noise<br />

4kTRgen<br />

40<br />

(2.6.9)


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

The output noise, vout,noise, will depend on the contributions from the <strong>in</strong>dividual <strong>in</strong>-<br />

ternal noise sources <strong>in</strong> the device <strong>and</strong> the <strong>in</strong>put noise. Because the HEMT is modeled<br />

by a small-signal circuit with noise sources, the output noise can be calculated with<br />

st<strong>and</strong>ard small-signal methods.<br />

The easiest contribution to the output noise to calculate is the channel noise. It is<br />

merely Ohm’s law, Ic(Rds||RL). Théven<strong>in</strong> <strong>and</strong> Norton representations can be used<br />

such that the voltage ga<strong>in</strong>s from Vgen, Egen, <strong>and</strong> Er to the output will be the same.<br />

This is the ga<strong>in</strong> of a common source configuration as found <strong>in</strong> circuit textbooks [31]:<br />

vout<br />

Av =<br />

Egen,Vgen,Er<br />

= − ωτ<br />

ω<br />

jRL||Rds<br />

(Rgen + R<strong>in</strong>)+j(Xgen − 1/(ωC ′ gs))<br />

with ω be<strong>in</strong>g the angular frequency (2π×frequency) <strong>and</strong> def<strong>in</strong><strong>in</strong>g<br />

ωτ = g ′ m/C ′ gs<br />

(2.6.10)<br />

(2.6.11)<br />

The other ga<strong>in</strong> needed is the shot noise contribution. Us<strong>in</strong>g Norton <strong>and</strong> Théven<strong>in</strong><br />

equivalent circuits aga<strong>in</strong>, it is easily found.<br />

As = vout<br />

Is<br />

= − ωτ<br />

ω<br />

jRL||Rds(Rgen + jXgen)<br />

(Rgen + R<strong>in</strong>)+j(Xgen − 1/(ωC ′ gs))<br />

As expected, it is similar to Av. The total output noise is then expressed as<br />

vout,noise = Av(Egen + Er) − (Rds||RL)Ic + AsIs<br />

(2.6.12)<br />

(2.6.13)<br />

but what is needed is the spectral density, <strong>and</strong> thus, noise power. Tak<strong>in</strong>g the complex<br />

conjugate, followed by a statistical average, gives<br />

�<br />

vout,noisev∗ � ��<br />

2<br />

out,noise = |Av| E2 �<br />

gen + 〈E2 r 〉 �<br />

+ | (Rds||RL) | 2 〈I2 c 〉<br />

41


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

+|As| 2 〈I2 �<br />

2<br />

s 〉 + |Av| ErE ∗ �<br />

gen + A∗ �<br />

vRL||Rds IcE∗ �<br />

gen<br />

+ ... 10 other cross terms (2.6.14)<br />

It is not possible to calculate the various correlations. Most cross-terms will be zero<br />

(shot <strong>and</strong> thermal noise will not be correlated). For simplicity, all sources are assumed<br />

to be uncorrelated. Equation 2.6.14 is then reduced to<br />

�<br />

vout,noisev ∗ � ��<br />

2<br />

out,noise = |Av| E 2 �<br />

gen + �<br />

E 2 ��<br />

�<br />

2<br />

r + | (Rds||RL) | I 2 � �<br />

2<br />

c + |As| I 2 �<br />

s<br />

(2.6.15)<br />

F<strong>in</strong>ally, comb<strong>in</strong><strong>in</strong>g equations 2.6.8, 2.6.9, 2.6.10, 2.6.12, 2.6.13, <strong>and</strong> equation 2.6.15<br />

we obta<strong>in</strong> an expression for noise figure:<br />

F =1+ R<strong>in</strong><br />

with<br />

Rgen<br />

+ b R2 gen + X 2 gen<br />

Rgen<br />

+ a<br />

�<br />

�<br />

�<br />

�<br />

Rgen � R<strong>in</strong><br />

�<br />

+ Rgen + j Xgen − 1<br />

ωC ′ ��<br />

���� 2<br />

gs<br />

(2.6.16)<br />

a = g ′ mΓ � �<br />

ω 2<br />

ωτ<br />

b =<br />

qIgs<br />

2kTreference<br />

(2.6.17)<br />

(2.6.18)<br />

The factor of 1 is the <strong>in</strong>put noise contribution to the noise figure. If the device does<br />

not generate noise, noise figure would be 1. There would then be no degradation <strong>in</strong><br />

the signal to noise ratio go<strong>in</strong>g <strong>in</strong>to <strong>and</strong> com<strong>in</strong>g out of the device. Equation 2.6.16<br />

describes the noise figure <strong>in</strong> terms of the match provided to the device, the device<br />

small-signal parameters, <strong>and</strong> the measured gate-leakage. There are no fitt<strong>in</strong>g param-<br />

42


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

eters. This is not the m<strong>in</strong>imum noise figure. To determ<strong>in</strong>e it, we f<strong>in</strong>d the source<br />

impedance that m<strong>in</strong>imizes equation 2.6.16. This is found by tak<strong>in</strong>g partial derivatives<br />

<strong>and</strong> solv<strong>in</strong>g equal to zero.<br />

�<br />

�<br />

F = Fm<strong>in</strong><br />

� ∂F ∂F =0, ∂Rgen ∂Xgen =0<br />

(2.6.19)<br />

This leads to the follow<strong>in</strong>g expressions for the optimal source impedance that provides<br />

the m<strong>in</strong>imum noise figure:<br />

�<br />

�<br />

�<br />

Ropt =<br />

Xopt = 1<br />

ωCgs<br />

a<br />

a + b<br />

�1+aR<strong>in</strong> a + b R<strong>in</strong> + ab<br />

(a + b) 2<br />

1<br />

(ωCgs) 2<br />

(2.6.20)<br />

(2.6.21)<br />

The last parameter, the noise resistance, can be obta<strong>in</strong>ed by us<strong>in</strong>g equation 2.4.3<br />

<strong>and</strong> the equation for noise figure, 2.6.16, at two different source impedances. The<br />

most convenient to choose are the m<strong>in</strong>imum noise figure <strong>and</strong> when the source is at<br />

its reference impedance, Γgen =0. We shall call the later FZ0. Enter<strong>in</strong>g this <strong>in</strong>to<br />

equation 2.4.3 <strong>and</strong> rearrang<strong>in</strong>g,<br />

Rn = Z0<br />

4 (FZ0<br />

�<br />

�<br />

� 1<br />

− Fm<strong>in</strong>) �1+<br />

�<br />

�<br />

�2<br />

�<br />

�<br />

Γopt �<br />

(2.6.22)<br />

An explicit expression is complicated. A Matlab script that implements all four of<br />

the noise parameters can be found <strong>in</strong> appendix B. These equations will be discussed<br />

43


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

more <strong>in</strong> § 2.6.5, <strong>and</strong> used <strong>in</strong> chapter 3. For the moment, let us turn our attention to the<br />

factor Γ, <strong>and</strong> the derivation of the channel noise source.<br />

2.6.3 Derivation of Dra<strong>in</strong> <strong>Noise</strong> Source<br />

We will quickly go through the derivation of Γ (<strong>and</strong>, therefore, the channel noise<br />

source) to show that it is not a fitt<strong>in</strong>g parameter. Van der Ziel has shown the formula-<br />

tion for a MOSFET [3]. Here, it will be explicitly done for a HEMT. The formulation<br />

<strong>in</strong>volves the DC dra<strong>in</strong> current, Id, so it is needed as well. Assume for the moment that<br />

the device is biased <strong>in</strong> the l<strong>in</strong>ear region (no velocity saturation). Follow<strong>in</strong>g arguments<br />

as found <strong>in</strong> [11,33], the dra<strong>in</strong> current can be written <strong>in</strong> general as<br />

Id = g(Vx) dVx<br />

dx<br />

= qµWns(x) dVx<br />

dx<br />

(2.6.23)<br />

W is the device width (cm), ns(x) is the sheet charge of the 2DEG (cm−2 ), <strong>and</strong> µ<br />

is the mobility � cm2 �<br />

. Vx is the potential difference at a distance x from the source,<br />

Vs<br />

relative to the source. g(Vx) is the channel conductivity per unit length at some po<strong>in</strong>t<br />

along the channel, which depends on Vx.<br />

An expression is needed for ns(x) <strong>in</strong> terms of Vx. First, an effective capacitance can<br />

be def<strong>in</strong>ed [11,33]<br />

C = Q<br />

V<br />

= εB<br />

d +∆d<br />

(2.6.24)<br />

with εB <strong>and</strong> d be<strong>in</strong>g the dielectric permittivity <strong>and</strong> thickness of the <strong>Al<strong>GaN</strong></strong> layer re-<br />

spectively <strong>and</strong> ∆d the centroid of the electron wave functions <strong>in</strong> the quantum well<br />

44


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

(that is, the 2DEG). The charge of this capacitance is qns(x), with a voltage along the<br />

channel of Vg − Vt − Vx. Vt is the threshold voltage. Comb<strong>in</strong><strong>in</strong>g all this together <strong>and</strong><br />

rearrang<strong>in</strong>g<br />

ns(x) =<br />

<strong>and</strong> substitut<strong>in</strong>g <strong>in</strong>to equation 2.6.23<br />

We can now def<strong>in</strong>e<br />

εB<br />

q(d +∆d) (Vg − Vt − Vx) (2.6.25)<br />

Id = µεBW<br />

d +∆d (Vg − Vt − Vx) dVx<br />

dx<br />

(2.6.26)<br />

g(Vx) = µεBW<br />

d +∆d (Vg − Vt − Vx) (2.6.27)<br />

Its usefulness will soon be apparent. Id can now be found. Because of cont<strong>in</strong>uity, the<br />

DC current enter<strong>in</strong>g <strong>and</strong> leav<strong>in</strong>g the device must be the same. Integrat<strong>in</strong>g over the<br />

length of the device, L,<br />

which gives<br />

� L<br />

0<br />

Iddx = IdL =<br />

� Vd<br />

0<br />

g(Vx)dVx<br />

Id = µεBW<br />

�<br />

(Vg − Vt)Vd −<br />

L(d +∆d)<br />

From this, the transconductance (gm)<br />

gm = ∂Id<br />

∂Vg<br />

= µεBW<br />

L(d +∆d) Vd<br />

45<br />

�<br />

2 Vd 2<br />

(2.6.29)<br />

(2.6.28)<br />

(2.6.30)


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

<strong>and</strong> the saturation current, Id,sat, can be found. The saturation current is found by<br />

differentiat<strong>in</strong>g equation 2.6.29 with respect to Vd <strong>and</strong> sett<strong>in</strong>g it equal to zero. This<br />

returns the voltage of Vd that maximizes Id, which we will call Vd,sat. However, if the<br />

device has a short channel, the velocity may saturate before the above condition. Then<br />

the voltage that the current saturates is the lesser of the follow<strong>in</strong>g two quantities:<br />

or<br />

Vd,sat = Vg − Vt<br />

Vd,sat = EcritLg<br />

(2.6.31)<br />

(2.6.32)<br />

Ecrit is the critical field strength <strong>and</strong> Lg is the device gate length. Putt<strong>in</strong>g equa-<br />

tion 2.6.31 back <strong>in</strong>to equation 2.6.29, we f<strong>in</strong>d the saturation current result<strong>in</strong>g from<br />

channel p<strong>in</strong>ch off.<br />

Id,sat = 1 µεBW<br />

2 L(d +∆d) (Vg − Vt) 2<br />

(2.6.33)<br />

equation 2.6.29 only works for a device <strong>in</strong> the l<strong>in</strong>ear region until the device saturates,<br />

Id = Id,sat. Beyond this, the current can be approximated to be the same as Id,sat (if<br />

short-channel effects are ignored).<br />

We will follow the work of van der Ziel to derive the channel noise [3]. Assume<br />

that a thermal voltage noise source, vn, creates a dra<strong>in</strong> noise current fluctuation, ∆Id,<br />

along the distributed channel. This is shown <strong>in</strong> figure 2.10. The thermal noise source<br />

can be written as<br />

〈vnv ∗ n〉 = 4kT<br />

g(V )<br />

46<br />

(2.6.34)


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Similar to equation 2.6.23, we can write an expression for the dra<strong>in</strong> current fluctua-<br />

tions:<br />

∆Id = g(V ) dV<br />

dx + vng(V ) (2.6.35)<br />

with V = Vx +∆V (x, t). Separat<strong>in</strong>g the derivative <strong>and</strong> <strong>in</strong>tegrat<strong>in</strong>g gives<br />

� L<br />

0<br />

∆Iddx =<br />

� L<br />

0<br />

g(V )dV +<br />

� L<br />

0<br />

vng(V )dx (2.6.36)<br />

An expression for the noise only is desired. If we consider the dra<strong>in</strong> AC shorted, the<br />

first term on the right of the equality <strong>in</strong> equation 2.6.36 will be zero. Cont<strong>in</strong>u<strong>in</strong>g we<br />

have<br />

∆IdL =<br />

Tak<strong>in</strong>g the spectral density we have<br />

� L<br />

〈∆Id∆I ∗ d〉 = 1<br />

L2 � L<br />

〈vnv<br />

0<br />

∗ n〉 g 2 (V )dx = 4kT<br />

L2 0<br />

vng(V )dx (2.6.37)<br />

� L<br />

0<br />

g(V )dx (2.6.38)<br />

with the use of equation 2.6.35. The variable be<strong>in</strong>g <strong>in</strong>tegrated over, x, can be changed<br />

to V through the use of equation 2.6.23. It is rearranged here as<br />

V x (x) + ∆V x (x,t)<br />

g(V )<br />

dx = dV (2.6.39)<br />

g(V)<br />

_<br />

Id<br />

+<br />

I d + ∆I d (t)<br />

... ...<br />

_<br />

+<br />

V n<br />

V x (x+∆x) + ∆V x (x+∆x,t)<br />

Figure 2.10: Cartoon used for deriv<strong>in</strong>g the channel noise.<br />

47<br />

_<br />

+


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Substitut<strong>in</strong>g <strong>and</strong> chang<strong>in</strong>g the limits of <strong>in</strong>tegration we arrive at the follow<strong>in</strong>g key equa-<br />

tion:<br />

〈∆Id∆I ∗ �<br />

d 〉 = i 2 �<br />

d = 4kT<br />

L2 � Vd<br />

g<br />

Id 0<br />

2 (Vx)dVx<br />

(2.6.40)<br />

It is now clear the usefulness of know<strong>in</strong>g Id <strong>and</strong> g(Vx) explicitly. Comb<strong>in</strong><strong>in</strong>g equa-<br />

tions 2.6.27, 2.6.29, <strong>and</strong> 2.6.40 leads to an expression for the noise<br />

with<br />

Γ=<br />

�<br />

i 2 �<br />

d =4kTgmΓ (2.6.41)<br />

1 − Vd<br />

Vg−Vt +<br />

1 − 1<br />

V 2 d<br />

3(Vg−Vt) 2<br />

Vd<br />

2 Vg−Vt<br />

(2.6.42)<br />

When the device saturates, the value of Vd that saturates the current is used <strong>in</strong><br />

equation 2.6.42 <strong>in</strong>stead of the value past saturation. Equation 2.6.42 then becomes<br />

Γ=2/3. A plot of Γ for different gate <strong>and</strong> dra<strong>in</strong> bias<strong>in</strong>gs is <strong>in</strong> figure 2.11. A thresh-<br />

old voltage of -6 V has been assumed. It can be <strong>in</strong>ferred from this graph that a device<br />

operat<strong>in</strong>g <strong>in</strong> the l<strong>in</strong>ear region has a larger Γ.<br />

One f<strong>in</strong>al note: while there was no evidence of hot electron effects <strong>in</strong> devices pre-<br />

sented <strong>in</strong> this chapter, van der Ziel mentions that they will change the expected noise.<br />

Instead of equation 2.6.40, the follow<strong>in</strong>g equation would have to be used [3]<br />

�<br />

i 2 �<br />

d = 4kT<br />

L2 � Vd<br />

Te(x)g(Vx)<br />

Id 0<br />

2 dVx<br />

(2.6.43)<br />

Te is the effective electron temperature. He gives an empirical relationship for it <strong>in</strong><br />

48


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

Γ<br />

1<br />

0.95<br />

0.9<br />

0.85<br />

0.8<br />

0.75<br />

0.7<br />

V g =-6<br />

V g =0<br />

Decreas<strong>in</strong>g V g<br />

0.65<br />

0 1 2 3 4 5 6 7 8<br />

V ds (V)<br />

Figure 2.11: Variation <strong>in</strong> Γ for different dra<strong>in</strong> <strong>and</strong> gate voltages. The gate voltage is<br />

varied from -6 to 0 V by steps of 1 V. The threshold voltage is -6 V.<br />

terms of the electric field relative to the critical field value:<br />

Te(x) =<br />

�<br />

1+ E<br />

Ecrit<br />

�n<br />

(2.6.44)<br />

n is either 0, 1, or 2. If it is zero, then we have equation 2.6.40. This leads to values of<br />

Γ a few times 1 (as opposed to less than 1 when n = 0).<br />

2.6.4 <strong>Noise</strong> Parameter Scal<strong>in</strong>g<br />

To the above model, scal<strong>in</strong>g can be added for vary<strong>in</strong>g gate width <strong>and</strong> number of gate<br />

f<strong>in</strong>gers. Simple l<strong>in</strong>ear scal<strong>in</strong>g based on a start<strong>in</strong>g device width of 150 µm <strong>and</strong> 1 gate<br />

f<strong>in</strong>ger was performed. This allows prediction for a comfortable range of usable gate<br />

49


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

f<strong>in</strong>gers <strong>and</strong> widths, but will not be accurate for large devices (1 mm or greater). Based<br />

on [34], the follow<strong>in</strong>g equations can be used to scale the small-signal parameters:<br />

with<br />

Rs,new = Rs,old/s1 Ri,new = Ri,old/s1 Rg,new = Rg,olds2<br />

Cgs,new = Cgs,olds1 Igs,new = Igs,olds1 gm,new = gm,olds1<br />

s1 = Wg,new<br />

Wg,old<br />

s2 = Wg,new/n 2 new<br />

Wg,old/n 2 old<br />

(2.6.45)<br />

(2.6.46)<br />

(2.6.47)<br />

The parameters labeled as new, are the values to be determ<strong>in</strong>ed based on the un-<br />

scaled extracted values, the old parameters. n is the number of gate f<strong>in</strong>gers. The<br />

simulated noise parameters versus gate width for a s<strong>in</strong>gle f<strong>in</strong>gered device are <strong>in</strong> fig-<br />

ure 2.12. With <strong>in</strong>creas<strong>in</strong>g unity-gate width, the noise <strong>in</strong>creases. However, the mag-<br />

nitude of the optimal match decreases, which may make match<strong>in</strong>g simpler. Also, the<br />

noise resistance drops, mak<strong>in</strong>g a design more robust <strong>in</strong> terms of the expected noise.<br />

These predicted results agree with measurements <strong>in</strong> [17]. At small gate widths, Rs<br />

<strong>and</strong> Ri will be large <strong>and</strong> keep a lower bound on NFm<strong>in</strong> (not shown <strong>in</strong> figure 2.12).<br />

The noise <strong>in</strong>creases with a wider gate f<strong>in</strong>ger because of <strong>in</strong>creas<strong>in</strong>g gate resistance <strong>and</strong><br />

gate leakage current (F αIgs,R 2 g <strong>in</strong> equation 2.6.16). So as the width <strong>in</strong>creases, NFm<strong>in</strong><br />

goes from be<strong>in</strong>g limited by Rs <strong>and</strong> Ri to Rg <strong>and</strong> Igs. The noise parameters are also<br />

simulated aga<strong>in</strong>st number of gate f<strong>in</strong>gers <strong>in</strong> figure 2.13. Due to layout constra<strong>in</strong>ts <strong>and</strong><br />

considerations for symmetric designs (particularly CPW implementations), the gate<br />

50


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

NF m<strong>in</strong> (dB)<br />

r n<br />

1.6<br />

1.4<br />

1.2<br />

1<br />

0.8<br />

0 200 400 600<br />

Gate Width (µm)<br />

6<br />

4<br />

2<br />

0<br />

0 200 400 600<br />

Gate Width (µm)<br />

|Γ opt |<br />

Phase Γ opt [deg.]<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0 200 400 600<br />

Gate Width (µm)<br />

(a) (b)<br />

40<br />

30<br />

20<br />

10<br />

0<br />

0 200 400 600<br />

Gate Width (µm)<br />

(c) (d)<br />

Figure 2.12: (a) M<strong>in</strong>imum noise figure, (b) magnitude <strong>and</strong> (d) phase of optimum<br />

source reflection, <strong>and</strong> noise resistance (c) all versus total gate width. Simulation frequency<br />

is 5 GHz.<br />

f<strong>in</strong>ger number was kept to even values or a s<strong>in</strong>gle gate. The total gate width is kept<br />

constant at 150 µm. A change <strong>in</strong> this parameter for small devices does little to affect<br />

the match or the noise resistance. In go<strong>in</strong>g from one to two gate f<strong>in</strong>gers, there is a<br />

slight improvement <strong>in</strong> noise figure. This can be expla<strong>in</strong>ed solely by an improvement<br />

<strong>in</strong> gate resistance. Beyond two f<strong>in</strong>gers, there is little additional benefit. These scal<strong>in</strong>g<br />

51


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

results will be used <strong>in</strong> § 3.5.4 <strong>and</strong> § 3.5.5.<br />

Therefore, the best geometry is a smaller total gate width, to keep Rg <strong>and</strong> Igs from<br />

becom<strong>in</strong>g to large, <strong>and</strong> two or four gate f<strong>in</strong>gers. Too small a device, <strong>and</strong> Rs <strong>and</strong> Ri<br />

will be too large. Many papers published for best NFm<strong>in</strong> performance typically have<br />

device widths of ∼100 µm .<br />

NF m<strong>in</strong> (dB)<br />

r n<br />

1<br />

0.95<br />

0.9<br />

0.85<br />

0 2 4 6 8 10<br />

Gate F<strong>in</strong>gers<br />

0.8<br />

0.78<br />

0.76<br />

0.74<br />

0.72<br />

0.7<br />

0 2 4 6 8 10<br />

Gate F<strong>in</strong>gers<br />

|Γ opt |<br />

(degrees)<br />

Phase Γ opt<br />

0.8<br />

0.79<br />

0.78<br />

0.77<br />

0.76<br />

0 2 4 6 8 10<br />

Gate F<strong>in</strong>gers<br />

(a) (b)<br />

18<br />

17<br />

16<br />

15<br />

14<br />

0 2 4 6 8 10<br />

Gate F<strong>in</strong>gers<br />

(c) (d)<br />

Figure 2.13: (a) M<strong>in</strong>imum noise figure, (b) magnitude <strong>and</strong> (d) phase of optimum<br />

source reflection, <strong>and</strong> noise resistance (c) all versus number of gate f<strong>in</strong>gers for a constant<br />

total gate width.<br />

52


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

2.6.5 Discussion of the Model<br />

The model is now checked aga<strong>in</strong>st measurements, compared to the other noise mod-<br />

els, <strong>and</strong> its limitations are discussed. Figure 2.14 shows noise parameter data for the<br />

35% device of table 2.1 <strong>and</strong> prediction us<strong>in</strong>g the model <strong>in</strong> Matlab <strong>and</strong> a circuit simula-<br />

tion. The fit for all four parameters is very good for both the small-signal simulation<br />

<strong>and</strong> the equations entered <strong>in</strong>to Matlab. The small-signal simulation us<strong>in</strong>g ADS is for a<br />

NF m<strong>in</strong> (dB)<br />

r n<br />

2.5<br />

2<br />

1.5<br />

1<br />

0.5<br />

4 6 8 10 12<br />

Frequency (GHz)<br />

1<br />

0.9<br />

0.8<br />

0.7<br />

0.6<br />

0.5<br />

4 6 8 10 12<br />

Frequency (GHz)<br />

|Γ opt |<br />

Angle Γ opt (Degress)<br />

0.8<br />

0.75<br />

0.7<br />

0.65<br />

0.6<br />

0.55<br />

4 6 8 10 12<br />

Frequency (GHz)<br />

60<br />

50<br />

40<br />

30<br />

20<br />

10<br />

4 6 8 10 12<br />

Frequency (GHz)<br />

Figure 2.14: <strong>Noise</strong> parameters predicted us<strong>in</strong>g the model (solid l<strong>in</strong>e) <strong>and</strong> <strong>in</strong> a full<br />

small-signal circuit simulation (dotted l<strong>in</strong>e) compared to measurements (crosses). The<br />

small-signal parameters are from table 2.1.<br />

53


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

full small-signal model (<strong>in</strong>clud<strong>in</strong>g Cgd) <strong>and</strong> noise sources as <strong>in</strong> figure 2.7. The phase<br />

match of Γopt with the Matlab script could be further improved by <strong>in</strong>clud<strong>in</strong>g the gate<br />

<strong>in</strong>ductance. Only the small-signal parameters <strong>and</strong> gate leakage were needed for the<br />

Matlab script. The measured noise parameters were not needed beforeh<strong>and</strong> as with<br />

the Pospieszalski <strong>and</strong> CN models. This predictive power can help <strong>in</strong> better designs of<br />

noise performance for devices, accurate estimations of the noise <strong>and</strong> the match to the<br />

device, <strong>and</strong> underst<strong>and</strong><strong>in</strong>g differences <strong>in</strong> noise performance for devices as performed<br />

<strong>in</strong> chapter 3.<br />

Of <strong>in</strong>terest is how much different noise sources contribute to the overall noise figure.<br />

This is presented <strong>in</strong> figure 2.15 us<strong>in</strong>g the model at an optimal bias for noise perfor-<br />

mance. The channel thermal noise accounts for half the contribution to noise figure.<br />

The resistances contribute roughly <strong>in</strong> proportion to their values relative to one another<br />

as one might expect; they are all lumped together at the <strong>in</strong>put. However, Rs effectively<br />

degrades the gm <strong>and</strong> Cgs through source degeneration <strong>and</strong> should be m<strong>in</strong>imized. The<br />

gate leakage contributes about 10% for a device with a reasonably low amount of gate<br />

leakage. This can be a much larger contributer, <strong>and</strong> will be discussed <strong>in</strong> § 3.5.4.<br />

Some <strong>in</strong>sightful parallels can be drawn between this model<strong>in</strong>g (equations 2.6.16,<br />

2.6.20, 2.6.21 <strong>and</strong> 2.6.22), <strong>and</strong> Pospieszalski’s model (equations 2.5.11, 2.5.12, 2.5.13,<br />

<strong>and</strong> 2.5.14). The simplest to see is that Xopt is the same for both models if the gate<br />

leakage tends to zero. If the gate leakage is negligible, then the reactance of the match<br />

54


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

R i 18%<br />

Gate<br />

Shot<br />

<strong>Noise</strong><br />

R g 7%<br />

11%<br />

R s 11%<br />

53%<br />

Dra<strong>in</strong><br />

<strong>Noise</strong><br />

Figure 2.15: Relative contributions of different noise sources to the overall noise<br />

figure.<br />

reduces to that predicted by both Pospieszalski <strong>and</strong> Fukui. Not as easy to see is that<br />

Ropt is of similar form for both models if b → 0. The other two parameters cannot be<br />

compared so easily. However, the modeled Rn for both show that it <strong>in</strong>creases slightly<br />

with <strong>in</strong>creas<strong>in</strong>g frequency <strong>and</strong> both predict that Fm<strong>in</strong> depends on the square of the<br />

frequency (which will be l<strong>in</strong>ear <strong>in</strong> plots of NFm<strong>in</strong> vs. frequency).<br />

All the noise models discussed so far have a noise source at the gate <strong>and</strong> dra<strong>in</strong>.<br />

Table 2.3 shows typical <strong>in</strong>put <strong>and</strong> output noise currents for the models. All the models<br />

predict relatively the same magnitude of output noise. The models with correlation<br />

show an <strong>in</strong>put noise similar to one another, but two orders of magnitude smaller than<br />

what Pospieszalski or this work predicts. This means that the cross-correlation terms<br />

55


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

(see, for example, equation 2.6.14) generate more noise than the gate noise. That<br />

the model <strong>in</strong>troduced here generates noise similar to that of Pospieszalski helps to<br />

re<strong>in</strong>force its validity.<br />

�<br />

2<br />

〈i<strong>in</strong>put〉<br />

�<br />

2<br />

〈ioutput〉 10<br />

Model: van der Ziel Pucel Pospieszalski This Work<br />

�<br />

−24 A2<br />

10 5.0 7.2 3100 1370<br />

Hz�<br />

−24 A2<br />

490 480 300 490<br />

Hz<br />

Table 2.3: Comparison of the various noise models’ <strong>in</strong>put <strong>and</strong> output noise currents.<br />

The model can help <strong>in</strong> predict<strong>in</strong>g NF, but it has its limitations. The most important<br />

is the model<strong>in</strong>g fails when the ga<strong>in</strong> drops because it does not take <strong>in</strong>to account Cgd<br />

or self-heat<strong>in</strong>g effects. This means it fails at frequencies close to fτ (f >fτ/2) <strong>and</strong><br />

at high DC bias<strong>in</strong>gs (Ids > 40 mA). A derivation with Cgd was undertaken, but no<br />

closed form solutions could be found for the noise parameters <strong>and</strong> the equations were<br />

too complicated to yield <strong>in</strong>sight. It is possible that the ga<strong>in</strong> could be corrected us<strong>in</strong>g<br />

the Miller effect [31].<br />

Also, the small-signal parameters must be accurately known. The parasitics re-<br />

sistances need to be correctly modeled <strong>and</strong> the measured S-parameters must be as<br />

accurate as possible. For example, the model<strong>in</strong>g was found to be poor when superior<br />

network analyzer was substituted with a less accurate network analyzer.<br />

56


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

2.7 Summary<br />

This chapter lead up to the presentation of a noise figure model that does not need<br />

prior noise parameter measurements for accurate prediction. It will be used <strong>in</strong> other<br />

parts of this work. All necessary background lead<strong>in</strong>g up to it was <strong>in</strong>troduced, its<br />

derivation presented <strong>in</strong> full, the model’s strengths <strong>and</strong> weakness expla<strong>in</strong>ed, <strong>and</strong> how it<br />

compares to the other popular noise models discussed.<br />

References<br />

[1] M. W. Pospieszalski, “Model<strong>in</strong>g of <strong>Noise</strong> Parameters of MESFET’s <strong>and</strong> MOD-<br />

FET’s <strong>and</strong> Their Frequency <strong>and</strong> Temperature Dependence,” IEEE Trans. <strong>Microwave</strong><br />

Theory Tech., vol. 37, no. 9, pp. 1340–1350, Sept. 1989.<br />

[2] R. A. Pucel, H. A. Haus, <strong>and</strong> H. Statz, “Signal <strong>and</strong> <strong>Noise</strong> Properties of Gallium Arsenide<br />

<strong>Microwave</strong> Field-Effect Transistors,” <strong>in</strong> Advances <strong>in</strong> <strong>Electronics</strong> <strong>and</strong> Electron<br />

Physics. New York: Academic Press, 1975, vol. 38, pp. 195–265.<br />

[3] A. van der Ziel, <strong>Noise</strong> <strong>in</strong> Solid State Devices <strong>and</strong> Circuits. New York: Wiley-<br />

Interscience, 1986.<br />

[4] J. B. Johnson, “Thermal Agitation of Electricity <strong>in</strong> Conductors,” Nature, vol. 119,<br />

pp. 50–51, 1927.<br />

[5] H. Nyquist, “Thermal Agitation of Electric Charge <strong>in</strong> Conductors,” Phys. Rev.,<br />

vol. 32, pp. 110–113, 1928.<br />

[6] C. Kittel <strong>and</strong> H. Kroemer, Thermal Physics, 2nd ed. New York: W. H. Freeman<br />

<strong>and</strong> Company, 2000.<br />

[7] S. A. Maas, <strong>Noise</strong> <strong>in</strong> L<strong>in</strong>ear <strong>and</strong> Nonl<strong>in</strong>ear Circuits. Boston: Artech House, 2005.<br />

[8] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, 2nd ed.<br />

New York: Cambridge University Pess, 2004.<br />

[9] M. J. Buck<strong>in</strong>gham, <strong>Noise</strong> <strong>in</strong> Electronic Devices <strong>and</strong> Systems. New York: John<br />

Wiley & Sons, 1983.<br />

[10] B. G. Streetman <strong>and</strong> S. Banerjee, Solid State Electronic Devices, 5th ed., ser. Solid<br />

State Physical <strong>Electronics</strong>. Upper Saddle River, NJ: Prentice Hall, 2000.<br />

57


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

[11] W. Liu, Fundamentals of III-V Devices: HBTs, MESFETs, <strong>and</strong> HFETs/<strong>HEMTs</strong>.<br />

New York: Wiley-Interscience, 1999.<br />

[12] B. Hughes <strong>and</strong> P. J. Tasker, “Bias Dependence of the MODFET Intr<strong>in</strong>sic Model Elements<br />

Values at <strong>Microwave</strong> Frequenices,” IEEE Trans. Electron Devices, vol. 36,<br />

no. 10, pp. 2267–2273, Oct. 1989.<br />

[13] G. Dambr<strong>in</strong>e, A. Cappy, F. Heliodore, <strong>and</strong> E. Playez, “A New Method for Determ<strong>in</strong><strong>in</strong>g<br />

the FET Small-Signal Equivalent Circuit,” IEEE Trans. <strong>Microwave</strong> Theory<br />

Tech., vol. 36, pp. 1151–1159, Jul. 1988.<br />

[14] P. H. Ladbrooke, MMIC Design: GaAs FETs <strong>and</strong> <strong>HEMTs</strong>. Boston: Artech House,<br />

Inc., 1989.<br />

[15] M. Berroth <strong>and</strong> R. Bosch, “Broad-B<strong>and</strong> Determ<strong>in</strong>ation of the FET Small-Signal<br />

Equivalent Circuit,” IEEE Trans. <strong>Microwave</strong> Theory Tech., vol. 38, pp. 891–895,<br />

Jul. 1990.<br />

[16] R. Anholt, Electrical <strong>and</strong> Thermal Characterization of MESFETs, <strong>HEMTs</strong>, <strong>and</strong><br />

HBTs. Norwood, MA: Artech House Publishers, 1994.<br />

[17] H. Xu, “MMICs us<strong>in</strong>g <strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> Th<strong>in</strong>-Film BST Capacitors,” Ph.D. dissertation,<br />

University of California, Santa Barbara, 2005.<br />

[18] H. Rothe <strong>and</strong> W. Dahlke, “Theory of Noisy Fourpoles,” Proceed<strong>in</strong>gs of the IRE,<br />

vol. 44, pp. 811–818, Jun. 1956.<br />

[19] G. Gonzalez, <strong>Microwave</strong> Transistor Amplifiers: Analysis <strong>and</strong> Design, 2nd ed. Upper<br />

Saddle River, NJ: Prentice-Hall, Inc., 1997.<br />

[20] M. J. Rodwell, “ECE 594F Class Notes, <strong>Noise</strong> <strong>in</strong> <strong>Electronics</strong> <strong>and</strong> Optoelectronics,”<br />

2000.<br />

[21] A. van der Ziel, “Gate <strong>Noise</strong> <strong>in</strong> Field Effect Transistors at Moderately High Frequencies,”<br />

Proc. IEEE, vol. 51, pp. 461–467, Mar. 1963.<br />

[22] ——, “Thermal <strong>Noise</strong> <strong>in</strong> Field-Effect Transistors,” Proc. IRE, vol. 50, pp. 1808–<br />

1812, Aug. 1962.<br />

[23] S. Lee, “Intr<strong>in</strong>sic <strong>Noise</strong> Characteriestics of Gallium Nitride High Electron Mobility<br />

Transistors,” Ph.D. dissertation, Purdue University, Aug. 2004.<br />

[24] A. Bal<strong>and</strong><strong>in</strong>, Ed., <strong>Noise</strong> <strong>and</strong> Fluctuations Control <strong>in</strong> Electronic Devices. Stevenson<br />

Ranch, CA: American Scientific Publishers, 2002.<br />

[25] S. Lee, K. J. Webb, V. Tilak, <strong>and</strong> L. Eastman, “Intr<strong>in</strong>sic <strong>Noise</strong> Equivalent-<br />

Circuit Paramters for <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” IEEE Trans. <strong>Microwave</strong> Theory<br />

Tech., vol. 51, pp. 1567–1577, May 2003.<br />

58


CHAPTER 2. NOISE FIGURE MODELING OF ALGAN/GAN HEMTS<br />

[26] H. Fukui, “Design of <strong>Microwave</strong> GaAs MESFET’s for Broad-B<strong>and</strong> Low-<strong>Noise</strong><br />

Amplifiers,” IEEE Trans. <strong>Microwave</strong> Theory Tech., vol. 27, no. 7, pp. 643–650,<br />

Jul. 1979.<br />

[27] M. W. Pospieszalski <strong>and</strong> W. Wiatr, “Comments on ”Design of <strong>Microwave</strong> GaAs<br />

MESFET’s for Broad-B<strong>and</strong>, Low-<strong>Noise</strong> Amplifiers,” IEEE Trans. <strong>Microwave</strong> Theory<br />

Tech., vol. MTT-34, no. 1, p. 194, Jan. 1986.<br />

[28] S. Nutt<strong>in</strong>ck, E. Gebara, J. Laskar, <strong>and</strong> M. Harris, “High-Frequency <strong>Noise</strong> <strong>in</strong> Al-<br />

<strong>GaN</strong>/<strong>GaN</strong> HFETs,” IEEE <strong>Microwave</strong> Components Lett., vol. 13, no. 4, pp. 149–<br />

151, Apr. 2003.<br />

[29] H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, S. Keller, U. Mishra, <strong>and</strong> R. York, “A C-B<strong>and</strong> High-<br />

Dynamic Range <strong>GaN</strong> HEMT Low-<strong>Noise</strong> Amplifier,” IEEE <strong>Microwave</strong> Components<br />

Lett., vol. 14, no. 6, pp. 262–264, Jun. 2004.<br />

[30] C. Sanabria, H. Xu, T. Palacios, A. Chakraborty, S. Heikman, U. Mishra, <strong>and</strong><br />

R. York, “Influence of Epitaxial Structure <strong>in</strong> the <strong>Noise</strong> Figure of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong><br />

<strong>HEMTs</strong>,” IEEE Trans. <strong>Microwave</strong> Theory Tech., vol. 53, pp. 762–769, Feb. 2005.<br />

[31] P. R. Gray, P. J. Hurst, S. H. Lewis, <strong>and</strong> R. G. Meyer, Analysis <strong>and</strong> Design of<br />

Analog Integrated Circuits, 4th ed. New York: Wiley, 2001.<br />

[32] F. Danneville, H. Happy, G. Dambr<strong>in</strong>e, J.-M. Belqu<strong>in</strong>, <strong>and</strong> A. Cappy, “Microscopice<br />

<strong>Noise</strong> Model<strong>in</strong>g <strong>and</strong> Macroscopic <strong>Noise</strong> Models: How Good a Connection?”<br />

IEEE Trans. Electron Devices, vol. 41, no. 5, pp. 779–786, May 1994.<br />

[33] M. Shur, Physics of Semiconductor Devices. Prentice Hall, 1990.<br />

[34] J. M. Golio, Ed., <strong>Microwave</strong> MESFETs <strong>and</strong> <strong>HEMTs</strong>. Norwood, MA: Artech<br />

House, Inc., 1991.<br />

59


3<br />

<strong>Noise</strong> Figure Measurements <strong>and</strong> Studies<br />

3.1 Introduction<br />

WITH the frame work for noise figure laid <strong>in</strong> the previous section, we now<br />

concentrate on noise figure measurements of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>. The<br />

technology used to fabricated the devices is presented first. The setup used for the<br />

measurements will be discussed, <strong>and</strong> the methodology used for comparisons expla<strong>in</strong>ed.<br />

The bulk of this chapter is devoted to several noise figure studies of <strong>Al<strong>GaN</strong></strong> <strong>HEMTs</strong>.<br />

These studies look at how different epitaxial material <strong>and</strong> device structures affect the<br />

noise figure performance. The effect of a field plate (FP) <strong>and</strong> gate leakage on noise<br />

figure are profound, <strong>and</strong> will be covered <strong>in</strong> § 3.5.4 <strong>and</strong> § 3.5.5. The model<strong>in</strong>g devel-<br />

oped <strong>in</strong> the previous chapter is used to underst<strong>and</strong> how a FP <strong>and</strong> gate leakage affect<br />

the noise performance. Another <strong>in</strong>terest<strong>in</strong>g type of <strong>GaN</strong> HEMT, known as a thick cap<br />

HEMT, was measured <strong>and</strong> will be briefly covered. It is of great <strong>in</strong>terest to know about<br />

the state-of-the-art of <strong>GaN</strong> HEMT noise performance <strong>and</strong> how it compares to <strong>HEMTs</strong><br />

<strong>in</strong> GaAs <strong>and</strong> Si material systems. This is done <strong>in</strong> § 3.6<br />

60


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

3.2 Device Details<br />

Devices from several samples with different epitaxial structures will be discussed,<br />

but they are similar from growth <strong>and</strong> process<strong>in</strong>g po<strong>in</strong>ts of view. The devices were<br />

all grown by metal-organic chemical vapor deposition (MOCVD) on both c-plane<br />

sapphire <strong>and</strong> c-plane 4H-SiC substrates. The epitaxial structures that will most com-<br />

monly appear <strong>in</strong> this work are <strong>in</strong> figure 3.1, which will be referenced throughout the<br />

chapter. The key differences <strong>in</strong> the samples are choice of substrate (sapphire or SiC)<br />

<strong>and</strong> <strong>in</strong>clusion of an AlN <strong>in</strong>terlayer between the <strong>Al<strong>GaN</strong></strong> barrier <strong>and</strong> the <strong>GaN</strong> channel.<br />

29 nm 27% <strong>Al<strong>GaN</strong></strong>: Si<br />

1700 nm UID <strong>GaN</strong><br />

65 nm AlN<br />

750 nm <strong>GaN</strong>: Fe<br />

Sapphire Substrate<br />

(a)<br />

29 nm 35% <strong>Al<strong>GaN</strong></strong>: Si<br />

0.6 nm AlN<br />

1700 nm UID <strong>GaN</strong><br />

65 nm AlN<br />

750 nm <strong>GaN</strong>: Fe<br />

Sapphire Substrate<br />

(b)<br />

29 nm 27% <strong>Al<strong>GaN</strong></strong>: Si<br />

1300 nm UID <strong>GaN</strong><br />

300 nm <strong>GaN</strong>: Fe<br />

160 nm AlN<br />

SiC Substrate<br />

Figure 3.1: Typical epitaxial structure for the devices <strong>in</strong> this work: (a) st<strong>and</strong>ard HEMT<br />

on a sapphire substrate (b) HEMT with AlN <strong>in</strong>terlayer (c) st<strong>and</strong>ard HEMT on siliconcarbide<br />

substrate<br />

After choice of substrate, growth consists of a nucleation layer. This layer may <strong>in</strong>-<br />

clude iron (Fe, an accepter) <strong>in</strong> it, as <strong>in</strong> figure 3.1 (a), to reduce the buffer conductive<br />

(caused by un<strong>in</strong>tentional dop<strong>in</strong>g). The decrease <strong>in</strong> conductivity improves the break-<br />

down of the device for power applications. An un<strong>in</strong>tentionally doped <strong>GaN</strong> layer grown<br />

61<br />

(c)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

next will be the channel, <strong>and</strong> a 29 nm <strong>Al<strong>GaN</strong></strong> Si-doped layer f<strong>in</strong>ishes the structure. The<br />

Al composition varies with the sample, but is 27 % if not stated explicity.<br />

Device process<strong>in</strong>g began with Ti/Al/Ni/Au electron beam evaporated source <strong>and</strong><br />

dra<strong>in</strong> contacts. These were annealed at 870 ◦ C for 30 s <strong>in</strong> a rapid thermal annealer<br />

(RTA). Device isolation was achieved by reactive ion etch<strong>in</strong>g (RIE) <strong>in</strong> Cl2. Stepper<br />

photolithography Ni/Au/Ni gates were electron beam evaporated with a gate length of<br />

0.7 µm. SiN passivation was achieved with plasma-enhanced chemical vapor deposi-<br />

tion (PECVD). If a field plate is used, it would now be added as a repeated <strong>and</strong> slightly<br />

shifted gate metal layer. All devices <strong>in</strong> this chapter have a gate width of 1x150 µm,<br />

a gate-source spac<strong>in</strong>g of 0.7 µm, <strong>and</strong> a gate-dra<strong>in</strong> spac<strong>in</strong>g of 2 µm. The pads are a<br />

coplanar waveguide (CPW) layout.<br />

Typical measurements of fτ <strong>and</strong> fmax are 23 <strong>and</strong> 47 GHz respectively. Contact<br />

resistance is from 0.3 to 0.6 Ω–mm. Charge <strong>and</strong> mobility vary with the Al composition<br />

of the barrier. For compositions rang<strong>in</strong>g from 15 % to 35 %, the mobility was found<br />

to be 1100-1565 cm 2 /Vs <strong>and</strong> the charge 0.4-1.3x10 13 cm −2 by Hall measurements.<br />

More details about the device details <strong>and</strong> process<strong>in</strong>g can be found <strong>in</strong> [1–4].<br />

3.3 <strong>Noise</strong> Figure Measurement Setup <strong>and</strong> Method<br />

<strong>Noise</strong> figure measurements were performed with a PC–controlled source–pull noise<br />

figure setup. A schematic is <strong>in</strong> figure 3.2. At the heart of the system is the noise figure<br />

62


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

Bias-T<br />

RF<br />

Switch<br />

<strong>Noise</strong><br />

Source<br />

Source<br />

Tuner<br />

RF Probe<br />

Station<br />

Tuner<br />

Controller<br />

Bias<br />

Controller<br />

Network<br />

Analyzer<br />

Computer<br />

Load<br />

Tuner<br />

GPIB<br />

Cable<br />

Bias-T<br />

RF<br />

Switch<br />

<strong>Noise</strong><br />

Figure<br />

Meter<br />

BNC<br />

Cable<br />

Figure 3.2: Schematic of the source-pull noise figure setup.<br />

meter, noise source, <strong>and</strong> source tuner. The HP 8970S noise-figure meter system con-<br />

sists of the HP 8970B noise figure meter, HP 8970C test set, <strong>and</strong> a signal generator<br />

that acts as a local oscillator for the 8970C. Typical error is ±0.15 dB. To f<strong>in</strong>d the de-<br />

vice m<strong>in</strong>imum noise figure, a variable source impedance is generated by a Maury Mi-<br />

crowave MT982A02 mechanical motorized tuner. The load tuner was set to 50 Ω. The<br />

device ga<strong>in</strong> dur<strong>in</strong>g measurement could have been improved by mov<strong>in</strong>g the load tuner<br />

63


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

to a small-signal match, but due to a comb<strong>in</strong>ation of hardware <strong>and</strong> software problems<br />

better measurements were obta<strong>in</strong>ed with it at the reference impedance. However, the<br />

software automatically calculates the ga<strong>in</strong> for a noise <strong>in</strong>put match <strong>and</strong> power output<br />

match.<br />

The device S-parameters at each bias are needed. A Maury <strong>Microwave</strong> MT998C<br />

RF switch box <strong>and</strong> an HP 8722D vector network analyzer (VNA) made it possible to<br />

seamlessly switch between noise <strong>and</strong> S-parameter measurements of devices. All mea-<br />

surements were performed on-wafer with Cascade-Microtech ACP40 ground-signal-<br />

ground CPW probes. The bias was set automatically by an HP 6625A DC–power<br />

supply system. All components were controlled over a general-purpose <strong>in</strong>terface bus<br />

(GPIB) by a Maury <strong>Microwave</strong> proprietary software program which calculates the<br />

noise parameters.<br />

The accuracy of the system was checked <strong>in</strong> two ways. The first was by measur-<br />

<strong>in</strong>g devices that were characterized elsewhere. The second was fabrication of CPW<br />

on-wafer 10 dB attenuators alongside devices <strong>and</strong> circuits. The schematic <strong>and</strong> a pho-<br />

tograph of the attenuator are <strong>in</strong> figure 3.3. Figure 3.3 (c) shows the measured loss <strong>and</strong><br />

m<strong>in</strong>imum noise figure (NFm<strong>in</strong>). A passive device’s loss <strong>and</strong> NFm<strong>in</strong> will be the same <strong>in</strong><br />

decimals. As can be seen, the agreement is good.<br />

To make NF comparisons, several factors need to be taken <strong>in</strong>to consideration. First,<br />

devices across a <strong>GaN</strong> wafer are not uniform due to the quality of research-grade mate-<br />

64


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

R1<br />

Ground<br />

2R 2<br />

In<br />

R1<br />

R1<br />

Out<br />

2R 2<br />

R1<br />

In Out<br />

(a)<br />

R2<br />

Ground<br />

dB<br />

11<br />

10.8<br />

10.6<br />

10.4<br />

10.2<br />

10<br />

9.8<br />

9.6<br />

9.4<br />

9.2<br />

NF m<strong>in</strong><br />

Loss<br />

9<br />

4 5 6 7 8 9 10 11 12 13 14 15<br />

Frequency (GHz)<br />

(b) (c)<br />

Figure 3.3: Coplanar waveguide attenuator (a) schematic, (b) photograph, <strong>and</strong> (c)<br />

measured loss <strong>and</strong> m<strong>in</strong>imum noise figure aga<strong>in</strong>st frequency.<br />

F (unitless)<br />

1.9<br />

1.8<br />

1.7<br />

1.6<br />

1.5<br />

1.4<br />

1.3<br />

F<br />

1/f τ<br />

1/f max<br />

0.10<br />

0.09<br />

0.08<br />

0.07<br />

0.06<br />

0.05<br />

0.04<br />

0.03<br />

0 20 40 60 80 100 120<br />

0.02<br />

140<br />

Current (mA)<br />

(1/GHz)<br />

F (unitless)<br />

2.2<br />

2.0<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

0 20 40 60 80 100 120 140<br />

Current (mA)<br />

Figure 3.4: <strong>Noise</strong> factor, fτ <strong>and</strong> fmax for devices from different samples versus<br />

current.<br />

65<br />

F<br />

1/f τ<br />

1/f max<br />

0.20<br />

0.18<br />

0.16<br />

0.14<br />

0.12<br />

0.10<br />

0.08<br />

0.06<br />

0.04<br />

0.02<br />

(1/GHz)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

rial <strong>and</strong> growth reactors. These variations across a sample cause changes <strong>in</strong> the device<br />

small-signal parameters <strong>and</strong> therefore <strong>in</strong> the fτ <strong>and</strong> fmax . <strong>Noise</strong> has a relationship to<br />

fτ <strong>and</strong> fmax such that improvements <strong>in</strong> one of the quantities usually means an improve-<br />

ment <strong>in</strong> the others. Figure 3.4 helps this argument. Here the <strong>in</strong>verse of fτ, <strong>in</strong>verse of<br />

fmax, <strong>and</strong> noise factor, F (NF = 10 log 10(F)), are plotted for devices from two samples<br />

as the dra<strong>in</strong>-source current is changed. As the current <strong>in</strong>creases, all three parameters<br />

<strong>in</strong>crease. There appears to be a close relationship between F <strong>and</strong> 1/fmax. This relation-<br />

ship can be seen from the model<strong>in</strong>g <strong>in</strong> the previous chapter. fτ <strong>and</strong> fmax are usually<br />

def<strong>in</strong>ed as<br />

gm<br />

fτ =<br />

Cgs + Cgd<br />

�<br />

Rds<br />

fmax = fτ<br />

4(Ri + Rg)<br />

(3.3.1)<br />

(3.3.2)<br />

Exam<strong>in</strong><strong>in</strong>g the last term <strong>in</strong> equation 2.6.16, we see a factor 1/ω 2 tau that is very similar<br />

to equation 3.3.1. An exact <strong>in</strong>stance of fmax is not seen <strong>in</strong> equation 2.6.16 (there is no<br />

Rds), but there is some similarity. The use of source degeneration, as demonstrated <strong>in</strong><br />

§ 2.6.1, modifies Cgs <strong>and</strong> gm. This would affect fτ, fmax, <strong>and</strong> thus NFm<strong>in</strong>. That NFm<strong>in</strong><br />

would depend on fmax <strong>in</strong>stead of fτ also makes sense from a power argument: noise<br />

figure <strong>and</strong> fmax are based on power quantities while fτ is a current ga<strong>in</strong> quantity. The<br />

conclusion of this aside is that devices with similar fτ <strong>and</strong> fmax should have similar<br />

noise.<br />

Device geometry is also important when mak<strong>in</strong>g comparisons. As seen <strong>in</strong> § 2.6.4,<br />

66


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

a wider device has a larger NFm<strong>in</strong>. Also, a device of smaller gate length usually has<br />

better fτ <strong>and</strong> fmax. For a fair comparison then, only devices of the same geometry<br />

should be compared.<br />

F<strong>in</strong>ally, there is the matter of device bias. As the DC bias to the transistor is changed<br />

the noise parameters also change. The optimum bias for NFm<strong>in</strong> may change with<br />

devices from different samples. In all measurements <strong>in</strong> this work, the bias is swept<br />

twice, first Vds then Ids, to f<strong>in</strong>d the best NFm<strong>in</strong> performance possible for the device.<br />

It is better to sweep Ids than Vgs as the threshold voltage changes across a sample.<br />

The transistor’s noise parameters were then measured versus frequency at this bias. It<br />

should be noted that the best bias for noise may not be the same as that of fmax, but it<br />

is a good start<strong>in</strong>g value.<br />

To show the importance of hav<strong>in</strong>g similar devices to NF, Rs, Ri, <strong>and</strong> Cgs variations<br />

were simulated over the range of extracted values <strong>and</strong> entered <strong>in</strong>to the model of chap-<br />

ter 2. The results of this are <strong>in</strong> figure 3.5. Each parameter spans a change of NFm<strong>in</strong> of<br />

0.2 to 0.3 dB. All measurements <strong>in</strong> this work are at a bias <strong>and</strong> source-impedance for the<br />

best NFm<strong>in</strong> atta<strong>in</strong>able from the device. For measurements versus bias, a frequency was<br />

used that gave the most stable <strong>and</strong> repeatable measurement (usually 5 GHz, sometimes<br />

10 GHz). As the bias was changed, S-parameters were re-measured <strong>and</strong> the optimum<br />

source-impedance re-evaluated. When compar<strong>in</strong>g devices from different samples, fτ<br />

<strong>and</strong> fmax measurements were first performed to f<strong>in</strong>d devices with similar performance.<br />

67


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

2.2<br />

2.1<br />

2.0<br />

1.9<br />

1.8<br />

1.7<br />

R s<br />

Capacitance (pF)<br />

0.20 0.21 0.22 0.23 0.24 0.25<br />

3 4 5 6 7 8 9 10 11 12<br />

R i<br />

Resistance (Ω)<br />

Figure 3.5: Variation <strong>in</strong> expected m<strong>in</strong>imum noise figure with changes <strong>in</strong> three smallsignal<br />

parameters.<br />

3.4 Bias Dependence<br />

Transistors <strong>in</strong> Low <strong>Noise</strong> Amplifiers (LNA) are biased at low currents <strong>and</strong> voltages<br />

for maximum NF performance <strong>and</strong> to reduce power consumption. A <strong>GaN</strong> HEMT<br />

biased at a “low” bias is likely beyond the breakdown of other material technologies.<br />

Therefore, it is of <strong>in</strong>terest to know how the noise parameters change with device bias,<br />

which leads to some <strong>in</strong>terest<strong>in</strong>g results <strong>in</strong> this work. Also, know<strong>in</strong>g how the noise<br />

changes with bias is necessary for noise sources extracted from NF measurements<br />

that are <strong>in</strong>cluded <strong>in</strong> device circuit simulations for oscillator phase noise.<br />

Figure 3.6 displays all four noise parameters, the associated ga<strong>in</strong>, <strong>and</strong> maximum<br />

ga<strong>in</strong> at a frequency of 10 GHz for dra<strong>in</strong>-source voltages, Vds, 2 to 20 V. The dra<strong>in</strong>-<br />

68<br />

C gs


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

Figure 3.6: Change <strong>in</strong> noise parameters with dra<strong>in</strong>-source voltage at 10 GHz: (a)<br />

M<strong>in</strong>imum noise figure (b) magnitude <strong>and</strong> phase of optimum reflection coefficient (c)<br />

noise resistance (d) device associated <strong>and</strong> maximum ga<strong>in</strong>s.<br />

69


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

source current, Ids, is kept constant at the best bias for NFm<strong>in</strong>. The small-signal asso-<br />

ciated power ga<strong>in</strong> is that available when the <strong>in</strong>put is matched for noise <strong>and</strong> the output<br />

for power. The maximum ga<strong>in</strong> is of a small-signal <strong>in</strong>put <strong>and</strong> output conjugate match.<br />

NFm<strong>in</strong> flattens out above Vds ∼3 V. The other noise parameters are nearly flat, except<br />

for the phase which decreases slightly. This is probably from a slight change <strong>in</strong> Cgd<br />

with Vgd. From these measurements, it can be concluded that once the device saturates<br />

the noise parameters can be considered constant.<br />

What is more <strong>in</strong>terest<strong>in</strong>g is how the noise parameters change with the dra<strong>in</strong>-source<br />

current. The measured noise parameters of this are the solid circles <strong>in</strong> figure 3.7<br />

for a sample with a sapphire substrate <strong>and</strong> 35 % Al <strong>in</strong> the barrier at a measurement<br />

frequency of 5 GHz. Now we f<strong>in</strong>d that NFm<strong>in</strong> <strong>and</strong> rn <strong>in</strong>crease with current. NFm<strong>in</strong> does<br />

not <strong>in</strong>crease because of a large <strong>in</strong>crease <strong>in</strong> device noise. The reason is that the ga<strong>in</strong><br />

drops with the <strong>in</strong>creas<strong>in</strong>g current. The ga<strong>in</strong> is lower because the channel is opened<br />

(loss of transconductance) <strong>and</strong> self-heat<strong>in</strong>g (from the high currents <strong>in</strong>volved). That<br />

the ga<strong>in</strong> affects the noise figure directly is seen <strong>in</strong> equation 2.6.9. The magnitude<br />

of the optimum source reflection coefficient, Γopt, decreases with <strong>in</strong>creas<strong>in</strong>g current<br />

while its phase decreases. This is a result of the match chang<strong>in</strong>g with loss of ga<strong>in</strong> <strong>and</strong><br />

<strong>in</strong>creas<strong>in</strong>g Cgs.<br />

To further <strong>in</strong>vestigate the usefulness of the Pospieszalski <strong>and</strong> CN models, they<br />

were applied to these noise measurements. Measurements of fτ <strong>and</strong> fmax (needed<br />

70


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

(a) (b)<br />

(c) (d)<br />

Figure 3.7: Typical plots of the noise parameters versus dra<strong>in</strong> source current as measured<br />

(circles) <strong>and</strong> us<strong>in</strong>g the Pospieszalski (dash-dot l<strong>in</strong>e) <strong>and</strong> CN (solid l<strong>in</strong>e) models:<br />

(a) M<strong>in</strong>imum noise figure (b) magnitude of optimum reflection coefficient (c) noise<br />

resistance (d) <strong>and</strong> phase of optimum reflection coefficient. Measurement frequency is<br />

5 GHz.<br />

71


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

for Pospieszalski’s formulation) <strong>and</strong> extracted small-signal parameters were collected<br />

at each bias the noise parameters were measured at <strong>in</strong> figure 3.7. Then, us<strong>in</strong>g meth-<br />

ods <strong>in</strong> § 2.5, a small-signal model with noise sources from either Pospieszalski’s or<br />

the CN formulation were created <strong>in</strong> ADS. ADS was then used to simulate the noise<br />

parameters. The results were added to figure 3.7. The solid l<strong>in</strong>es are simulations us-<br />

<strong>in</strong>g the CN model <strong>and</strong> the dash-dot l<strong>in</strong>es us<strong>in</strong>g the Pospieszalski model. Both models<br />

predict the noise parameters for the <strong>GaN</strong> HEMT versus bias well <strong>and</strong> can be used for<br />

bias-dependent noise parameter model<strong>in</strong>g of <strong>GaN</strong> <strong>HEMTs</strong>.<br />

The model<strong>in</strong>g <strong>in</strong> § 2.6 does not work to predict the noise parameters versus bias.<br />

It only works at low bias<strong>in</strong>gs (Ids < ∼40 mA <strong>in</strong> this work) because it does not take<br />

<strong>in</strong>to account reduction <strong>in</strong> ga<strong>in</strong> from self-heat<strong>in</strong>g. Because the Pospieszalski <strong>and</strong> CN<br />

models are fitt<strong>in</strong>g to the data at these high biases, they predict the noise parameters<br />

correctly. However, the model can still be used to expla<strong>in</strong> trends seen <strong>in</strong> the noise<br />

parameters versus bias. Equation 2.6.22 shows that as NFm<strong>in</strong> <strong>in</strong>creases <strong>and</strong> |Γopt|<br />

decreases with <strong>in</strong>creas<strong>in</strong>g current, then Rn should <strong>in</strong>crease as well.<br />

Figure 3.8 shows how the noise variables used for the simulations <strong>in</strong> figure 3.7<br />

change with bias. These variables are the quantitative values of noise sources entered<br />

<strong>in</strong> a circuit simulator, such as ADS. For example: shot noise changes with DC cur-<br />

rent, thermal noise changes with resistance, <strong>and</strong> a Pospieszalski thermal noise source<br />

changes with the noise temperature. We see <strong>in</strong> figure 3.8 (a) that the correlation coef-<br />

72


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

(a)<br />

(c)<br />

Figure 3.8: <strong>Noise</strong> variables for the Pospieszalski <strong>and</strong> a CN noise model versus dra<strong>in</strong><br />

source current: (a) magnitude <strong>and</strong> phase of the correlation coefficient <strong>and</strong> (b) gate <strong>and</strong><br />

dra<strong>in</strong> noise for a CN model. The device gm, multiplied by a factor of 10 to better fit<br />

the scale, is also <strong>in</strong> (b). (c) Dra<strong>in</strong> <strong>and</strong> gate noise temperatures of the Pospieszalski<br />

model for vary<strong>in</strong>g current.<br />

ficient magnitude <strong>and</strong> phase are relatively flat versus current. This supports the work<br />

of S. Lee on <strong>GaN</strong> <strong>HEMTs</strong> [5,6]. After detailed analysis us<strong>in</strong>g the CN model, Lee con-<br />

cluded that the correlation coefficient could be considered constant with a magnitude<br />

73<br />

(b)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

of 0.7 <strong>and</strong> phase of 90 ◦ after de-embedd<strong>in</strong>g extr<strong>in</strong>sic thermal <strong>and</strong> shot noise. Here,<br />

we see the same magnitude but a slightly higher phase. The <strong>in</strong>crease <strong>in</strong> phase is likely<br />

from the shot noise of the gate leakage. While Lee’s results were only versus fre-<br />

quency, we see here that it holds versus bias as well. To the author’s knowledge, there<br />

is no reference <strong>in</strong> the <strong>GaN</strong> literature that has the noise variables us<strong>in</strong>g the CN model<br />

versus bias. We might expect the <strong>in</strong>put <strong>and</strong> output noise currents <strong>in</strong> figure 3.8 (b) to<br />

behave as van der Ziel predicts <strong>in</strong> equations 2.5.1 <strong>and</strong> 2.5.2. Then the dra<strong>in</strong> noise,<br />

〈i 2 d 〉, should follow changes <strong>in</strong> gm versus bias. Ten times the value of gm (for the con-<br />

venience of plott<strong>in</strong>g) is also <strong>in</strong> figure 3.8 (b). 〈i 2 d〉 appears to follow the same trend.<br />

The gate noise, �<br />

i2 �<br />

g , also seems to follow what van der Ziel would predict (a C2 gs/gm<br />

dependence).<br />

Turn<strong>in</strong>g to the Pospieszalski model <strong>in</strong> figure 3.8 (c), the two noise temperatures are<br />

plotted aga<strong>in</strong>st Ids. The gate noise temperature is relatively flat. This agrees with the<br />

one <strong>GaN</strong> reference <strong>in</strong> the literature [7], <strong>and</strong> what is seen <strong>in</strong> GaAs <strong>HEMTs</strong>. Here, the<br />

dra<strong>in</strong> noise temperature drops. This is the opposite of what would be expected but is<br />

easily expla<strong>in</strong>ed. The sapphire sample the device was on did not have a good buffer,<br />

mak<strong>in</strong>g Rds low (less than 1 kΩ <strong>in</strong>stead of ∼1.5 kΩ). As the current <strong>in</strong>creased, Rds<br />

dropped off dramatically (900 Ω to 100 Ω). The term Td/Rds is prevalent through<br />

Pospieszalski’s equations. In fitt<strong>in</strong>g to the data, Td decreased along with Rds to keep<br />

the correct proportion. In other samples with a good buffer (only a slight drop <strong>in</strong> Rds),<br />

74


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

Td will <strong>in</strong>crease.<br />

3.5 <strong>GaN</strong> HEMT <strong>Noise</strong> Figure Studies<br />

3.5.1 Substrate<br />

The effect of substrate on the noise parameters was carried out on samples with<br />

SiC <strong>and</strong> sapphire substrates hav<strong>in</strong>g 25% Al <strong>in</strong> the barrier <strong>and</strong> structures similar to<br />

figure 3.1 (a) <strong>and</strong> (c). Versus frequency, devices from both samples had very similar<br />

Figure 3.9: M<strong>in</strong>imum noise figure, small signal associated <strong>and</strong> maximum ga<strong>in</strong> for<br />

devices on sapphire <strong>and</strong> SiC substrates.<br />

75


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

noise parameters. As the current was varied, a difference <strong>in</strong> performance of NFm<strong>in</strong><br />

emerged. Figure 3.9 shows that the sapphire sample had worse noise figure perfor-<br />

mance at higher bias<strong>in</strong>gs. At 100 mA, the difference is more than 0.5 dB. Also plotted<br />

are the associated <strong>and</strong> max ga<strong>in</strong>s for both devices shown. The sapphire sample ga<strong>in</strong>s<br />

fall much quicker as the bias <strong>in</strong>creases. As stated earlier, if the ga<strong>in</strong> drops then the<br />

noise figure will too. Devices on sapphire have degraded power performance at high<br />

bias<strong>in</strong>gs because of self-heat<strong>in</strong>g. Here, we see it affects noise figure performance as<br />

well. A previous study confirms the noise degradation is caused by self-heat<strong>in</strong>g with<br />

temperature dependent measurements [8].<br />

3.5.2 Al Composition <strong>in</strong> the Barrier<br />

To <strong>in</strong>vestigate if chang<strong>in</strong>g the Al composition <strong>in</strong> the <strong>Al<strong>GaN</strong></strong> barrier had an effect<br />

on noise, four samples were grown with 15%, 25%, 27%, <strong>and</strong> 35% Al on a sapphire<br />

substrate. The rest of the structure for all four samples was the same as <strong>in</strong> figure 3.1 (a).<br />

Lu previously published this study for <strong>GaN</strong> <strong>HEMTs</strong> [9]. He found that devices of high<br />

Al composition (35 %) had better noise performance than low Al composition (15 %).<br />

However, the reported fτ <strong>and</strong> fmax for the samples were not similar. fτ <strong>in</strong>creased with<br />

higher Al composition from 25 GHz to 50 GHz <strong>and</strong> fmax <strong>in</strong>creased from 55 GHz to<br />

101 GHz. The four samples <strong>in</strong> this study had reasonably similar fτ <strong>and</strong> fmax. <strong>Noise</strong><br />

parameters of the four samples for chang<strong>in</strong>g frequency are plotted <strong>in</strong> figure 3.10. Also,<br />

76


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong><br />

r n<br />

2.4<br />

2.2<br />

2.0<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

1.0<br />

2.0<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

1.0<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

35%<br />

27%<br />

25%<br />

15%<br />

3 4 5 6 7 8 9 10 11 12 13<br />

Frequency (GHz)<br />

(a)<br />

0.0<br />

3 4 5 6 7 8 9 10 11 12 13<br />

Frequency (GHz)<br />

(c)<br />

|Γ|<br />

Ga<strong>in</strong> (dB)<br />

1.0<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0.0<br />

16<br />

14<br />

12<br />

10<br />

8<br />

35%<br />

27%<br />

25%<br />

15%<br />

4 6 8 10 12<br />

Frequency (GHz)<br />

(b)<br />

6<br />

3 4 5 6 7 8 9 10 11 12 13<br />

Frequency (GHz)<br />

(d)<br />

Figure 3.10: <strong>Noise</strong> parameters versus frequency for devices of different alum<strong>in</strong>um<br />

composition <strong>in</strong> the barrier: (a) M<strong>in</strong>imum noise figure (b) magnitude <strong>and</strong> phase of<br />

optimum reflection coefficient (c) noise resistance (d) device associated ga<strong>in</strong>.<br />

77<br />

180<br />

160<br />

140<br />

120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

0<br />

Phase Γ (Degrees)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

3.8<br />

3.6<br />

3.4<br />

3.2<br />

3.0<br />

2.8<br />

2.6<br />

2.4<br />

2.2<br />

2.0<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

35%<br />

27%<br />

25%<br />

15%<br />

1.0<br />

0 20 40 60 80 100 120 140<br />

Current (mA)<br />

Figure 3.11: M<strong>in</strong>imum noise figure of samples with different alum<strong>in</strong>um composition<br />

<strong>in</strong> the barrier at vary<strong>in</strong>g dra<strong>in</strong>-source current.<br />

NFm<strong>in</strong> for the samples versus Ids are <strong>in</strong> figure 3.11. The data are astonish<strong>in</strong>gly similar<br />

for noise measurements, <strong>and</strong> challenge the previous report. The application of the CN<br />

<strong>and</strong> Pospieszalski models to these devices, presented earlier <strong>in</strong> table 2.2, also agrees<br />

that the noise parameters should be very similar. It is <strong>in</strong>terest<strong>in</strong>g that the dra<strong>in</strong>-source<br />

current bias for best NFm<strong>in</strong> was similar for the different samples (15±5mA). For the<br />

15 % Al sample this was nearly half Ids,sat while only 8 % Ids,sat for the 35 % Al<br />

sample.<br />

3.5.3 AlN Interlayer<br />

The addition of an extremely th<strong>in</strong> (one or two monolayers) AlN layer between the<br />

<strong>GaN</strong> channel <strong>and</strong> the <strong>Al<strong>GaN</strong></strong> barrier has been found to <strong>in</strong>crease the conduction-b<strong>and</strong><br />

78


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

4.0<br />

3.5<br />

3.0<br />

2.5<br />

2.0<br />

1.5<br />

1.0<br />

0 20 40 60 80 100 120 140<br />

Current (mA)<br />

NF m<strong>in</strong> AlN<br />

NF m<strong>in</strong> no AlN<br />

f τ AlN<br />

f τ no AlN<br />

f max AlN<br />

f max no AlN<br />

Figure 3.12: M<strong>in</strong>imum noise figure, fτ, <strong>and</strong> fmax versus dra<strong>in</strong>-source current for a<br />

sample with (squares) <strong>and</strong> without (circles) an AlN <strong>in</strong>terlayer.<br />

offset, better conf<strong>in</strong><strong>in</strong>g the 2-DEG <strong>and</strong> improv<strong>in</strong>g mobility [10]. This should improve<br />

the device fτ <strong>and</strong> fmax. It is therefore worth determ<strong>in</strong><strong>in</strong>g if it improves the noise<br />

performance as well. Samples with structures as <strong>in</strong> figure 3.1 (a) <strong>and</strong> (b) with 35%<br />

Al <strong>in</strong> the barrier on sapphire substrates were measured for noise. Overall, devices on<br />

the sample with the AlN-<strong>in</strong>terlayer had fτ <strong>and</strong> fmax values marg<strong>in</strong>ally higher than the<br />

sample without. However, for the noise measurements devices with as similar of fτ<br />

<strong>and</strong> fmax as could be found were used.<br />

There were no differences <strong>in</strong> the noise parameters versus frequency, but as the dra<strong>in</strong><br />

current <strong>in</strong>creased a large difference <strong>in</strong> NFm<strong>in</strong> was apparent. This is <strong>in</strong> figure 3.12 as<br />

the solid symbols. At a current of 100 mA, there is a 0.4 dB difference <strong>and</strong> at the<br />

79<br />

45<br />

40<br />

35<br />

30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

0<br />

Frequency (GHz)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

(dB)<br />

16<br />

14<br />

12<br />

10<br />

8<br />

6<br />

4<br />

Ga<strong>in</strong> AlN layer<br />

Ga<strong>in</strong> no AlN<br />

Gmax AlN layer<br />

Gmax no AlN<br />

2<br />

0 20 40 60 80 100 120 140<br />

Current (mA)<br />

(a)<br />

R s (Ω)<br />

26<br />

24<br />

22<br />

20<br />

18<br />

16<br />

14<br />

12<br />

10<br />

8<br />

6<br />

with AlN <strong>in</strong>terlayer<br />

without AlN<br />

0 10 20 30 40 50 60 70 80 90<br />

Current (mA)<br />

(b)<br />

Figure 3.13: (a) Associated <strong>and</strong> maximum ga<strong>in</strong> <strong>and</strong> (b) source resistance for devices<br />

with <strong>and</strong> without an AlN <strong>in</strong>terlayer at different applied currents.<br />

max current a 0.8 dB difference. While the ga<strong>in</strong> (figure 3.13 (a)), f τ, <strong>and</strong> fmax (also<br />

<strong>in</strong> figure 3.12) do drop off slightly faster for the sample without the AlN, it is not as<br />

dramatic as the sapphire/SiC substrate comparison <strong>and</strong> not enough to expla<strong>in</strong> such a<br />

large NFm<strong>in</strong> difference.<br />

Rs appears to be the cause of the difference. Palacios has shown that Rs <strong>in</strong>creases<br />

drastically with dra<strong>in</strong>-source current [11]. Figure 3.13 (b) shows the measured source<br />

resistance (us<strong>in</strong>g the method <strong>in</strong> [11]) for devices from both samples. The devices with<br />

the AlN-<strong>in</strong>terlayer see an Rs <strong>in</strong>crease of about two times, while the devices without<br />

the <strong>in</strong>terlayer see an <strong>in</strong>crease of three times. These Rs measurements are only out to<br />

80 mA (equipment limitations), <strong>in</strong>stead of 135 mA as <strong>in</strong> figure 3.12. The difference<br />

<strong>in</strong> Rs would be even larger at higher biases. As seen with the model<strong>in</strong>g us<strong>in</strong>g source-<br />

degeneration <strong>in</strong> chapter 2, Rs can be thought of as directly add<strong>in</strong>g noise at the <strong>in</strong>put,<br />

80


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

which <strong>in</strong>creases the noise.<br />

As mentioned, devices on the sample with the AlN-<strong>in</strong>terlayer overall tend to have<br />

higher fτ <strong>and</strong> fmax. These devices will have a slightly better NFm<strong>in</strong>. However, the<br />

improvement is at best 0.1 dB <strong>in</strong> the X-b<strong>and</strong>.<br />

3.5.4 Gate Leakage<br />

While try<strong>in</strong>g to do a comparative study, devices from a sample (27% Al content<br />

barrier with SiC substrate) that had similar fτ <strong>and</strong> fmax did not have similar NFm<strong>in</strong>.<br />

In fact, NFm<strong>in</strong> differed by almost 1 dB despite the ga<strong>in</strong>s be<strong>in</strong>g identical as seen <strong>in</strong><br />

figure 3.14. It was realized that the gate leakage was very different for the three devices<br />

<strong>in</strong> figure 3.14. The three term<strong>in</strong>al gate leakage, Igs, at a bias of Ids = 10 mA, Vds =5V<br />

for the devices was found to be: 22 µA (or 140 µA/mm); 73 µA (or 486 µA/mm);<br />

141 µA (or 940 µA/mm). The devices are labeled as such <strong>in</strong> figure 3.14. We see that<br />

as the gate leakage <strong>in</strong>creases, so does the noise figure. To explore how gate leakage<br />

affects the noise parameters, the model<strong>in</strong>g <strong>in</strong> § 2.6 was used. Small-signal parameters<br />

that approximately matched all three devices <strong>in</strong> figure 3.14 were entered <strong>in</strong>to Matlab<br />

us<strong>in</strong>g the script <strong>in</strong> appendix B. The gate leakage was swept from 10 −8 to 10 −2 A <strong>and</strong><br />

the calculated noise parameters plotted (as l<strong>in</strong>es) <strong>in</strong> figure 3.15 with data (crosses) for<br />

the devices <strong>in</strong> figure 3.14. The circles are data from another sample that exhibits an<br />

expected amount of gate leakage. The agreement of simulation <strong>and</strong> data for NFm<strong>in</strong><br />

81


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

Figure 3.14: (a) M<strong>in</strong>imum noise figure, (b) device associated ga<strong>in</strong>, <strong>and</strong> maximum ga<strong>in</strong><br />

versus frequency for devices with different gate leakage currents. The gate leakage<br />

quoted is at the same bias as the noise measurements (Ids = 10 mA, Vds = 5 V).<br />

is excellent. The other parameters do not agree perfectly, but it should be reiterated<br />

that the small-signal parameters used were typical for all three devices (they were not<br />

identical). No change <strong>in</strong> rn was predicted or measured.<br />

Gate leakage has a large effect on the noise parameters. It must be monitored, along<br />

with fτ <strong>and</strong> fmax, when measur<strong>in</strong>g noise figure of <strong>GaN</strong> devices. While gate leakage<br />

<strong>and</strong> NFm<strong>in</strong> have been studied previously [6, 12–14], these studies are all lack<strong>in</strong>g <strong>in</strong><br />

at least one of the follow<strong>in</strong>g ways: use of fitt<strong>in</strong>g, <strong>in</strong>adequate data, only predict<strong>in</strong>g<br />

NFm<strong>in</strong>, or not applied to <strong>GaN</strong>. The comb<strong>in</strong>ation of analytical model<strong>in</strong>g (without fitt<strong>in</strong>g<br />

parameters), data presented, <strong>and</strong> agreement of model<strong>in</strong>g <strong>and</strong> data here is unique <strong>and</strong><br />

82


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

r n<br />

NF m<strong>in</strong> (dB)<br />

5<br />

4<br />

3<br />

2<br />

10 -8 10 -7 10 -6 10 -5 10 -4 10 -3 10 -2<br />

1<br />

I (A)<br />

gs<br />

2<br />

1.75<br />

1.5<br />

1.25<br />

1<br />

0.75<br />

0.5<br />

0.25<br />

|Γ opt |<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

10 -8 10 -7 10 -6 10 -5 10 -4 10 -3 10 -2<br />

0<br />

I (A)<br />

gs<br />

(a) (b)<br />

140<br />

10 -8 10 -7 10 -6 10 -5 10 -4 10 -3 10 -2<br />

0<br />

I (A)<br />

gs<br />

Phase Γ opt (degress)<br />

120<br />

100<br />

80<br />

60<br />

40<br />

10 -8 10 -7 10 -6 10 -5 10 -4 10 -3 10 -2<br />

20<br />

I (A)<br />

gs<br />

(c) (d)<br />

Figure 3.15: Simulated (l<strong>in</strong>e) <strong>and</strong> measured (crosses) noise parameters for devices<br />

with different gate leakage currents: (a) M<strong>in</strong>imum noise figure (b) magnitude of optimum<br />

reflection coefficient (c) noise resistance (d) phase of optimum reflection coefficient.<br />

Frequency is 10 GHz. The circles are data from another sample with the<br />

typically expected amount of gate leakage.<br />

clear.<br />

A f<strong>in</strong>al note: <strong>GaN</strong> <strong>HEMTs</strong> on MBE were measured by the author. Their power <strong>and</strong><br />

small-signal performance are comparable to MOCVD-based <strong>HEMTs</strong>, but the devices<br />

had higher gate leakage <strong>and</strong> thus ∼0.15 dB more noise.<br />

83


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

3.5.5 Field-Plated Devices<br />

The addition of a FP has led to <strong>GaN</strong> <strong>HEMTs</strong> hav<strong>in</strong>g impressive power h<strong>and</strong>l<strong>in</strong>g<br />

capability at microwave frequencies [15, 16]. There is a penalty, which is shown <strong>in</strong><br />

figure 3.16. 1 A FP <strong>in</strong>creases Cgd, <strong>and</strong> that reduces fτ <strong>and</strong> fmax. In fact, the longer<br />

the FP, the higher the breakdown <strong>and</strong> lower the fτ ,fmax , <strong>and</strong> ga<strong>in</strong> of the device. The<br />

author decided to <strong>in</strong>vestigate the noise performance. Based on the studies presented<br />

thus far, one would predict that a FP device would have worse NF performance.<br />

Frequency (GHz)<br />

60<br />

55<br />

50<br />

45<br />

40<br />

35<br />

30<br />

25<br />

20<br />

15<br />

f τ<br />

f max<br />

0.0 0.2 0.4 0.6 0.8 1.0 1.2<br />

Field-Plate Length (µm)<br />

Figure 3.16: fτ <strong>and</strong> fmax of devices with different field-plate lengths.<br />

This turned out not to be the case. Figure 3.17 (a) demonstrates that as the FP<br />

length <strong>in</strong>creases the NFm<strong>in</strong> improves despite decreas<strong>in</strong>g ga<strong>in</strong> (Figure 3.17 (d)). The<br />

1For all measurements <strong>in</strong> this section, the devices with different FP lengths are on the same die of a<br />

sample.<br />

84


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

r n<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

1.0<br />

0.8<br />

4 6 8 10<br />

Frequency (GHz)<br />

(a)<br />

1.2<br />

1.0<br />

0.8<br />

0.6<br />

0.4<br />

1.1 µm<br />

0.9 µm<br />

0.7 µm<br />

0.5 µm<br />

None<br />

4 6 8 10<br />

Frequency (GHz)<br />

(c)<br />

|Γ|<br />

Ga<strong>in</strong> (dB)<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

14<br />

12<br />

10<br />

8<br />

6<br />

4<br />

1.1 µm<br />

0.9 µm<br />

0.7 µm<br />

0.5 µm<br />

None<br />

0.0<br />

20<br />

4 6 8<br />

Frequency (GHz)<br />

(b)<br />

10<br />

4 6 8 10<br />

Frequency (GHz)<br />

(d)<br />

Figure 3.17: <strong>Noise</strong> parameters versus frequency for devices with field plates of different<br />

length: (a) M<strong>in</strong>imum noise figure (b) magnitude <strong>and</strong> phase of optimum reflection<br />

coefficient (c) noise resistance (d) device ga<strong>in</strong> when matched at the <strong>in</strong>put for best noise<br />

performance <strong>and</strong> the output for power performance.<br />

85<br />

180<br />

160<br />

140<br />

120<br />

100<br />

80<br />

60<br />

40<br />

Phase Γ (degree)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

improvement is almost 0.2 dB. There is little additional improvement for a FP longer<br />

than 0.9 µm. The match changes monotonically with the FP <strong>in</strong> Figure 3.17 (b). The<br />

improvement <strong>in</strong> NF with a FP was first reported by the author <strong>in</strong> [17] <strong>and</strong> was later<br />

confirmed <strong>in</strong> [18].<br />

Several explanations to underst<strong>and</strong> this result were explored. The first was to look<br />

at the gate leakage. It was thought that maybe the FP was reduc<strong>in</strong>g the electric field<br />

<strong>and</strong> possibly the gate-dra<strong>in</strong> contribution to gate leakage. The three <strong>and</strong> two-term<strong>in</strong>al<br />

gate leakage of 100 µm wide devices with different FP lengths was measured. Some<br />

of the three-term<strong>in</strong>al measurements are <strong>in</strong> figure 3.18. The bias was set to Vds 5V<br />

(similar to noise measurements) <strong>and</strong> Vds 20 V (a low bias for power) <strong>and</strong> Vgs a volt<br />

past the threshold voltage. It does not appear that the FP lowers gate leakage. In fact,<br />

it appears to <strong>in</strong>crease slightly with the FP.<br />

Another thought was that perhaps there was a difference <strong>in</strong> electric field caus<strong>in</strong>g a<br />

change elsewhere than gate leakage. The analysis of Γ <strong>in</strong> § 2.6.3 was exam<strong>in</strong>ed to see<br />

if it might change because of a FP. This proved to not be true. A change <strong>in</strong> Γ would<br />

mean a change <strong>in</strong> the DC I–V characteristics of a FP from a non-FP device. The only<br />

real change is that the knee voltage is better def<strong>in</strong>ed for a FP device [19]. That would<br />

only cause Γ to go to its f<strong>in</strong>al value of 2/3 quicker. More proof that it was not the<br />

electric field-profile was found by two other means. One was to look at NFm<strong>in</strong> versus<br />

dra<strong>in</strong> source voltage. No difference was apparent. Another was to run an ATLAS<br />

86


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

Gate Leakage (µA)<br />

700<br />

600<br />

500<br />

400<br />

300<br />

200<br />

100<br />

0<br />

, V ds 20 V<br />

, V ds 5 V<br />

0.0 0.2 0.4 0.6 0.8 1.0 1.2<br />

Field-Plate Length (µm)<br />

Figure 3.18: Typical change <strong>in</strong> gate leakage for devices of <strong>in</strong>creas<strong>in</strong>g field-plate<br />

length. Bias<strong>in</strong>gs are Vds 5V,Ids 10 mA <strong>and</strong> Vds 20 V, Vgs one volt past threshold.<br />

Electric Field (V/cm)<br />

6.0M<br />

4.0M<br />

2.0M<br />

no FP, V ds 5 V<br />

no FP, V ds 20 V<br />

FP, V ds 5 V<br />

FP, V ds 20 V<br />

0.0<br />

0.0 0.5 1.0 1.5 2.0 2.5<br />

Distance (µm)<br />

Figure 3.19: Electric field profile for a device with <strong>and</strong> without a FP at a bias of Vds 5<br />

<strong>and</strong> 20 V with the gate biased close to Vt. The gray bars are the “physical lengths” of<br />

the gate <strong>and</strong> field plate. Atlas simulation provided by Yuvaraj Dora.<br />

87


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

simulation <strong>and</strong> look at the difference <strong>in</strong> the electric field profile. This was done, <strong>and</strong><br />

the results are <strong>in</strong> figure 3.19. 2 The simulations at Vds 5 <strong>and</strong> 20 V show the total electric<br />

field versus distance. The gate is biased close to the threshold voltage, as is the case<br />

for best NFm<strong>in</strong> results. The bars sitt<strong>in</strong>g atop the x-axis represent the gate length (short)<br />

<strong>and</strong> FP (long) lengths. While the peak electric field at the dra<strong>in</strong> edge of the gate will<br />

be significantly reduced when a HEMT is biased to 80 V or more, here for 5 <strong>and</strong> 20 V,<br />

there is no difference. In fact, at Vds 5 V, there is practically no change <strong>in</strong> the field.<br />

The small-signal parameters were also exam<strong>in</strong>ed. Those that were found to change<br />

with a FP are <strong>in</strong> figure 3.20. Cgd <strong>in</strong>creases because of the extra capacitance the FP<br />

provides. That Rgd decreases h<strong>in</strong>ts that it is either easier to charge Cgd or that the<br />

leakage has <strong>in</strong>creased the conductance between gate <strong>and</strong> dra<strong>in</strong>. Rd decreases because<br />

the effective distance between gate <strong>and</strong> dra<strong>in</strong> is reduced by the length of the FP. Ri<br />

does not appear to change. Any change <strong>in</strong> it is probably due to error <strong>in</strong> the small-signal<br />

extraction hav<strong>in</strong>g difficulty differentiat<strong>in</strong>g Rg <strong>and</strong> Ri. As the gate now has another set<br />

of f<strong>in</strong>gers <strong>in</strong> parallel, its resistance, <strong>and</strong> thus Rg, decreases. The FP <strong>and</strong> gate f<strong>in</strong>gers<br />

can be connected at their ends, further reduc<strong>in</strong>g gate resistance <strong>and</strong> the noise [20].<br />

When try<strong>in</strong>g to expla<strong>in</strong> FP device NF performance, it was first thought that Rg was<br />

the cause. However, the model<strong>in</strong>g at the time argued that the improvement <strong>in</strong> Rg was<br />

not enough to cause the difference, particularly for the results <strong>in</strong> [18]. It was later<br />

realized the devices <strong>and</strong> the model<strong>in</strong>g were not the same. Devices of different widths,<br />

2 ATLAS simulation graciously provided by Yuvaraj Dora to the author’s specifications<br />

88


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

C gd (fF)<br />

70<br />

60<br />

50<br />

40<br />

30<br />

20<br />

10<br />

0<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

Field-Plate Length (µm)<br />

70<br />

60<br />

50<br />

40<br />

30<br />

20<br />

10<br />

0<br />

R gd (Ω)<br />

Resistance (Ω)<br />

19<br />

18<br />

17<br />

16<br />

15<br />

8<br />

7<br />

6<br />

5<br />

4<br />

3<br />

2<br />

1<br />

R d<br />

R i<br />

R g<br />

0.0 0.2 0.4 0.6 0.8 1.0<br />

Field-Plate Length (µm)<br />

Figure 3.20: Small-signal parameters that change with a field plate.<br />

<strong>and</strong> with/without the FP plus gate ends shorted, were be<strong>in</strong>g compared. The scal<strong>in</strong>g<br />

<strong>and</strong> model<strong>in</strong>g of chapter 2 were used yet aga<strong>in</strong>. How NFm<strong>in</strong> changes with gate width<br />

for devices with <strong>and</strong> without a FP were simulated <strong>in</strong> Matlab <strong>and</strong> compared to the<br />

author’s <strong>and</strong> Wu’s [18] measurements. Figure 3.21 makes this clear. The agreement<br />

is very good, <strong>and</strong> it can now be concluded that the FP lower<strong>in</strong>g the gate resistance<br />

is what improves NFm<strong>in</strong>. That NFm<strong>in</strong> decreases despite the lower ga<strong>in</strong> is because the<br />

FP is act<strong>in</strong>g as an external feedback capacitance between gate <strong>and</strong> dra<strong>in</strong> (<strong>in</strong> parallel<br />

with Cgd). Lossless feedback does not harm the noise figure, but it does lower the<br />

ga<strong>in</strong> [21]).<br />

The reduction <strong>in</strong> NF does not justify the use of a FP as it reduces the ga<strong>in</strong> consider-<br />

ably (several decibels). Ga<strong>in</strong> is <strong>in</strong> short supply at microwave frequencies <strong>and</strong> reduc<strong>in</strong>g<br />

it so much may not justify NFm<strong>in</strong> improvement of just a few tenths of a decimal. In<br />

89


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

3<br />

2.5<br />

2<br />

1.5<br />

1<br />

0.5<br />

No FP<br />

Long FP<br />

Difference<br />

0<br />

0 50 100 150 200 250 300<br />

Gate Width (µm)<br />

Figure 3.21: M<strong>in</strong>imum noise figure versus gate width for devices with <strong>and</strong> without a<br />

long field plate at a simulation frequency of 10 GHz. The difference between them, <strong>in</strong><br />

decimals, is plotted as the dotted l<strong>in</strong>e. The “x’s” are from this work (150 µm ) <strong>and</strong> Y.<br />

Wu (243 µm ) [18].<br />

NF m<strong>in</strong><br />

1.8<br />

1.7<br />

1.6<br />

1.5<br />

1.4<br />

1.3<br />

1.2<br />

1.1<br />

1.0<br />

0.9<br />

0.8<br />

0.7<br />

4 GHz<br />

7 GHz<br />

10 GHz<br />

0.0 0.2 0.4 0.6 0.8 1.0 1.2<br />

Field-Plate Length (µm)<br />

Figure 3.22: M<strong>in</strong>imum noise figure of the field-plated devices at different measurement<br />

frequencies of 4, 7, <strong>and</strong> 10 GHz. As the frequency <strong>in</strong>creases, the improvement<br />

from a field plate decreases.<br />

90


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

addition, the benefit is reduced the higher the operat<strong>in</strong>g frequency because of the sk<strong>in</strong><br />

effect on the gate resistance. This is demonstrated <strong>in</strong> figure 3.22. At 4 GHz, the im-<br />

provement can be as much as 0.3 dB, but at 10 GHz the improvement is 0.1 dB. This<br />

agrees with the results <strong>in</strong> [18].<br />

3.5.6 Thick-Epitaxial Cap Devices<br />

Shen has proposed a HEMT with a thick <strong>GaN</strong> cap on top of the <strong>Al<strong>GaN</strong></strong> [22]. The<br />

advantage of this structure is that a SiN passivation is not needed. As passivation<br />

leads to problems of reliability for st<strong>and</strong>ard <strong>HEMTs</strong>, this new HEMT holds great<br />

promise. It has the record for power performance without passivation. Its small-signal<br />

characteristics are similar to st<strong>and</strong>ard <strong>HEMTs</strong>. Dr. Shen allowed the devices to be<br />

measured for noise. The performance was similar versus frequency (<strong>in</strong> particular,<br />

the ga<strong>in</strong>s), although the dra<strong>in</strong>-source current needed to be higher for optimum noise<br />

performance. The reason for this is clear after view<strong>in</strong>g the noise parameters versus<br />

bias <strong>in</strong> figure 3.23. The cap devices have a NFm<strong>in</strong> optimum bias at 45 mA <strong>in</strong>stead of<br />

10 mA as seen with normal <strong>HEMTs</strong>. Rn <strong>and</strong> NFm<strong>in</strong> follow remarkably similar trends,<br />

further re<strong>in</strong>forc<strong>in</strong>g arguments made earlier <strong>in</strong> the chapter. NF m<strong>in</strong> for the thick cap is<br />

much lower at higher bias<strong>in</strong>gs than a st<strong>and</strong>ard HEMT, but is higher at low bias<strong>in</strong>gs.<br />

The reason for this is likely the large gate leakage the device has at a low bias (68 µA<br />

at Ids 10 mA for a 150 µm wide device). If the leakage can be controlled, these devices<br />

91


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

r n<br />

3.6<br />

3.4<br />

3.2<br />

3.0<br />

2.8<br />

2.6<br />

2.4<br />

2.2<br />

2.0<br />

0 20 40 60 80 100<br />

2.0<br />

1.8<br />

1.6<br />

1.4<br />

1.2<br />

1.0<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

St<strong>and</strong>ard HEMT<br />

Thick Cap<br />

Current (mA)<br />

(a)<br />

0.0<br />

0 20 40 60 80<br />

Current (mA)<br />

(c)<br />

100<br />

|Γ|<br />

Ga<strong>in</strong> (dB)<br />

1.0<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0.0<br />

0 20 40 60 80 100<br />

18<br />

16<br />

14<br />

12<br />

10<br />

8<br />

6<br />

Current (mA)<br />

(b)<br />

4<br />

0 20 40 60 80<br />

Current (mA)<br />

(d)<br />

100<br />

Figure 3.23: <strong>Noise</strong> parameters versus dra<strong>in</strong>-source current of a thick cap device <strong>and</strong><br />

a st<strong>and</strong>ard HEMT: (a) M<strong>in</strong>imum noise figure (b) magnitude <strong>and</strong> phase of optimum<br />

reflection coefficient (c) noise resistance (d) <strong>and</strong> associated ga<strong>in</strong>. Measurement frequency<br />

is 10 GHz.<br />

92<br />

180<br />

160<br />

140<br />

120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

0<br />

Phase Γ (degrees)


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

show great promise for noise performance.<br />

3.6 Comparison of High-Performance <strong>GaN</strong> <strong>HEMTs</strong> to<br />

Other Material Systems<br />

By now, the reader might have asked the question, “How does <strong>GaN</strong> HEMT noise<br />

figure performance compare to other material systems?” This will now be addressed.<br />

Most all the devices <strong>in</strong> this work have a gate length of 0.7 µm due to equipment lim-<br />

itations. It is hard to f<strong>in</strong>d published noise results <strong>in</strong> the literature at such a relatively<br />

“long” gate length. In addition, many published noise figure results have large varia-<br />

tions <strong>in</strong> the accuracy of their measurements.<br />

Bear<strong>in</strong>g this disclaimer <strong>in</strong> m<strong>in</strong>d, 0.15 µm gate-length devices optimized for high-<br />

frequency performance were obta<strong>in</strong>ed from Tomás Palacios. These devices are sim-<br />

ilar to those <strong>in</strong> a recent publication that had an fτ of 150 GHz <strong>and</strong> fmax of over<br />

200 GHz [23]. The measured NFm<strong>in</strong> aga<strong>in</strong>st frequency is <strong>in</strong> figure 3.24 for two de-<br />

vices. Repeated measurements at 10 GHz gave a consistent NFm<strong>in</strong> of 0.4 dB.<br />

Keep<strong>in</strong>g <strong>in</strong> m<strong>in</strong>d this very good value, let us turn our attention to table 3.1. Here<br />

we have an abundance of data on <strong>HEMTs</strong> from different material systems: <strong>GaN</strong>,<br />

SiGe, InAlGaAs systems on GaAs <strong>and</strong> on InP. The gate lengths, gate widths, NFm<strong>in</strong>,<br />

measurement frequency for NFm<strong>in</strong>, relevant <strong>in</strong>formation, <strong>and</strong> the reference number<br />

found at the end of the chapter are all listed. Most obvious is that SiGe has some work<br />

to do before its noise performance can be competitive with the other materials. InP<br />

93


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

Lg (µm) Wg ( µm ) NFm<strong>in</strong> (dB) Freq. (GHz) Note Ref.<br />

<strong>GaN</strong> <strong>HEMTs</strong><br />

0.12 100 0.72 12 SiC substrate [24]<br />

0.15 150 0.4 10 T. Palacios, UCSB —<br />

0.15 200 0.6 10 SiC substrate [25]<br />

0.15 100 0.75 10 SiC substrate [26]<br />

0.17 100 1.1 10 Si substrate [27]<br />

0.18 100 0.7 12 Sapphire substrate [8]<br />

0.25 100 0.8 10 SiC sub., from graph [28]<br />

0.25 100 1.05 18 Sap., from graph [9]<br />

0.25 100 1.04 10 Sapphire substrate [29]<br />

0.25 200 1.9 10 SiC substrate [30]<br />

Si/SiGe <strong>HEMTs</strong><br />

0.1 100 1.6 10 from graph, pads de-emb. [31]<br />

0.1 90 1.7 10 [32]<br />

0.1 — 2.1 10 pads de-embedded [33]<br />

0.1 100 3.2 10 [34]<br />

(In,Al)GaAs/(In,Ga,Al)As on GaAs <strong>HEMTs</strong><br />

0.1 200 0.3 10 from graph [35]<br />

0.1 — 0.51 18 [36]<br />

0.1 100 0.64 26 mHEMT [37]<br />

0.1 100 1.1 40 mHEMT [37]<br />

0.13 140 0.31 12 pHEMT [38]<br />

0.13 140 0.45 18 pHEMT [38]<br />

0.25 — 0.7 18 [36]<br />

(In,Al)GaAs/(In,Ga,Al)As on InP <strong>HEMTs</strong><br />

0.1 50 0.8 60 [39]<br />

0.1 80 0.45 10 from graph [40]<br />

0.1 80 0.61 20 from graph [40]<br />

0.15 – 0.4 10 [41]<br />

0.2 – 0.48 10 [42]<br />

0.2 – 0.8 26 [42]<br />

Table 3.1: M<strong>in</strong>imum noise figure for HEMT devices <strong>in</strong> many technologies. Also listed<br />

are the gate length <strong>and</strong> width, measurement frequency, some necessary <strong>in</strong>formation,<br />

<strong>and</strong> the reference.<br />

94


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

NF m<strong>in</strong> (dB)<br />

0.9<br />

0.8<br />

0.7<br />

0.6<br />

0.5<br />

0.4<br />

0.3<br />

0.2<br />

0.1<br />

0.0<br />

6 8 10 12 14 16<br />

Frequency (GHz)<br />

Figure 3.24: M<strong>in</strong>imum noise figure of two 0.15 µm gate length transistors provided<br />

by Tomás Palacios.<br />

provides the best performance, but GaAs <strong>and</strong> <strong>GaN</strong> are close competition <strong>in</strong> the X-<br />

b<strong>and</strong>. GaAs might have slightly better performance than <strong>GaN</strong>, but a 0.1 dB advantage<br />

can be lost once the device is put <strong>in</strong> a circuit.<br />

3.7 Summary<br />

This chapter looked at a plethora of noise figure measurements of <strong>GaN</strong> <strong>HEMTs</strong>.<br />

The methodology for procedures was expla<strong>in</strong>ed <strong>and</strong> its importance made clear. Fac-<br />

tors that changed NFm<strong>in</strong> (such as an AlN-<strong>in</strong>terlayer <strong>and</strong> choice of substrate) were <strong>in</strong>-<br />

vestigated. The importance of monitor<strong>in</strong>g gate leakage was found <strong>and</strong> analyzed with<br />

the model<strong>in</strong>g <strong>in</strong> § 2.6. The unexpected result of a FP improv<strong>in</strong>g NFm<strong>in</strong> was discovered<br />

95


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

<strong>and</strong> fully <strong>in</strong>vestigated, also with the use of the model<strong>in</strong>g from § 2.6.<br />

Together with model<strong>in</strong>g from the previous chapter, this chapter helps po<strong>in</strong>t out ways<br />

of obta<strong>in</strong><strong>in</strong>g the best NFm<strong>in</strong> possible. Most important is reduc<strong>in</strong>g the gate leakage <strong>and</strong><br />

parasitic resistances at the <strong>in</strong>put. Rg <strong>and</strong> Rs also need to be m<strong>in</strong>imized. The small-<br />

signal power ga<strong>in</strong> needs to be kept as high as possible. Self-heat<strong>in</strong>g causes an <strong>in</strong>crease<br />

<strong>in</strong> NFm<strong>in</strong> because of loss of device ga<strong>in</strong>.<br />

References<br />

[1] V. K. Paidi, “MMIC Power Amplifiers <strong>in</strong> <strong>GaN</strong> HEMT <strong>and</strong> InP HBT Technologies,”<br />

Ph.D. dissertation, University of California, Santa Barbara, Sept. 2004.<br />

[2] J. Xu, “<strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> High-Electron-Mobility Transistors Based Flip-Chip Integrated<br />

Broadb<strong>and</strong> Power Amplifiers,” Ph.D. dissertation, University of California,<br />

Santa Barbara, Dec. 2000.<br />

[3] R. Vetury, “Polarization Induced 2DEG <strong>in</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>: On the Orig<strong>in</strong>,<br />

DC <strong>and</strong> Transient Characterization,” Ph.D. dissertation, University of California,<br />

Santa Barbara, Dec. 2000.<br />

[4] K. Krishnamurthy, “Ultra-Broadb<strong>and</strong>, Efficient, <strong>Microwave</strong> Power Amplifiers <strong>in</strong><br />

Gallium Nitride HEMT Technology,” Ph.D. dissertation, University of California,<br />

Santa Barbara, May 2000.<br />

[5] S. Lee, K. J. Webb, V. Tilak, <strong>and</strong> L. Eastman, “Intr<strong>in</strong>sic <strong>Noise</strong> Equivalent-<br />

Circuit Paramters for <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” IEEE Trans. <strong>Microwave</strong> Theory<br />

Tech., vol. 51, pp. 1567–1577, May 2003.<br />

[6] S. Lee, “Intr<strong>in</strong>sic <strong>Noise</strong> Characteriestics of Gallium Nitride High Electron Mobility<br />

Transistors,” Ph.D. dissertation, Purdue University, Aug. 2004.<br />

[7] S. Nutt<strong>in</strong>ck, E. Gebara, J. Laskar, <strong>and</strong> M. Harris, “High-Frequency <strong>Noise</strong> <strong>in</strong> Al-<br />

<strong>GaN</strong>/<strong>GaN</strong> HFETs,” IEEE <strong>Microwave</strong> Components Lett., vol. 13, no. 4, pp. 149–<br />

151, Apr. 2003.<br />

[8] W. Lu, V. Kumar, R. Schw<strong>in</strong>dt, E. P<strong>in</strong>er, <strong>and</strong> I. Adesida, “DC, RF, <strong>and</strong> <strong>Microwave</strong><br />

<strong>Noise</strong> Performance of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> on Sapphire Substrates,” IEEE Trans.<br />

<strong>Microwave</strong> Theory Tech., vol. 50, pp. 2499–2503, Nov. 2002.<br />

96


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

[9] ——, “DC, RF, <strong>and</strong> <strong>Microwave</strong> <strong>Noise</strong> Performance of <strong>Al<strong>GaN</strong></strong>-<strong>GaN</strong> Field Effect<br />

Transistors Dependence of Alum<strong>in</strong>um Concentration,” IEEE Trans. Electron Devices,<br />

vol. 50, pp. 1069–1074, Apr. 2003.<br />

[10] L. Shen, S. Heikman, B. Moran, R. Coffie, N.-Q. Zhang, D. Buttair,<br />

P. Smorchkova, S. Keller, S. DenBaars, <strong>and</strong> U. Mishra, “<strong>Al<strong>GaN</strong></strong>/AlN/<strong>GaN</strong> High-<br />

Power <strong>Microwave</strong> HEMT,” IEEE Electron Devices Lett., vol. 22, no. 10, pp. 457–<br />

459, Oct. 2001.<br />

[11] T. Palacios, S. Rajan, A. Chakraborty, S. Heikman, S. Keller, S. DenBaars, <strong>and</strong><br />

U. Mishra, “Influence of the Dynamic Access Resistance <strong>in</strong> the gm <strong>and</strong> fτ L<strong>in</strong>earity<br />

of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” IEEE Trans. Electron Devices, vol. 52, no. 10, pp. 2117–<br />

2123, Oct. 2005.<br />

[12] C. H. Oxley, “A Simple Approach Includ<strong>in</strong>g Gate Leakage for Calculat<strong>in</strong>g the<br />

M<strong>in</strong>imum <strong>Noise</strong> Figure of <strong>GaN</strong> <strong>HEMTs</strong>,” <strong>Microwave</strong> <strong>and</strong> Optical Technology Letters,<br />

vol. 33, no. 2, pp. 113–115, Apr. 2002.<br />

[13] F. Danneville, G. Dambr<strong>in</strong>e, H. Happy, <strong>and</strong> A. Cappy, “Influence of the Gate Leakage<br />

Current on the <strong>Noise</strong> Performance of MESFETs <strong>and</strong> MODFETs,” IEEE <strong>Microwave</strong><br />

Theory <strong>and</strong> Tech. Symp., pp. 373–376, 1993.<br />

[14] D.-S. Sh<strong>in</strong>, J. B. Lee, S. M<strong>in</strong>, J.-E. Oh, Y. J. Park, W. Jung, <strong>and</strong> D. S. Ma, “Analytical<br />

<strong>Noise</strong> Model with the Influence of Shot <strong>Noise</strong> Induced by the Gate Leakage<br />

Current for Submicrometer Gate-Length High-Electron-Mobility Transistors,”<br />

IEEE Trans. Electron Devices, vol. 44, no. 11, pp. 1883 – 1887, Nov. 1997.<br />

[15] Y.-F. Wu, A. Saxler, M. Moore, R. P. Smith, S. Sheppard, P. Chavarkar, T. Wisleder,<br />

U. Mishra, <strong>and</strong> P. Parikh, “30-W/mm <strong>GaN</strong> <strong>HEMTs</strong> by Field Plate Optimization,”<br />

IEEE Electron Devices Lett., vol. 25, no. 3, pp. 117–119, Mar. 2004.<br />

[16] T. Palacios, “Optimization of the High Frequency Performance of Nitride-Based<br />

Transistors,” Ph.D. dissertation, University of California, Santa Barbara, Mar.<br />

2006.<br />

[17] C. Sanabria, H. Xu, T. Palacios, A. Chakraborty, S. Heikman, U. Mishra, <strong>and</strong><br />

R. York, “ Influence of the Heterostructure Design on <strong>Noise</strong> Figure of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong><br />

<strong>HEMTs</strong>,” <strong>in</strong> Device Research Conference, vol. 1, Jun. 2004, pp. 43–44.<br />

[18] Y.-F. Wu, M. Moore, T. Wisleder, P. Chavarkar, <strong>and</strong> P. Parikh, “<strong>Noise</strong> Characteristics<br />

of Field-Plated <strong>GaN</strong> <strong>HEMTs</strong>,” International Journal of High Speed <strong>Electronics</strong><br />

<strong>and</strong> Systems, vol. 14, pp. 192–194, 2004.<br />

[19] A. Ch<strong>in</strong>i, D. Buttari, R. Coffie, S. Heikman, S. Keller, <strong>and</strong> U. Mishra, “12 W/mm<br />

Power Density <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> on Sapphire Substrate,” IEEE Electron. Lett.,<br />

vol. 40, no. 1, pp. 73–74, Jan. 2004.<br />

97


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

[20] H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, Y. Wei, S. Heikman, S. Keller, U. Mishra, <strong>and</strong><br />

R. York, “A New Field-Plated <strong>GaN</strong> HEMT Structure with Improved Power <strong>and</strong><br />

<strong>Noise</strong> Performance,” <strong>in</strong> Lester Eastman Conference on High Performance Devices,<br />

Aug. 2004.<br />

[21] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, 2nd ed.<br />

New York: Cambridge University Pess, 2004.<br />

[22] L. Shen, R. Coffie, D. Buttari, S. Heikman, A. Chakraborty, A. Ch<strong>in</strong>i,<br />

S. Keller, S. DenBaars, <strong>and</strong> U. Mishra, “High-Power Polarization-Eng<strong>in</strong>eered<br />

<strong>GaN</strong>/<strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> Without Surface Passivation,” IEEE Electron Devices<br />

Lett., vol. 25, no. 1, pp. 7–9, Jan. 2004.<br />

[23] T. Palacios, A. Chakraborty, S. Heikman, S. Keller, S. DenBaars, <strong>and</strong> U. Mishra,<br />

“Algan/gan high electron mobility transistors with <strong>in</strong>gan back-barriers,” IEEE<br />

Electron Devices Lett., vol. 27, no. 1, pp. 13–15, Jan. 2006.<br />

[24] W. Lu, J. Yang, M. Khan, <strong>and</strong> I. Adesida, “<strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> on SiC with over<br />

100 GHz fT <strong>and</strong> Low <strong>Microwave</strong> <strong>Noise</strong>,” IEEE Trans. Electron Devices, vol. 48,<br />

no. 3, pp. 581–585, Mar. 2001.<br />

[25] N. Nguyen, M. Micovic, W.-S. Wong, P. Hashimoto, P. Janke, D. Harvey, <strong>and</strong><br />

C. Nguyen, “Robust Low <strong>Microwave</strong> <strong>Noise</strong> <strong>GaN</strong> MODFETs with 0.6 dB <strong>Noise</strong><br />

Figure at 10 GHz,” IEEE Electron. Lett., vol. 36, pp. 469–471, March 2000.<br />

[26] J. Moon, M. Micovic, A. Kurdoghlian, P. Janke, P. Hashimoto, W.-S. Wong, L. Mc-<br />

Cray, <strong>and</strong> C. Nguyen, “<strong>Microwave</strong> <strong>Noise</strong> Performance of <strong>Al<strong>GaN</strong></strong><strong>GaN</strong> <strong>HEMTs</strong><br />

With Small DC Power Dissipation,” IEEE Electron Devices Lett., vol. 23, no. 11,<br />

pp. 637–639, Nov. 2002.<br />

[27] A. M<strong>in</strong>ko, V. Hol, S. Lepilliet, G. Dambr<strong>in</strong>e, J. C. De Jaeger, Y. Cordier, F. Semond,<br />

F. Natali, <strong>and</strong> J. Massies, “High <strong>Microwave</strong> <strong>and</strong> <strong>Noise</strong> Performance of 0.17-<br />

µm <strong>Al<strong>GaN</strong></strong><strong>GaN</strong> <strong>HEMTs</strong> on High-Resistivity Silicon Substrates,” IEEE Electron<br />

Devices Lett., vol. 25, no. 4, pp. 167–169, Apr. 2004.<br />

[28] J.-W. Lee, V. Kumar, R. Schw<strong>in</strong>dt, A. Kuliev, R. Birkhahn, D. Gotthold, <strong>and</strong><br />

S. Guo, “<strong>Microwave</strong> <strong>Noise</strong> Performances of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong> on Semi-<br />

Insulat<strong>in</strong>g 6H-SiC Substrates,” Electron. Lett., vol. 40, no. 1, pp. 80–81, Jan. 2004.<br />

[29] A. P<strong>in</strong>g, E. P<strong>in</strong>er, J. Redw<strong>in</strong>g, M. Khan, <strong>and</strong> I. Adesida, “<strong>Microwave</strong> <strong>Noise</strong> Performance<br />

of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” Electron. Lett., vol. 36, no. 2, pp. 175–176, Jan.<br />

2000.<br />

[30] S. Hsu <strong>and</strong> D. Pavlidis, “Low <strong>Noise</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> MODFETs with High Breakdown<br />

<strong>and</strong> Power Characteristics,” Gallium Arsenide Integrated Circuit (GaAs IC)<br />

Symposium, 2001. 23rd Annual Technical Digest, pp. 229–232, Oct. 2001.<br />

98


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

[31] F. Aniel, M. Enciso-Aguilar, L. Giguerre, P. Crozat, R. Adde, T. Mack, U. Seiler,<br />

T. Hackbarth, H. Herzog, U. Konig, <strong>and</strong> B. Raynor, “High Performance 100 nm Tgate<br />

Stra<strong>in</strong>ed Si/Si0.6Ge0.4 n-MODFET,” <strong>in</strong> International Semiconductor Device<br />

Research Symposium, vol. 47, Dec. 2001, pp. 482–485.<br />

[32] S. J. Koester, J. O. Chu, <strong>and</strong> C. S. Webster, “High-Frequency <strong>Noise</strong> Performance<br />

of SiGe p-Channel MODFETs,” IEEE Electron. Lett., vol. 36, no. 7, pp. 674–765,<br />

Mar. 2000.<br />

[33] M. Enciso, F. Aniel, P. Crozat, R. Adde, M. Zeuner, A. Fox, <strong>and</strong> T. Hackbarth, “0.3<br />

dB M<strong>in</strong>imum <strong>Noise</strong> Figure at 2.5 GHz of 0.13 µm Si/Si0.58Ge0.42 n-MODFETs,”<br />

Electron. Lett., vol. 37, no. 17, pp. 1089–1090, Aug. 2001.<br />

[34] W. Lu, A. Kuliev, S. Koester, X.-W. Wang, J. Chu, T.-P. Ma, <strong>and</strong> I. Adesida,<br />

“High Performance 0.1 µm Gate-Length p-type SiGe MODFET’s <strong>and</strong> MOS-<br />

MODFET’s,” IEEE Trans. Electron Devices, vol. 47, no. 8, pp. 1645–1652, Aug.<br />

2000.<br />

[35] H. Kawasaki, T. Sh<strong>in</strong>o, M. Kawano, <strong>and</strong> K. Kamei, “Super Low <strong>Noise</strong> Al-<br />

GaAs/GaAs HEMT with One Tenth Micron gate,” <strong>Microwave</strong> Symposium Digest,<br />

1989., IEEE MTT-S International, pp. 423–426, 13-15 June 1989.<br />

[36] J. Mateos, D. Pardo, T. Gonzlez, P. Tadyszak, F. Danneville, <strong>and</strong> A. Cappy, “Influence<br />

of Al Mole Fraction on the <strong>Noise</strong> Performance of GaAs/Al xGa1−xAs<br />

HEMT’s,” IEEE Trans. Electron Devices, vol. 45, pp. 2081–2083, Sept. 1998.<br />

[37] H. S. Yoon, J. H. Lee, J. Y. Shim, J. Y. Hong, D. M. Kang, <strong>and</strong> K. H. Lee, “Extremely<br />

Low <strong>Noise</strong> Characteristics of 0.1 µm Gamma-Gate Power Metamorphic<br />

<strong>HEMTs</strong> on GaAs Substrate,” <strong>in</strong> International Conference on Indium Phosphide<br />

<strong>and</strong> Related Materials, 2005, pp. 133–136.<br />

[38] J.-H. Lee, H.-S. Yoon, C.-S. Park, <strong>and</strong> P. Hyung-Moo, “Ultra Low <strong>Noise</strong> Characteristics<br />

of AlGaAs/InGaAs/GaAs Pseudomorphic HEMT’s with Wide Head T-<br />

Shaped Gate,” IEEE Electron Devices Lett., vol. 16, pp. 271–273, June 1995.<br />

[39] M.-Y. Kao, K. Duh, P. Ho, <strong>and</strong> P.-C. Chao, “An Extremely Low-<strong>Noise</strong> InP-Based<br />

HEMT with Silicon Nitride Passivation,” Electron Devices Meet<strong>in</strong>g Technical Digest.,<br />

International, pp. 907–910, Dec. 1994.<br />

[40] M. Murti, J. Laskar, S. Nutt<strong>in</strong>ck, S. Yoo, A. Raghavan, J. Bergman, J. Bautista,<br />

R. Lai, R. Grundbacher, M. Barsky, P. Ch<strong>in</strong>, <strong>and</strong> P. Liu, “Temperature-Dependent<br />

Small-Signal <strong>and</strong> <strong>Noise</strong> Parameter Measurements <strong>and</strong> Model<strong>in</strong>g on InP <strong>HEMTs</strong>,”<br />

IEEE Trans. <strong>Microwave</strong> Theory Tech., vol. 48, no. 12, pp. 2579–2587, Dec. 2000.<br />

[41] Y. Ando, A. Cappy, K. Marubashi, K. Onda, H. Miyamoto, <strong>and</strong> M. Kuzuhara,<br />

“<strong>Noise</strong> Parameter Model<strong>in</strong>g for InP-Based Pseudomorphic <strong>HEMTs</strong>,” IEEE Trans.<br />

Electron Devices, vol. 44, no. 9, pp. 1367–1374, Sept. 1997.<br />

99


CHAPTER 3. NOISE FIGURE MEASUREMENTS AND STUDIES<br />

[42] H. C. Duran, B.-U. H. Klepser, <strong>and</strong> W. Bachtold, “Low-<strong>Noise</strong> Properties of Dry<br />

Gate Recess Etched InP HEMT’s,” IEEE Electron Devices Lett., vol. 17, pp. 482–<br />

484, Oct. 1996.<br />

100


4<br />

Low-Frequency <strong>Noise</strong> of <strong>GaN</strong> <strong>HEMTs</strong><br />

4.1 Introduction<br />

IT is surpris<strong>in</strong>g that a subject that has been studied for almost 80 years would not be<br />

well-understood. But such is the case of low-frequency noise (LFN), the largest<br />

contributor to phase noise of oscillators [1]. The ma<strong>in</strong> motivation of this chapter is to<br />

create a model for circuit simulation. Most of the LFN literature for <strong>GaN</strong> consists of<br />

theoretical explorations of devices (<strong>in</strong> some studies, just films), or measurements at<br />

very low device bias<strong>in</strong>g not suitable for practical device model<strong>in</strong>g. <strong>GaN</strong> HEMT LFN<br />

studies at bias<strong>in</strong>gs typical for circuits are presented <strong>in</strong> this chapter. A new empirical<br />

model is also presented. It is difficult to make LFN comparisons, particularly to most<br />

published works due to the many ways the data are presented, but measurements by<br />

the author for both GaAs <strong>and</strong> <strong>GaN</strong> <strong>HEMTs</strong> will be exam<strong>in</strong>ed. In addition to the above<br />

topics, the LFN setup will be described <strong>in</strong> detail, as most LFN setups are either poorly<br />

described <strong>in</strong> papers or cannot bias as high as is necessary for <strong>GaN</strong> <strong>HEMTs</strong>.<br />

101


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

4.2 Review of Low-Frequency <strong>Noise</strong><br />

The terms LFN, flicker, 1/f, <strong>and</strong> generation-recomb<strong>in</strong>ation (G-R) noise need clari-<br />

fication. Flicker (also called 1/f) noise refers to phenomena that generate noise with a<br />

slope <strong>in</strong>versely proportional to the frequency, hence “one on f.” A sample spectrum,<br />

show<strong>in</strong>g flicker noise <strong>and</strong> other contributions, is shown <strong>in</strong> figure 4.1. Resistors <strong>and</strong><br />

bulk material tend to be strictly 1/f, but devices, <strong>in</strong>clud<strong>in</strong>g <strong>HEMTs</strong>, deviate as 1/f γ<br />

where γ is usually <strong>in</strong> the range of 0.7 to 1.3. For <strong>GaN</strong> <strong>HEMTs</strong>, experiments show γ<br />

to be between 1 <strong>and</strong> 1.3 [2–4]. Generation-recomb<strong>in</strong>ation noise refers to trap-related<br />

capture <strong>and</strong> emission of carriers creat<strong>in</strong>g a spectrum of the form<br />

SX(f) = �<br />

4∆X 2� τ<br />

1+ω2τ 2<br />

(4.2.1)<br />

that is measured as noise at low frequencies. Here, τ is the trap life time, ω the angular<br />

frequency, <strong>and</strong> X a quantity that fluctuates (usually charge or mobility). Sometimes<br />

the G-R spectrum, which will have a Lorentzian power spectral density, manifests<br />

itself as a “bulge” on a 1/f spectrum, shown <strong>in</strong> figure 4.1. There is also a theory, most<br />

often credited to McWhorter [5], 1 that a cont<strong>in</strong>uum of traps, with spectra of the form<br />

<strong>in</strong> equation 4.2.1, of different life times is the source of the 1/f noise spectrum. This<br />

model appears to work for CMOS [2], but it is not accepted for all device technologies.<br />

Therefore, when referr<strong>in</strong>g to the noise at these low frequencies (less than ∼10 MHz),<br />

be it from g-r, 1/f, or other unknown sources, the expression LFN will be used.<br />

1 In truth, he did not create the concept but extended the theory.<br />

102


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

<strong>Noise</strong><br />

Spectral<br />

Density<br />

(I 2 or V 2 )<br />

1/f γ noise<br />

G-R "Bulge"<br />

1/f Reference L<strong>in</strong>e<br />

<strong>Noise</strong> spurs<br />

(outside source/<strong>in</strong>terference)<br />

Corner Frequency<br />

<strong>Noise</strong> Floor<br />

Frequency<br />

Figure 4.1: Sketch of the key features of low-frequency noise.<br />

It has already been h<strong>in</strong>ted that traps with energies <strong>in</strong> the material b<strong>and</strong>-gap are<br />

one source of LFN. There are other possible sources: surface effects (such as surface<br />

states), the bulk material itself, dislocations, tunnel<strong>in</strong>g, non-uniform channel resis-<br />

tance, quantum effects, <strong>and</strong> others. Evidence for try<strong>in</strong>g to expla<strong>in</strong> LFN from any one<br />

of these possible sources can be found <strong>in</strong> the literature. It is likely due to a comb<strong>in</strong>ation<br />

of several sources <strong>and</strong> that there will never be a unify<strong>in</strong>g 1/f model for all devices.<br />

<strong>HEMTs</strong> show orders of magnitudes more LFN noise than bipolar junction-transistor<br />

(BJT) types of devices. This suggests a surface area or high-field dependence. Early<br />

<strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> material showed much worse LFN noise than results reported later.<br />

The excess <strong>in</strong> the LFN was attributed to traps <strong>and</strong> dislocations [3]. Le<strong>in</strong>shte<strong>in</strong> re-<br />

103


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

ported [3] that <strong>GaN</strong> <strong>HEMTs</strong> had less LFN than observed <strong>in</strong> measurements of <strong>GaN</strong><br />

th<strong>in</strong> films, <strong>and</strong> suggested there was some suppression of LFN caused by the device.<br />

Some types of LFN vary with the square of bias current, such as that of diodes. A<br />

relationship that describes this, <strong>and</strong> that is used <strong>in</strong> circuit simulators is:<br />

�<br />

i 2 �<br />

g,1/f = Kf<br />

I Af<br />

DC<br />

f Ffe<br />

(4.2.2)<br />

Kf, Af, <strong>and</strong> Ffe are all fitt<strong>in</strong>g parameters, IDC is the DC current, <strong>and</strong> f is fre-<br />

quency. For mak<strong>in</strong>g theoretical comparisons of voltage, current, <strong>and</strong> resistance LFN<br />

spectrums (SV (f), SI(f), <strong>and</strong> SR(f) respectively), the spectrum is often normalized<br />

(SV (f)/V 2 DC , SI(f)/I 2 DC , <strong>and</strong> SR(f)/R 2 respectively). These spectrums are related<br />

through Ohm’s law, <strong>and</strong> lead to the question, “What’s really fluctuat<strong>in</strong>g <strong>and</strong> caus<strong>in</strong>g<br />

1/f noise?” If it is the resistance (which for <strong>HEMTs</strong>, means the channel resistance),<br />

it has been reasoned that it is because of changes <strong>in</strong> mobility or charge. Because the<br />

resistance varies with the <strong>in</strong>verse of mobility <strong>and</strong> charge, fluctuations <strong>in</strong> one of these<br />

two quantities generates 1/f noise (<strong>in</strong> theory at least). This has lead to the two ma-<br />

jor pr<strong>in</strong>ciples on which most 1/f theories are based: carrier density fluctuation <strong>and</strong><br />

mobility fluctuation model<strong>in</strong>g. The former is used <strong>in</strong> the G-R model (equation 4.2.1)<br />

discussed earlier. In fact, equation 4.2.1 can be used to describe both models. It is<br />

debated whether charge or mobility fluctuations are the source of noise; data supports<br />

both. These models have been applied to describe <strong>GaN</strong> HEMT LFN [2–4].<br />

104


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

An <strong>in</strong>terest<strong>in</strong>g empirical relationship was proposed <strong>in</strong> 1969 by Hooge:<br />

SI(f)<br />

I 2 DC<br />

= αH<br />

Nf<br />

(4.2.3)<br />

where N is the number of carriers <strong>in</strong> the device <strong>and</strong> αH is a dimensionless quantity<br />

known as the Hooge parameter. It has s<strong>in</strong>ce been used to characterize many materials<br />

<strong>and</strong> devices. It has also been used as a figure of merit <strong>and</strong> at times as a constant, such<br />

as for characteriz<strong>in</strong>g 1/f noise of materials. However, accord<strong>in</strong>g to Hooge himself, it<br />

was not meant to be considered a constant [6]. αH was extracted for measurements<br />

that appear <strong>in</strong> this chapter. As seen <strong>in</strong> figure 4.2 (a) <strong>and</strong> (b), αH changes considerably,<br />

even with frequency of extraction (not surpris<strong>in</strong>g as the noise is not strictly 1/f). That<br />

αH is not constant can be seen from other published results as well [3]. Therefore,<br />

the author believes it should not be used for device comparisons <strong>and</strong> only for material<br />

comparisons.<br />

It is the aim of this chapter to create a LFN model that can be used <strong>in</strong> the computer-<br />

aided design (CAD) software program Advanced Design System (ADS) <strong>and</strong> to make<br />

comparisons of LFN <strong>in</strong> <strong>HEMTs</strong>. The theory discussed so far is not applicable to<br />

circuit model<strong>in</strong>g. There is not yet agreement <strong>in</strong> the theory, <strong>and</strong> data are generally<br />

at low bias<strong>in</strong>gs or not even for <strong>HEMTs</strong>. At high bias<strong>in</strong>gs where the electric fields<br />

are high, traps may no longer be the dom<strong>in</strong>ant source of LFN. There is little published<br />

literature of LFN models of <strong>HEMTs</strong> for use <strong>in</strong> circuit simulators. Therefore, work was<br />

started from scratch. A setup that could h<strong>and</strong>le the bias<strong>in</strong>gs typical of <strong>GaN</strong> <strong>HEMTs</strong><br />

105


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

(a) (b)<br />

Figure 4.2: Variation of α with (a) dra<strong>in</strong>-source voltage bias <strong>and</strong> (b) frequency of<br />

extraction for two devices on the same sample.<br />

was built. Measurements of bias-dependence <strong>and</strong> geometry were performed. Other<br />

studies of LFN performance were performed as time permitted.<br />

A f<strong>in</strong>al consideration is how to present the data. There is little consistency <strong>in</strong> units<br />

<strong>and</strong> an astonish<strong>in</strong>g number of different choices: A 2 /Hz, V 2 /Hz, A/ √ Hz, V/ √ Hz,<br />

A 2 , V 2 , nV/ √ Hz, <strong>in</strong>put-referred versions of these <strong>and</strong> representations <strong>in</strong> dB or l<strong>in</strong>ear<br />

formats. Sufficient <strong>in</strong>formation is not always presented <strong>in</strong> published articles to be<br />

able to convert from one type of units to another. How the data is presented can also<br />

give different results. For example, normaliz<strong>in</strong>g SI(f) by I 2 gives a different trend<br />

of noise versus gate width than not normaliz<strong>in</strong>g. The author decided that present<strong>in</strong>g<br />

noise as A 2 /Hz would be most familiar <strong>and</strong> convenient for a circuit designer <strong>and</strong> for<br />

comparison purposes.<br />

More <strong>in</strong>formation about LFN can be found <strong>in</strong> [3, 7–9]. A very good <strong>and</strong> easy-to-<br />

106


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

follow review can be found <strong>in</strong> [10]. The book edited by Bal<strong>and</strong><strong>in</strong>, [3], is a collection of<br />

articles about noise <strong>in</strong> <strong>GaN</strong> <strong>in</strong>clud<strong>in</strong>g two that treat LFN of <strong>GaN</strong> <strong>HEMTs</strong> extensively.<br />

4.3 Low-Frequency <strong>Noise</strong> Setup<br />

It is hard to f<strong>in</strong>d a complete description for a LFN setup. Much time was spent<br />

creat<strong>in</strong>g a satisfactory setup for <strong>GaN</strong>. Therefore, this section will expla<strong>in</strong> what an ex-<br />

perimentalist needs to know to create a setup. There are four ma<strong>in</strong> equipment concerns<br />

for a LFN setup. The first is <strong>in</strong>strumentation to measure the power spectrum versus<br />

frequency. The measurement range of <strong>in</strong>terest is usually 10 Hz to 1 MHz (possibly<br />

wider). Provided that a given spectrum analyzer can even measure such frequencies,<br />

the LFN of the analyzer itself is usually larger than the device’s noise. Increased aver-<br />

ag<strong>in</strong>g <strong>and</strong> reduction of the resolution b<strong>and</strong>width can help these problems; however, the<br />

time for a s<strong>in</strong>gle measurement becomes prohibitively long. Other <strong>in</strong>struments that can<br />

be used <strong>in</strong>clude computer-controlled oscilloscopes (a long time sample is measured<br />

<strong>and</strong> a FFT is then performed on a computer) <strong>and</strong>, the preferred tool, a dedicated FFT<br />

<strong>in</strong>strument. An HP 3561A dynamic signal analyzer (DSA) is a dedicated <strong>in</strong>strument<br />

for these types of measurements <strong>and</strong> was used <strong>in</strong> this work. It is part of the HP 3048<br />

phase-noise system. The DSA noise floor was superior to other options available <strong>and</strong><br />

the HP 3048 software could automatically control it. However, the DSA could only<br />

measure a maximum frequency of 100 kHz. Most measurements <strong>in</strong> this chapter were<br />

107


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

preformed from 10 Hz to 100 kHz with ∼300 po<strong>in</strong>ts per decade.<br />

Even with the DSA, the noise floor was not adequate for some measurements.<br />

Therefore, the second equipment concern is the noise floor for which a low-noise<br />

amplifier (LNA) is needed. A Stanford Research Systems SR560 LNA was used to<br />

improve the noise floor. Its voltage ga<strong>in</strong> was typically set to 40 dB (100 V/V, the<br />

m<strong>in</strong>imum to meet the <strong>in</strong>strument’s specified noise floor) <strong>and</strong> sometimes as high as<br />

60 dB (for gate noise measurements). The HP3048 software could be adjusted for<br />

the ga<strong>in</strong>. The LNA was run off of <strong>in</strong>ternal lead-acid batteries while measur<strong>in</strong>g. The<br />

measured noise floor for the 3561A when its <strong>in</strong>put was shorted is shown together with<br />

the measured noise floor of the LNA (also shorted <strong>in</strong>put) plus 3561A <strong>in</strong> figure 4.3. An<br />

improvement of more than four orders of magnitude is seen. The measured noise floor<br />

of the LNA <strong>in</strong> figure 4.3 is identical to its specifications.<br />

The third equipment concern is bias<strong>in</strong>g the device without <strong>in</strong>terfer<strong>in</strong>g with the mea-<br />

surement. Almost all DC power supplies, <strong>in</strong>clud<strong>in</strong>g older analog models, have active<br />

circuitry to ma<strong>in</strong>ta<strong>in</strong> the bias set po<strong>in</strong>t. This, comb<strong>in</strong>ed with noise from the AC l<strong>in</strong>es,<br />

makes them impractical to use for LFN measurements. Hence, batteries are used.<br />

9 V batteries are usually preferred as they are recognized to be low-noise. For <strong>GaN</strong><br />

<strong>HEMTs</strong>, even small devices can require high current (easily 100 mA or more). An<br />

attempt to use several 9 V batteries for the dra<strong>in</strong> lead to unexpected measurement<br />

results once the device bias <strong>in</strong>creased above the knee voltage. A rechargeable 12 V<br />

108


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

HP3561A alone<br />

with SR560<br />

Figure 4.3: Measured noise floor of the HP 3561A DSA only <strong>and</strong> with the SRS SR560<br />

LNA (short-circuited <strong>in</strong>put).<br />

lead-acid battery, that could output 90 mA without disturb<strong>in</strong>g the measurement, was<br />

used for the dra<strong>in</strong> bias. A 9 V battery was used for the gate bias<strong>in</strong>g (two <strong>in</strong> series for<br />

large threshold devices). These were used with a bias box (discussed more below).<br />

Biases of Vds from 0.1 to 12 V, Ids from 5 to 90 mA, <strong>and</strong> Vgs from -0.05 to -9 V could<br />

be obta<strong>in</strong>ed. This is far superior to most other setups, which are limited to bias<strong>in</strong>g<br />

conditions <strong>in</strong> the l<strong>in</strong>ear region of a <strong>GaN</strong> HEMT.<br />

The f<strong>in</strong>al equipment concern is to protect the setup from outside electrical <strong>and</strong> me-<br />

chanical <strong>in</strong>terference. The setup was located on a vibration-isolation table, which<br />

helped with mechanical <strong>in</strong>terference (although it could still be determ<strong>in</strong>ed when con-<br />

struction work took place <strong>in</strong> the build<strong>in</strong>g or when a peer would walk by the setup).<br />

109


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

A metal box, that could be grounded to the rest of the setup, was made to fit over<br />

the sample <strong>and</strong> probes of the RF probe station to shield aga<strong>in</strong>st electrical <strong>in</strong>terference.<br />

This was required to get a good measurement <strong>and</strong> helped to reduce spurs by ∼30 dB.<br />

Figure 4.4 shows the full schematic of the setup. The multimeters were Fluke<br />

8012As <strong>and</strong> various h<strong>and</strong>-held battery-operated units. Multimeters for voltage <strong>and</strong><br />

current could be connected while measur<strong>in</strong>g LFN, but ohm-meters had to be discon-<br />

nected. A covered bias box was built. 10 turn, 3 W, 2 kΩ potentiometers were used<br />

to vary the bias. The value of 2 kΩ was carefully selected to allow the maximum<br />

range of measurable bias<strong>in</strong>gs. Ground loops destroy a measurement, so care must be<br />

12V<br />

Bias Box<br />

9V<br />

3W<br />

2kΩ<br />

Pot.<br />

220Ω<br />

3W<br />

2kΩ<br />

Pot.<br />

10Ω<br />

+ -<br />

RL 25pF<br />

100MΩ<br />

Voltmeter<br />

Ammeter<br />

+ -<br />

Capacitors<br />

50pF to1000µF<br />

Voltmeter<br />

SR560 LNA<br />

Ga<strong>in</strong> = 1000 V/V<br />

(60 dB)<br />

BW: 0.01 to 1MHz<br />

R ds<br />

50Ω<br />

Shielded<br />

Device<br />

1MΩ<br />

HP3561A<br />

FFT Box<br />

Figure 4.4: Schematic of the setup used for device dra<strong>in</strong>-side low-frequency noise<br />

measurements.<br />

110


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

taken with the ground<strong>in</strong>g. The bias-box chassis <strong>and</strong> device shield were grounded to<br />

the negative term<strong>in</strong>al of the battery. While other setups may use a load resistance, RL,<br />

to keep the source resistance seen by the LNA a constant (mak<strong>in</strong>g it easy to convert<br />

the measured voltage spectrum to a current spectrum), the author found this additional<br />

parallel resistance to limit the measurable range of biases <strong>and</strong> the dynamic range of<br />

the DSA. Instead, the effective RL is determ<strong>in</strong>ed for each LFN measurement. This<br />

resistance is the parallel comb<strong>in</strong>ation of that seen on the dra<strong>in</strong> side of the bias box <strong>and</strong><br />

Rds of the transistor. A bank of capacitors were connected at the gate to provide an<br />

AC short while measur<strong>in</strong>g the dra<strong>in</strong> LFN.<br />

Most aspects of the measurement must be done manually, as automation would only<br />

add noise <strong>and</strong> ru<strong>in</strong> the data. The steps for a measurement are:<br />

1. Determ<strong>in</strong>e the DC Rds of the device near the bias of <strong>in</strong>terest (∆Vds/∆Ids).<br />

2. Unplug power to dra<strong>in</strong> <strong>and</strong> gate, <strong>and</strong> ground both ends of the dra<strong>in</strong> pot (us<strong>in</strong>g<br />

switches on bias-box).<br />

3. Disconnect the device <strong>and</strong> LNA.<br />

4. Measure the resistance of the bias box, Rbox. This is the parallel comb<strong>in</strong>ation of<br />

both halves of the potentiometer <strong>and</strong> associated circuitry.<br />

5. Disconnect the multimeter used to measure Rbox (this would add a voltage that<br />

disturbs the measurement).<br />

6. Reconnect the device, LNA, <strong>and</strong> bias. Wait 1 m<strong>in</strong>ute for the LNA to settle (a volt<br />

change at its <strong>in</strong>put, such as turn<strong>in</strong>g on the device or chang<strong>in</strong>g the bias, causes it<br />

to saturate).<br />

7. Measure. Adjust the measured voltage spectrum by the parallel comb<strong>in</strong>ation of<br />

resistances (SI = SV /(Rbox||Rds) 2 ).<br />

111


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

Figure 4.5: A typical low-frequency plot.<br />

The setup was checked by measur<strong>in</strong>g the noise spectrum of resistors. This is flat<br />

with a voltage spectrum equal to its thermal noise. 2 A typical HEMT’s LFN is shown<br />

<strong>in</strong> figure 4.5. The dotted l<strong>in</strong>e is a 1/f reference l<strong>in</strong>e, show<strong>in</strong>g the data is very nearly 1/f<br />

<strong>in</strong> slope. There are several spurs, <strong>in</strong>clud<strong>in</strong>g one at 60 Hz (from lights <strong>and</strong> various AC<br />

sources) <strong>and</strong> another large one at ∼70 kHz (believed to be a radio signal). The spurs<br />

were found to always be present, <strong>and</strong> are removed from all other measurements <strong>in</strong> this<br />

chapter.<br />

2 Only when DC current flows through a resistor does it generate low-frequency noise.<br />

112


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

4.4 <strong>GaN</strong> HEMT Low-Frequency <strong>Noise</strong> Model<strong>in</strong>g<br />

For model<strong>in</strong>g, the dra<strong>in</strong> <strong>and</strong> gate LFN bias-dependence need to be determ<strong>in</strong>ed. The<br />

setup <strong>in</strong> § 4.3 was used to measure dra<strong>in</strong> noise while an AC short was applied to the<br />

gate. LFN was measured with the dra<strong>in</strong> voltage held constant at Vds 5 V <strong>and</strong> the gate<br />

voltage swept (chang<strong>in</strong>g Ids). Then noise for a constant Ids of 30 mA <strong>and</strong> vary<strong>in</strong>g Vds<br />

was measured. Both of these results are plotted <strong>in</strong> figure 4.6. In this figure only three<br />

of the decade values are plotted from each full LFN measurement for convenience<br />

of display<strong>in</strong>g the data. Once the device saturates, the LFN does not change further<br />

with Ids (<strong>and</strong> hence Vgs). However, LFN <strong>in</strong>creases with Vds. This change, shown <strong>in</strong><br />

figure 4.6 (b), is more than an order of magnitude for the twelve volt bias range.<br />

The magnitude of these LFN measurements at low bias<strong>in</strong>gs is consistent with the<br />

literature. The only other published work of LFN noise of <strong>GaN</strong> <strong>HEMTs</strong> at high bias-<br />

<strong>in</strong>gs is by Hsu [4]. The measurements presented here for a saturated device agree with<br />

that work, <strong>in</strong>clud<strong>in</strong>g the key result that the noise changes with Vds. In fact, Hsu claims<br />

more bulg<strong>in</strong>g at higher biases (Vds > 12 V), <strong>and</strong> that the bulg<strong>in</strong>g broadens. This <strong>in</strong>-<br />

crease of LFN with Vds was also observed <strong>in</strong> the GaAs devices to be discussed <strong>in</strong> § 4.6.<br />

The slope, γ, was extracted <strong>and</strong> found to typically be about 1.15, vary<strong>in</strong>g from less<br />

than 1.1 to nearly 1.3. Bias dependence was not clear, but usually γ would decrease<br />

with <strong>in</strong>creas<strong>in</strong>g Vds <strong>and</strong> very slightly decrease with <strong>in</strong>creas<strong>in</strong>g Ids. The chang<strong>in</strong>g of<br />

γ with bias is not well understood. It has been shown to vary with temperature as<br />

113


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

(a) (b)<br />

Figure 4.6: Plots of the measured dra<strong>in</strong> low-frequency noise with (a) change <strong>in</strong> dra<strong>in</strong>source<br />

current <strong>and</strong> (b) voltage.<br />

well. Previous research po<strong>in</strong>ts to trap effects, tunnel<strong>in</strong>g, <strong>and</strong> hot-carriers as directions<br />

to explore [2,4].<br />

Despite the measurements be<strong>in</strong>g of noise, devices biased the same had very similar<br />

measurements. This means that devices at the same bias can be compared. This is<br />

used as a basis for device comparison studies later <strong>in</strong> the chapter.<br />

The gate LFN was measured with the dra<strong>in</strong> AC shorted. Vds was set to 5 V <strong>and</strong> the<br />

gate voltage was varied. As seen <strong>in</strong> figure 4.7, even at its noisiest the gate LFN is more<br />

than three orders of magnitude smaller than the dra<strong>in</strong> noise. Measurements above a<br />

few kHz hit the noise floor of the setup. Hsu f<strong>in</strong>ds similar results [4], as does Riddle<br />

for GaAs MESFETs [11]. It was discovered that devices that had a large change <strong>in</strong><br />

gate leakage with applied gate bias also had a large change <strong>in</strong> LFN. Conversely, if the<br />

device gate leakage was relatively constant with gate bias, the LFN did not change<br />

114


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

V,<br />

30 µA<br />

V,<br />

15 µA<br />

V,<br />

8.5 µA<br />

V gs , I gs<br />

V,<br />

4.5 µA<br />

V,<br />

2.9 µA<br />

Figure 4.7: Measured gate low-frequency noise versus gate-source voltage (<strong>and</strong><br />

current).<br />

much. The device <strong>in</strong> figure 4.7 did have a large change <strong>in</strong> leakage as evident from the<br />

x-axis label<strong>in</strong>g. γ was also very close to unity. These observations mean the gate LFN<br />

can be modeled by the st<strong>and</strong>ard diode equation, 4.2.2. This helps re<strong>in</strong>force the noise<br />

figure model<strong>in</strong>g <strong>in</strong> chapter 2 us<strong>in</strong>g a shot noise source at the gate of the transistor.<br />

It is desirable to add scal<strong>in</strong>g to circuit model<strong>in</strong>g of noise, so the effect of device<br />

geometry on LFN was also measured. Figure 4.8 shows that as the gate is made wider<br />

the dra<strong>in</strong> LFN noise <strong>in</strong>creases. This trend agrees with the low-bias <strong>GaN</strong> HEMT LFN<br />

measurements of Kuksenkov [12] <strong>and</strong> the HEMT model<strong>in</strong>g of Angelov [13]. This<br />

trend is opposite to what is exhibited by MOSFETs [14]. Given this result, it makes<br />

the gate length dependence all the more <strong>in</strong>terest<strong>in</strong>g. Figure 4.9 (a) <strong>and</strong> (b) show that a<br />

shorter gate length <strong>in</strong>creases the dra<strong>in</strong> LFN. This means that the LFN does not display<br />

115


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

Figure 4.8: Change <strong>in</strong> low-frequency noise with gate width at various decade frequencies<br />

for three devices.<br />

(a) (b)<br />

Figure 4.9: (a) Change <strong>in</strong> low-frequency noise as the gate length, Lg is changed. (b)<br />

The 1 kHz data from (a) with a best fit l<strong>in</strong>e.<br />

116


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

a direct area dependence. It can be speculated that the gate length dependence LFN is<br />

related to the Vds dependence.<br />

The Vds LFN dependence will harm the phase noise performance of <strong>GaN</strong> HEMT-<br />

based oscillators. Also, it is not typical of previous LFN model<strong>in</strong>g, which has a current<br />

dependence <strong>in</strong>stead. Based on the measurements presented thus far, the gate <strong>and</strong> dra<strong>in</strong><br />

have different LFN processes <strong>and</strong> should not be modeled <strong>in</strong> the same manner. It<br />

needs to be determ<strong>in</strong>ed whether the gate <strong>and</strong> dra<strong>in</strong> LFN are mathematically correlated.<br />

Measurements were performed with the gate AC shorted <strong>and</strong> not AC shorted, show<strong>in</strong>g<br />

no change <strong>in</strong> the dra<strong>in</strong> LFN. The same was done for gate LFN measurements <strong>and</strong> the<br />

dra<strong>in</strong> did not appear to have an effect. Hsu found similar results [4]. Lee shows<br />

that the gate <strong>and</strong> dra<strong>in</strong> LFN are uncorrelated [15]. Therefore, a two-current noise<br />

source model with no correlation as illustrated <strong>in</strong> figure 4.10 can be used. Based on<br />

the measurements <strong>in</strong> this work, the follow<strong>in</strong>g model of the dra<strong>in</strong> LFN for a circuit<br />

simulator is proposed:<br />


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

�<br />

i 2 �<br />

d,1/f = Ki<br />

� ��<br />

Wg,new Lg,old<br />

Wg,old<br />

Lg,new<br />

f γ<br />

� c<br />

(msVds +1)<br />

(4.4.1)<br />

where Ki, ms, <strong>and</strong> c are fitt<strong>in</strong>g parameters represent<strong>in</strong>g the magnitude of the noise,<br />

the slope change with Vds , <strong>and</strong> an exponent of the change with gate length respec-<br />

tively. Wg <strong>and</strong> Lg are def<strong>in</strong>ed the same as <strong>in</strong> § 2.6.4. Values found to work for devices<br />

presented are Ki = 4e-14, γ = 1.1, c = 1.5, <strong>and</strong> m = 2.6. As already mentioned, the<br />

gate noise can be modeled with equation 4.2.2. Due to the limitations of hitt<strong>in</strong>g the<br />

noise floor, the gate LFN geometry dependence could not be measured. However, as<br />

it depends on the Schottky contact reverse-bias gate leakage, it might be assumed that<br />

the gate LFN scales with the gate leakage as <strong>in</strong> equation 2.6.45. Parameters found to<br />

work for the gate LFN are Af = 1.15, Ffe = 1, <strong>and</strong> Kf = 2e-11.<br />

The model was used to estimate the dra<strong>in</strong> LFN corner frequency of the devices,<br />

as equipment limitations prevented the measurement of this important quantity. The<br />

noise figure model<strong>in</strong>g work <strong>in</strong> chapter 2 was used to estimate the noise floor, <strong>and</strong><br />

its <strong>in</strong>tersection with the LFN was calculated. This value varies with bias because of<br />

changes <strong>in</strong> LFN <strong>and</strong> ga<strong>in</strong>, but was estimated to be close be 1-10 MHz. This is a<br />

typical value for GaAs <strong>HEMTs</strong>. It was also attempted to use this model <strong>in</strong> phase noise<br />

circuit simulations <strong>in</strong> ADS. However, limitations <strong>in</strong> the device model <strong>and</strong> the lack<br />

of a native voltage bias-dependent low-frequency noise source component <strong>in</strong> ADS<br />

prevented accurate prediction of phase noise.<br />

118


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

4.5 <strong>GaN</strong> HEMT Low-Frequency <strong>Noise</strong> Studies<br />

4.5.1 Substrate<br />

There has already been some work compar<strong>in</strong>g the noise of <strong>GaN</strong> <strong>HEMTs</strong> on different<br />

substrates [3]. The f<strong>in</strong>d<strong>in</strong>gs were that SiC was less noisy than sapphire, hav<strong>in</strong>g Hooge<br />

parameters one to two orders of magnitude smaller. But the measurements were at<br />

a low bias <strong>and</strong> it has already been expla<strong>in</strong>ed why the Hooge parameter should not<br />

be used for device comparisons. There is no data published compar<strong>in</strong>g devices that<br />

are <strong>in</strong> the saturation region. This was undertaken, <strong>and</strong> is presented <strong>in</strong> figure 4.11.<br />

The devices are both biased at Vds 5 V <strong>and</strong> Ids 10 mA. Across the spectrum, there<br />

is no apparent difference. While there appears to be a slight amount of bulg<strong>in</strong>g, it is<br />

not stronger for either device. This suggests that at typical device bias<strong>in</strong>gs the LFN<br />

is suppressed or another noise mechanism is stronger. Also, identical <strong>GaN</strong> HEMT-<br />

based oscillators on both substrates constructed by the author (§ 5.4.1) did not show a<br />

difference <strong>in</strong> LFN.<br />

4.5.2 Passivation<br />

There are two previous publications on the effect passivation has on LFN <strong>in</strong> <strong>GaN</strong>.<br />

The first [16] measured <strong>HEMTs</strong> at a very low bias <strong>and</strong> normalized the noise to the<br />

dra<strong>in</strong> current. They showed that the passivation does improve the LFN performance,<br />

119


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

Figure 4.11: Measurement of devices on a sapphire <strong>and</strong> SiC substrate at a bias of Vds<br />

5V,Ids 30 mA<br />

(a) (b)<br />

Figure 4.12: (a) Low-frequency noise of a device before <strong>and</strong> after passivation. (b)<br />

Low-frequency noise at 10 Hz <strong>and</strong> 1 kHz of a device before <strong>and</strong> after passivation at<br />

different Vgs .<br />

120


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

but based on their measurements the improvement varied with gate bias from as large<br />

as an order of magnitude to almost no improvement. The other study also found<br />

an improvement with passivation but only measured TLM structures [17]. LFN was<br />

measured for a sample before <strong>and</strong> after passivation. In figure 4.12 (a), the LFN is<br />

plotted for a device with <strong>and</strong> without passivation at Vds 5 V <strong>and</strong> Ids 10 mA. Passivation<br />

improves the LFN by an order of magnitude across the spectrum. Passivation causes<br />

a slight change <strong>in</strong> threshold voltage, so measurements were taken at different gate<br />

bias<strong>in</strong>gs with Vds still at 5 V. The measured data at 10 Hz <strong>and</strong> 1 kHz is plotted <strong>in</strong><br />

figure 4.12 (b). The improvement stays relatively constant with gate voltage. It is<br />

believed that suppression of surface state traps, along with the <strong>in</strong>crease of channel<br />

charge, cause the improvement.<br />

4.5.3 Thick-Epitaxial Cap Devices<br />

After consider<strong>in</strong>g the result of passivation on LFN, it is <strong>in</strong>terest<strong>in</strong>g to look now at a<br />

device that does not require passivation. Devices of similar design to those <strong>in</strong> § 3.5.6<br />

(<strong>Al<strong>GaN</strong></strong> was used to cap the device <strong>in</strong>stead of <strong>GaN</strong>) were measured. Figure 4.13 shows<br />

LFN measurements of a few thick-cap devices (that have no passivation) <strong>and</strong> st<strong>and</strong>ard<br />

passivated <strong>HEMTs</strong>. The performance of the thick-cap devices is as good as, if not<br />

slightly better than, a passivated st<strong>and</strong>ard HEMT device. It can be concluded that a<br />

LFN mechanism on the channel surface can be suppressed with either passivation or<br />

121


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

Figure 4.13: Comparison of st<strong>and</strong>ard passivated <strong>HEMTs</strong> to an unpassivated thick cap<br />

<strong>HEMTs</strong>. Bias is Vds =5V<strong>and</strong>Ids = 30 mA.<br />

with another epitaxial layer.<br />

4.5.4 Field-Plated Devices<br />

A field plate (FP) was shown to lower NF <strong>in</strong> this work (§ 3.5.5). It has also been<br />

shown that a FP improves oscillator phase noise <strong>in</strong> [18]. In fact, the longer the FP, the<br />

less the phase noise. As the 30 dB/decade slope of this data suggests that a 1/f type of<br />

noise is dom<strong>in</strong>at<strong>in</strong>g the phase noise, it might be reasoned that the LFN of FP <strong>and</strong> non-<br />

FP devices is different. If there is a difference, it is not apparent <strong>in</strong> the measurements<br />

shown <strong>in</strong> figure 4.14. In part (a) of the figure, all FP <strong>and</strong> non-FP transistors display<br />

the same LFN at different frequencies when biased under the same conditions. The<br />

second plot, (b), shows that as Vds <strong>in</strong>creases, all devices’ LFN <strong>in</strong>crease at nearly equal<br />

122


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

(a) (b)<br />

Figure 4.14: (a) Decade low-frequency noise data for devices with different FP lengths<br />

at a bias of Vds 5 V <strong>and</strong> Ids 30 mA. (b) 100 Hz low-frequency noise for different FP<br />

lengths at Vds 3, 5, <strong>and</strong> 8 V.<br />

rates. The answer to the orig<strong>in</strong> of the phase noise improvement with a FP will need to<br />

be answered elsewhere.<br />

4.6 Comparison to GaAs <strong>HEMTs</strong><br />

In the previous chapter the NF of <strong>GaN</strong> was shown to be similar to GaAs. Let us now<br />

exam<strong>in</strong>e the LFN of these two materials. A few <strong>GaN</strong> LFN reports have claimed that<br />

the noise is comparable (such as [2]). These previous results are usually done with a<br />

measured <strong>GaN</strong> device <strong>and</strong> <strong>in</strong>formation for a GaAs device from a data sheet or another<br />

paper. In addition, the Hooge parameter is used as the figure of merit.<br />

The author obta<strong>in</strong>ed GaAs <strong>HEMTs</strong> <strong>and</strong> measured the LFN of these devices. They<br />

are from the TriQu<strong>in</strong>t TQP13 pHEMT process, with an fτ of ∼95 GHz. The device<br />

123


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

gate geometry is 0.13 x 93 µm. This is not the same as the 0.7 x 100 µm <strong>GaN</strong> HEMT<br />

devices to which they are compared to. However, based on the model presented <strong>in</strong> this<br />

chapter, approximat<strong>in</strong>g the magnitude of the GaAs devices to be 3 times smaller than<br />

what is actually measured should make for a fair comparison. The question arises<br />

of how to bias the different devices. The follow<strong>in</strong>g bias was chosen: 1.5 times the<br />

knee voltage of the fully open channel <strong>and</strong> 3/4 the total gate bias (positive turn-on to<br />

negative cut-off).<br />

Figure 4.15 shows the measurements of (a) the full spectrum of one <strong>GaN</strong> <strong>and</strong> one<br />

GaAs device <strong>and</strong> (b) selected frequency po<strong>in</strong>ts for a few <strong>GaN</strong> <strong>and</strong> GaAs devices.<br />

The GaAs has not been corrected for its difference <strong>in</strong> geometry. Consider<strong>in</strong>g this<br />

difference, the LFN from a few kHz to 1 MHz is similar for both types of devices.<br />

The slope (γ) of the GaAs devices was found to typically be 0.8 (it is common <strong>in</strong> the<br />

(a) (b)<br />

Figure 4.15: Low-frequency noise comparison of (a) a <strong>GaN</strong> <strong>and</strong> GaAs HEMT (full<br />

spectrum) <strong>and</strong> (b) multiple <strong>GaN</strong> <strong>and</strong> GaAs <strong>HEMTs</strong> (decade measurements). Bias for<br />

all devices is Vgs = 3/4 Vtotal, Vds = 1.5 Vknee.<br />

124


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

GaAs FET literature for γ to be between 0.7 <strong>and</strong> 1), which is much less than the 1.2 for<br />

<strong>GaN</strong>. Therefore, at lower frequencies GaAs has less noise. In § 5.5, it will be shown<br />

that the close-to-carrier phase noise of <strong>GaN</strong> oscillators is worse than for GaAs-based<br />

oscillators because of these different slopes.<br />

4.7 Summary<br />

This chapter has provided a useful collection of quality high-bias low-frequency<br />

noise data. It was shown that the noise depends heavily on Vds while rema<strong>in</strong><strong>in</strong>g nearly<br />

constant to a chang<strong>in</strong>g gate voltage. The geometry dependence has also been shown to<br />

be different than what is seen <strong>in</strong> other devices. A scalable, bias-dependent, empirical<br />

model was proposed that can be used <strong>in</strong> circuit simulators. It was demonstrated that<br />

neither the choice of substrate nor the addition of a FP change the LFN. Suppression<br />

of LFN from surface effects was demonstrated through measurements of unpassivated<br />

<strong>and</strong> thick-cap devices. References to the literature were added to support all measure-<br />

ments where previous work exists. The full details of a LFN setup were expla<strong>in</strong>ed. A<br />

f<strong>in</strong>al note is that bulg<strong>in</strong>g (g-r noise presumably from traps) would occasionally show<br />

up. The bulge would be very broad with a spectrum from around 100 Hz to 1 kHz. It<br />

was not always present, even across the same sample. With improvement of material<br />

it can be expected that the LFN will improve.<br />

125


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

References<br />

[1] A. Hajimiri <strong>and</strong> T. Lee, The Design of Low <strong>Noise</strong> <strong>Oscillators</strong>. Boston: Kluwer<br />

Academic Publishers, 1999.<br />

[2] A. Bal<strong>and</strong><strong>in</strong>, S. Morozov, S. Cai, R. Li, K. Wang, G. Wijeratne, <strong>and</strong><br />

C. Viswanathan, “Low Flicker-<strong>Noise</strong> <strong>GaN</strong>/<strong>Al<strong>GaN</strong></strong> Heterostructure Field-Effect<br />

Transistors for <strong>Microwave</strong> Communications,” <strong>Microwave</strong> Theory <strong>and</strong> Techniques,<br />

IEEE Transactions on, vol. 47, no. 8, pp. 1413–1417, 1999.<br />

[3] A. Bal<strong>and</strong><strong>in</strong>, Ed., <strong>Noise</strong> <strong>and</strong> Fluctuations Control <strong>in</strong> Electronic Devices. Stevenson<br />

Ranch, CA: American Scientific Publishers, 2002.<br />

[4] S. Hsu, P. Valizadeh, D. Pavlidis, J. Moon, M. Micovic, D. Wong, <strong>and</strong> T. Hussa<strong>in</strong>,<br />

“Characterization <strong>and</strong> Analysis of Gate <strong>and</strong> Dra<strong>in</strong> Low-Frequency <strong>Noise</strong> <strong>in</strong><br />

<strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> <strong>HEMTs</strong>,” <strong>in</strong> High Performance Devices, 2002. Proceed<strong>in</strong>gs. IEEE<br />

Lester Eastman Conference on, 2002, pp. 453–460.<br />

[5] A. L. McWhorter, “Semiconductor Surface Physics,” R. H. K<strong>in</strong>ston, Ed. Philadelphia:<br />

Univ. of Pennsylvania Press, 1956, pp. 207–228.<br />

[6] F. N. Hooge, “1/f <strong>Noise</strong> Sources,” IEEE Trans. Electron Devices, vol. 41, no. 11,<br />

pp. 1926–35, Nov. 1994.<br />

[7] A. van der Ziel, “Unified Presentation of 1/f <strong>Noise</strong> <strong>in</strong> Electron Devices: Fundamental<br />

1/f <strong>Noise</strong> Sources,” Proc. IEEE, vol. 76, no. 3, pp. 233–258, 1988.<br />

[8] ——, <strong>Noise</strong> <strong>in</strong> Solid State Devices <strong>and</strong> Circuits. New York: Wiley-Interscience,<br />

1986.<br />

[9] M. J. Buck<strong>in</strong>gham, <strong>Noise</strong> <strong>in</strong> Electronic Devices <strong>and</strong> Systems. New York: John<br />

Wiley & Sons, 1983.<br />

[10] D. A. Bell, “A survey of 1/f noise <strong>in</strong> electrical conductors.” Journal of Physics C:<br />

Solid State Physics, vol. 13, no. 24, pp. 4425–37, Aug. 1980.<br />

[11] A. N. Riddle, “Oscillator <strong>Noise</strong>: Theory <strong>and</strong> Characterization,” Ph.D. dissertation,<br />

North Carol<strong>in</strong>a State University, 1986.<br />

[12] D. V. Kuksenkov, H. Temk<strong>in</strong>, R. Gaska, <strong>and</strong> J. W. Yang, “Low-Frequency <strong>Noise</strong><br />

<strong>in</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> Heterostructure Field Effect Transistors,” IEEE Electron Devices<br />

Lett., vol. 19, no. 7, pp. 222–224, July 1998.<br />

[13] I. Angelov, R. Kozhuharov, <strong>and</strong> H. Zirath, “A Simple Bias Dependant LF FET<br />

<strong>Noise</strong> Model for CAD,” <strong>in</strong> <strong>Microwave</strong> Symposium Digest, 2001 IEEE MTT-S International,<br />

vol. 1, 2001, pp. 407–410 vol.1.<br />

126


CHAPTER 4. LOW-FREQUENCY NOISE OF GAN HEMTS<br />

[14] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, 2nd ed.<br />

New York: Cambridge University Pess, 2004.<br />

[15] S. Lee, “Intr<strong>in</strong>sic <strong>Noise</strong> Characteriestics of Gallium Nitride High Electron Mobility<br />

Transistors,” Ph.D. dissertation, Purdue University, Aug. 2004.<br />

[16] A. V. Vertiatchikh <strong>and</strong> L. Eastman, “Effect of the Surface <strong>and</strong> Barrier Defects<br />

on the <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT Low-Frequency <strong>Noise</strong> Performance,” IEEE Electron<br />

Devices Lett., vol. 24, no. 9, pp. 535–537, Sept. 2003.<br />

[17] S. A. Vitusevich, M. V. Petrychuk, S. V. Danylyuk, A. M. Kurak<strong>in</strong>, N. Kle<strong>in</strong>, <strong>and</strong><br />

A. E. Belyaev, “Influence of Surface Passivation on Low-Frequency <strong>Noise</strong> Properties<br />

of <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> High Electron Mobility Transistor Structures,” phys. stat. sol.<br />

(a), no. 5, pp. 816–819, Mar. 2005.<br />

[18] H. Xu, C. Sanabria, S. Heikman, S. Keller, U. Mishra, <strong>and</strong> R. York, “High Power<br />

<strong>GaN</strong> <strong>Oscillators</strong> Us<strong>in</strong>g Field-Plated HEMT Structure,” <strong>in</strong> <strong>Microwave</strong> Symposium<br />

Digest, 2005 IEEE MTT-S International, 2005, pp. 1345–1348.<br />

127


5.1 Introduction<br />

5<br />

<strong>GaN</strong> HEMT Based <strong>Oscillators</strong><br />

OSCILLATORS are a key component of many communication systems. They<br />

set fundamental limits of channel spac<strong>in</strong>g because of their phase noise. This<br />

work has demonstrated that NF <strong>and</strong> LFN of <strong>GaN</strong> <strong>HEMTs</strong> are only slightly worse<br />

than the far more technologically mature GaAs <strong>HEMTs</strong>. Here the phase noise will<br />

be compared to similar <strong>in</strong>tegrated circuit designs <strong>in</strong> Si <strong>and</strong> GaAs. Some guidel<strong>in</strong>es<br />

for low-phase noise design are discussed. A recently-developed MMIC process at<br />

UCSB [1] was ideal for mak<strong>in</strong>g the designs, <strong>and</strong> it will be reviewed briefly. The focus<br />

of the chapter is the design <strong>and</strong> measurement of two LC differential oscillators. The<br />

first did not have impressive phase noise performance but its l<strong>in</strong>earity was excellent.<br />

It is also the first <strong>GaN</strong> differential oscillator to appear <strong>in</strong> the literature [2]. A second<br />

similar oscillator was fabricated that had fairly good phase noise performance. These<br />

oscillators are compared to other published results of differential oscillators <strong>in</strong> Si <strong>and</strong><br />

GaAs, as well as <strong>GaN</strong> HEMT oscillators of all designs.<br />

128


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

5.2 Concern<strong>in</strong>g Phase <strong>Noise</strong><br />

The study of phase noise is a difficult undertak<strong>in</strong>g. Entire books <strong>and</strong> dissertations<br />

are devoted to its study [3–5]. Even today there is still not a consensus of how best to<br />

approach the problem [6]. The reason for the difficulty of underst<strong>and</strong><strong>in</strong>g phase noise<br />

stems from three challenges. The first is that an oscillator is a non-l<strong>in</strong>ear problem, <strong>and</strong><br />

simple analytical analysis cannot capture all key aspects. The rigorous work is usually<br />

too complicated for clear <strong>in</strong>sight or practical design. LFN is the largest contributor to<br />

phase noise, <strong>and</strong> the lack of its full underst<strong>and</strong><strong>in</strong>g underm<strong>in</strong>es any complete study of<br />

phase noise. It is not unheard of for a designer to simulate phase noise without LFN<br />

model<strong>in</strong>g <strong>in</strong> the circuit because of this shortcom<strong>in</strong>g. F<strong>in</strong>ally, the circuit design impacts<br />

the phase noise, possibly <strong>in</strong> profound ways [5].<br />

Def<strong>in</strong><strong>in</strong>g phase noise is much simpler than design<strong>in</strong>g for it. Any signal source is<br />

bound to have fluctuations <strong>in</strong> phase <strong>and</strong> amplitude. We can write these fluctuations as<br />

v(t) =v0(1 − a(t)) cos(ω0t + ψ(t)) (5.2.1)<br />

v0 <strong>and</strong> ω0 are the amplitude <strong>and</strong> angular frequency of resonance of the oscillator.<br />

The amplitude fluctuations, a(t), <strong>and</strong> phase fluctuations, ψ(t), are both stochastic pro-<br />

cesses. The power spectral density close to the carrier frequency is found to be [7]<br />

Sv(ω) = v2 0<br />

2 {4π(1 −〈φψ(0)〉)δ(ω − ω0)+Sa(ω − ω0)<br />

+Sψ(ω − ω0)+2Im[Sψa(ω − ω0)]} (5.2.2)<br />

129


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

Here we see the signal amplitude (v 2 0/2) at the resonant frequency, along with an<br />

autocorrelation term, φψ, from the phase fluctuations that reduces it. Sa is noise added<br />

to the spectrum from the amplitude fluctuations, <strong>and</strong> is commonly known as amplitude<br />

modulation (AM) noise. The phase fluctuations contribute to the spectrum through Sψ,<br />

<strong>and</strong> is known as phase modulation (PM) or phase noise. A correlation term between<br />

the amplitude <strong>and</strong> phase fluctuations can also contribute to the spectrum through the<br />

cross spectral density, Sψa, called AM-PM noise. AM-PM noise is an odd function<br />

(the others are even), <strong>and</strong> can lead to asymmetry of the noise spectrum around the<br />

carrier. Proper circuit design reduces AM-PM noise. The nonl<strong>in</strong>earities of the ga<strong>in</strong>-<br />

produc<strong>in</strong>g element <strong>in</strong> an oscillator (usually, but not always, a transistor) provides a<br />

restor<strong>in</strong>g force that not only keeps the amplitude stable but also greatly reduces AM<br />

noise. This means the noise near an oscillator’s resonance frequency is dom<strong>in</strong>ated by<br />

phase noise.<br />

Bear<strong>in</strong>g <strong>in</strong> m<strong>in</strong>d that there is AM, PM, <strong>and</strong> AM-PM noise, but PM (phase noise)<br />

dom<strong>in</strong>ates, the phase noise can be observed with a spectrum analyzer. In fact, an Ag-<br />

ilent E4440 spectrum analyzer with a phase noise personality was used for measure-<br />

ments that appear <strong>in</strong> this chapter. Phase noise appears as a broaden<strong>in</strong>g of the frequency<br />

of oscillation, shown <strong>in</strong> figure 5.1 (a). Its f<strong>in</strong>e detail of shape varies, <strong>and</strong> is typically<br />

one of three cases as seen <strong>in</strong> figure 5.1 (b-d). The resonator acts as a filter. Above or<br />

below the resonant frequency a voltage signal would drop proportionally to 1/f, but<br />

130


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

Power<br />

Power<br />

1/f 3<br />

1/f 2<br />

(a)<br />

(c)<br />

ω 0<br />

<strong>Noise</strong> Floor<br />

Frequency<br />

<strong>Noise</strong> Floor<br />

Frequency<br />

Power<br />

Power<br />

1/f 3<br />

1/f 3<br />

1/f<br />

(b)<br />

(d)<br />

<strong>Noise</strong> Floor<br />

Frequency<br />

<strong>Noise</strong> Floor<br />

Frequency<br />

Figure 5.1: (a) A typical spectrum of an oscillator. (b-d) are zoomed <strong>in</strong> plots of the<br />

circled portion <strong>in</strong> (a). The power <strong>in</strong> the various plots are not scaled to one another.<br />

the power would drop 1/f 2 (20 dB/decade). This is exactly what happens to both flat<br />

(such as thermal) noise sources <strong>and</strong> up-converted LFN lead<strong>in</strong>g to 20 dB/decade <strong>and</strong><br />

30 dB/decade slopes respectively. An oscillator with a very high Q, <strong>and</strong> the best noise<br />

performance (the plots <strong>in</strong> figure 5.1 are not to scale to each other), would have a shape<br />

such as that <strong>in</strong> figure 5.1 (b). The 1/f 3 region is the resonator filter<strong>in</strong>g up-converted<br />

LFN, while the 1/f regions are up-converted LFN that is outside the resonator b<strong>and</strong>-<br />

width. If there were no LFN (a blissful ideality), the slope of the entire range would be<br />

131


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

just 20 dB/decade. Figure 5.1 (c) is the most typical spectrum observed. Here aga<strong>in</strong><br />

is filter<strong>in</strong>g of LFN, but now we see thermal noise be<strong>in</strong>g filtered as well. The last case,<br />

figure 5.1 (d), is for an oscillator with an enormous amount of LFN. We will return<br />

to this qualitative analysis later. Phase noise is specified <strong>in</strong> decibels below the carrier<br />

(dBc) <strong>in</strong> a 1 Hz b<strong>and</strong>width at a offset-frequency from the carrier. For example, a phase<br />

noise of -132 dBc/Hz at 100 kHz means the noise at a frequency 100 kHz <strong>in</strong> addition<br />

to the oscillation frequency is 132 dB below the power the oscillator has at its resonant<br />

frequency (ω0).<br />

A concise background of phase noise would require more than a chapter. Despite<br />

the complexity of phase noise, there are some guidel<strong>in</strong>es that tend to show up repeat-<br />

edly [3,5–9]:<br />

• Increase the tank Q: This is the most important factor for improv<strong>in</strong>g phase<br />

noise. Higher Q means more suppression of off-carrier frequency components.<br />

• M<strong>in</strong>imize the LFN: Probably the next most important parameter after Q. Choice<br />

of device (bipolars <strong>in</strong> general have much less LFN than FETs) <strong>and</strong> device geometry<br />

can help, as well as improvement <strong>in</strong> material quality. But usually little<br />

can be done to quell LFN.<br />

• Increase the Signal Amplitude: If all other <strong>in</strong>fluences could be considered<br />

constant, a larger signal amplitude <strong>in</strong>creases signal to noise ratio <strong>and</strong> reduces<br />

the phase noise.<br />

• M<strong>in</strong>imize Other <strong>Noise</strong> Sources: Flat noise sources determ<strong>in</strong>e the phase noise<br />

<strong>in</strong> the 20 dB/decade range.<br />

• Properly Design the Loop Ga<strong>in</strong>: Design<strong>in</strong>g the phase shift of the loop to have<br />

a maximum derivative at the center of the resonance of the loop ga<strong>in</strong> maximizes<br />

the resonator’s ability to attenuate phase noise.<br />

132


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

• L<strong>in</strong>earize Cgs: The non-l<strong>in</strong>earity of Cgs with Vgs makes the oscillator less symmetric<br />

<strong>and</strong> <strong>in</strong>creases noise. Additional capacitance can smooth Cgs <strong>and</strong> improve<br />

phase noise. [3,10]<br />

• Optimize the Circuit: Some oscillator topologies, particularly the Colpitts,<br />

provide better phase noise than others. Design of the circuit <strong>in</strong>fluences the<br />

phase noise performance. Use of techniques, such as automatic ga<strong>in</strong> control<br />

<strong>and</strong> tapped resonators, can help as well.<br />

The last po<strong>in</strong>t is the largest topic to expound upon. However, the other po<strong>in</strong>ts go<br />

far toward the goal of improv<strong>in</strong>g phase noise. Because of the very large powers <strong>GaN</strong><br />

<strong>HEMTs</strong> can produce, the improvement of phase noise with signal amplitude is of great<br />

<strong>in</strong>terest to <strong>GaN</strong> circuits <strong>and</strong> provides the motivation for build<strong>in</strong>g the oscillators <strong>in</strong> this<br />

chapter.<br />

To simulate phase noise, very accurate device small-signal, large-signal, <strong>and</strong> noise<br />

model<strong>in</strong>g must be available. In addition, simulators vary <strong>in</strong> their accuracy <strong>in</strong> predict<strong>in</strong>g<br />

phase noise. Model<strong>in</strong>g from the previous chapters was used to attempt simulations<br />

of phase noise. However, the device model had some shortcom<strong>in</strong>gs (discussed <strong>in</strong><br />

§ 6.2) that prevented even accurate model<strong>in</strong>g of the 20 dB/decade phase noise from<br />

filtered thermal noise. Some <strong>in</strong>terest<strong>in</strong>g results could be determ<strong>in</strong>ed: the gate-leakage<br />

shot noise was found to be negligible compared to contributions of the gate resistance<br />

thermal noise, the channel noise, <strong>and</strong> the LFN. It was also noted that a field plate (FP)<br />

did help the simulated phase noise results. A surpris<strong>in</strong>g simulated improvement of<br />

10 dB of a very long FP HEMT oscillator over a non-FP oscillator agrees with the<br />

work of Dr. Xu [11]. The practices outl<strong>in</strong>ed above were followed as well as possible.<br />

133


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

5.3 MMIC Process Description<br />

The MMIC process developed by Hongtao Xu was used for circuit fabrication [1].<br />

It <strong>in</strong>tegrates capacitors, <strong>in</strong>ductors, resistors, <strong>and</strong> transmission l<strong>in</strong>es. This follows the<br />

HEMT process described <strong>in</strong> § 3.2, but with additional steps. There are now two addi-<br />

tional metal layers (metal 1, 1 µm of gold, <strong>and</strong> metal 2, 3 µm of gold), a resistor layer<br />

(NiCr), <strong>and</strong> a dielectric spacer layer (3 µm of PMGI). Inductors are made with metal<br />

1 as an under-pass metal, then the dielectric spacer, <strong>and</strong> metal 2 to f<strong>in</strong>ish the over-pass<br />

<strong>and</strong> most of the metalization of the square <strong>in</strong>ductors. Capacitors are made with the<br />

gate metal as a bottom electrode <strong>and</strong> the same film used for passivation as the capac-<br />

itor dielectric. Metal 1 becomes the top electrode, <strong>and</strong> metal 2 connects the capacitor<br />

bottom <strong>and</strong> top to the rest of the circuit (metal 2 is used over the PMGI to contact the<br />

top electrode without short<strong>in</strong>g the capacitor). Resistor composition is Ti/SiO2/NiCr<br />

<strong>and</strong> is protected by the same SiN film used for the capacitors <strong>and</strong> passivation. Trans-<br />

mission l<strong>in</strong>es are made exclusively with the thick metal 2. The same material structure<br />

<strong>in</strong> § 3.2 is used here. Full details of the process, along with a process flow chart, can<br />

be found <strong>in</strong> Hongtao Xu’s dissertation [1].<br />

134


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

5.4 Differential <strong>Oscillators</strong><br />

5.4.1 High L<strong>in</strong>earity Oscillator<br />

Some familiarity with oscillators will be assumed <strong>in</strong> this <strong>and</strong> the next subsection.<br />

A background can be found <strong>in</strong> textbooks such as [8]. The oscillators were designed<br />

primarily as test vehicles, mak<strong>in</strong>g the circuit design more flexible <strong>and</strong> less conven-<br />

tional than typical commercial configurations. Design goals were to m<strong>in</strong>imize phase<br />

noise <strong>and</strong> maximize power <strong>and</strong> l<strong>in</strong>earity without regard to other constra<strong>in</strong>ts such as<br />

device size or ease of implementation. Design started with a basic cross-coupled pair<br />

of 2 × 100 µm-wide <strong>HEMTs</strong>, seen <strong>in</strong> figure 5.2. Typically the dra<strong>in</strong> bias would be set<br />

with a current mirror at the HEMT sources (marked with an S) but it was desired to<br />

have the freedom to change the bias. This also elim<strong>in</strong>ates the LFN from devices <strong>in</strong> the<br />

current mirror.<br />

The tank, represented by the capacitor Ct <strong>and</strong> <strong>in</strong>ductors Lt, primarily set the reso-<br />

nance frequency. A fixed capacitor was used <strong>in</strong>stead of a varactor to improve the Q of<br />

the resonator <strong>and</strong> allow better <strong>in</strong>sight of the phase noise performance of the <strong>HEMTs</strong>.<br />

The passive components used for the tank were based on measurements of fabricated<br />

capacitors <strong>and</strong> <strong>in</strong>ductors at 5 GHz, with the best Q components picked for the desired<br />

oscillation frequency. Typical unloaded Q values at 5 GHz are 25 <strong>and</strong> 65 for the square<br />

spiral <strong>in</strong>ductors <strong>and</strong> the SiN capacitors respectively.<br />

135


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

Load<br />

C 2<br />

L 1<br />

S<br />

L t<br />

C 1<br />

C 3<br />

L 2<br />

L 3<br />

C t<br />

L 3<br />

C 3<br />

L 2<br />

L t<br />

C 1<br />

S<br />

L 1<br />

C 2<br />

Load<br />

Figure 5.2: Circuit schematic of the oscillator (bias<strong>in</strong>g not shown).<br />

Lt<br />

L2<br />

L1<br />

C1<br />

C3<br />

C2<br />

Ct<br />

Figure 5.3: Photograph of the high l<strong>in</strong>earity oscillator. Darker areas are the two<br />

<strong>HEMTs</strong>. Passive components are labeled as <strong>in</strong> figure 5.2.<br />

136<br />

C2<br />

C3<br />

C1<br />

L1<br />

L2<br />

Lt


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

The output is taken on either side of the tank. To keep from load<strong>in</strong>g the tank, <strong>and</strong><br />

to match to a 50 Ω output, an L-match (L1 <strong>and</strong> C2) is used. Bias is provided at the<br />

gate <strong>and</strong> through the L-matches with the output. This required the use of off-chip<br />

bias-Ts for the RF/DC ports, but not for the gate bias port. The gate uses large square<br />

<strong>in</strong>ductors (L2) as RF chokes. C1 is a DC block that prevents the dra<strong>in</strong> of the <strong>HEMTs</strong><br />

from short<strong>in</strong>g, <strong>and</strong> C3 isolates the DC of the gate <strong>and</strong> dra<strong>in</strong>. The lengths of l<strong>in</strong>e used<br />

to cross-connect the <strong>HEMTs</strong> are modeled by the <strong>in</strong>ductors L3. The photograph of the<br />

oscillator <strong>in</strong> figure 5.3 labels most of the circuit elements found <strong>in</strong> figure 5.2. Circuit<br />

size is 2 x 1.65 mm 2 .<br />

The circuit was simulated us<strong>in</strong>g ADS. Transient <strong>and</strong> harmonic simulations were<br />

performed. Monitor<strong>in</strong>g these, along with the dynamic load-l<strong>in</strong>e at the dra<strong>in</strong> port of the<br />

<strong>HEMTs</strong>, the circuit was optimized to give as l<strong>in</strong>ear an output as possible.<br />

Measurements were performed on-wafer with air-coplanar (Cascade Microtech ACP-<br />

40 GSG) probes. Off-chip bias-Ts were used for both the dra<strong>in</strong>s <strong>and</strong> gates. One side<br />

of the oscillator was term<strong>in</strong>ated <strong>in</strong> a dummy 50 Ω load. All measurements that follow<br />

(power, l<strong>in</strong>earity, phase noise) were performed us<strong>in</strong>g an Agilent E4440 spectrum ana-<br />

lyzer with phase noise personality. Figure 5.4 (a) shows the measured spectrum from<br />

one side provid<strong>in</strong>g 22.9 dBm of power at a 4.166 GHz oscillation frequency. The cir-<br />

cuit is biased at Vds 20 V, Ids 233 mA, <strong>and</strong> Vgs -1 V. The analyzer had a 10 MHz span<br />

<strong>and</strong> 33 kHz resolution b<strong>and</strong>width for the measurement. How the power <strong>and</strong> efficiency<br />

137


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

(a) (b)<br />

Figure 5.4: Measurements of the oscillator: (a) power spectrum (b) frequency pull<strong>in</strong>g.<br />

(a) (b)<br />

Figure 5.5: Output power (s<strong>in</strong>gle-sided), second harmonic power, <strong>and</strong> efficiency (full<br />

circuit) of the high-l<strong>in</strong>earity oscillator for changes <strong>in</strong> device (a) dra<strong>in</strong>-source voltage<br />

<strong>and</strong> (b) gate-source voltage.<br />

138


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

compare to other oscillators will be discussed <strong>in</strong> § 5.5.<br />

Of <strong>in</strong>terest is the oscillator pull<strong>in</strong>g, which is the change <strong>in</strong> oscillation frequency<br />

with DC bias. This was measured for both the gate <strong>and</strong> dra<strong>in</strong> voltages <strong>and</strong> plotted <strong>in</strong><br />

figure 5.4 (b). It is typical to express the pull<strong>in</strong>g as a ratio of change <strong>in</strong> frequency to<br />

change <strong>in</strong> bias. Approximat<strong>in</strong>g the data as l<strong>in</strong>ear <strong>in</strong> figure 5.4 (b) gives a pull<strong>in</strong>g of<br />

4.3 MHz/V for the gate <strong>and</strong> 0.6 MHz/V for the dra<strong>in</strong>. These changes are less than<br />

0.4% of the frequency of oscillation.<br />

The power, 2nd harmonic, <strong>and</strong> efficiency for changes <strong>in</strong> bias appear <strong>in</strong> figure 5.5 (a)<br />

<strong>and</strong> (b). The third harmonic was so small as to be buried <strong>in</strong> the noise floor of the spec-<br />

trum analyzer, mak<strong>in</strong>g it ∼70 dB below the carrier. The 2nd harmonic was typically<br />

better than 30 dBc for all bias<strong>in</strong>gs measured. If the output were taken differentially,<br />

even harmonics would cancel <strong>and</strong> the oscillator would make for an extremely l<strong>in</strong>ear<br />

source.<br />

This oscillator was the first <strong>GaN</strong> differential oscillator to be reported <strong>in</strong> the litera-<br />

ture, <strong>and</strong> also the best reported l<strong>in</strong>earity for a <strong>GaN</strong> oscillator [2]. However, the phase<br />

noise was not impressive, be<strong>in</strong>g only -86.3 dBc <strong>and</strong> -115.7 dBc at best. A second,<br />

similar, oscillator was constructed <strong>and</strong> is now presented.<br />

139


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

5.4.2 Low-Phase <strong>Noise</strong> Oscillator<br />

The previous design was modified to improve phase noise performance. The Q<br />

of the various <strong>in</strong>ductors was reasoned to be a limit<strong>in</strong>g factor, <strong>and</strong> were replaced. Mi-<br />

crostrip l<strong>in</strong>es, Lt <strong>in</strong> figure 5.6, were used <strong>in</strong> place of the tank <strong>in</strong>ductors. The L-matches<br />

were changed to tapped capacitors to help preserve the loaded tank Q. This meant the<br />

dra<strong>in</strong> bias<strong>in</strong>g needed to be applied through the tank microstrip <strong>in</strong>ductors. To provide<br />

an AC ground across Lt, the large block<strong>in</strong>g capacitor 2C3 was added. The circuit,<br />

shown <strong>in</strong> figure 5.7, was slightly smaller than the previous at 2 x 1.4 mm 2 .<br />

Load<br />

C 1<br />

C 2<br />

S<br />

2C 3<br />

L t Lt<br />

C 4<br />

L 2<br />

L 1<br />

C t<br />

L 1<br />

C 4<br />

L 2<br />

S<br />

C 2<br />

C 1<br />

Load<br />

Figure 5.6: Circuit schematic of the low-phase noise oscillator (bias<strong>in</strong>g not shown).<br />

140


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

L 2<br />

L t<br />

C 1<br />

C 2<br />

C 3<br />

C 4<br />

C t<br />

Figure 5.7: Photograph of the low-phase noise oscillator. Passive components are<br />

labeled as <strong>in</strong> figure 5.6.<br />

Figure 5.8 is a typical measured phase noise spectrum. Above 10 kHz the oscillator<br />

drift obstructs the measurement. From 10 kHz to 1 MHz the spectrum decreases with<br />

a 30 dB per decade slope, similar to figure 5.1 (d). This means the low-frequency<br />

noise (LFN) of the oscillator dom<strong>in</strong>ates <strong>and</strong> hides the 20 dB per decade thermal noise<br />

region. So far, no published <strong>GaN</strong>-oscillator phase-noise measurements have shown<br />

a strict 20 dB per decade slope below a 1 MHz offset [2, 11–15]. A phase noise<br />

measurement above a 1 MHz offset usually approaches a spectrum analyzer’s noise<br />

floor, hence the lack of phase noise measurements <strong>in</strong> the literature at larger offsets.<br />

Now to answer the key question: does an <strong>in</strong>crease of oscillator power improve the<br />

phase noise? After the optimal bias for Vgs was determ<strong>in</strong>ed <strong>and</strong> set, Vds was varied.<br />

141<br />

C 4<br />

C 3<br />

L t<br />

C 1<br />

C 2<br />

L 2


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

Figure 5.8: Measured phase noise of the oscillator.<br />

Figure 5.9: Phase noise at 100 kHz <strong>and</strong> 1 MHz offsets versus dra<strong>in</strong>-source bias for a<br />

few oscillators.<br />

142


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

The power of the oscillators <strong>in</strong>creases from 15 dBm to 18 dBm as Vds <strong>in</strong>creased from<br />

6 V to 35 V. Below Vds 5 V the power drops quickly (it was typically 7 dBm at Vds<br />

4 V). Phase noise for devices with different Vds is plotted <strong>in</strong> figure 5.9. The phase<br />

noise does not improve with an <strong>in</strong>crease <strong>in</strong> signal power, but actually <strong>in</strong>creases. This<br />

agrees with measurements <strong>in</strong> [12]. Of <strong>in</strong>terest is that the general shape of figure 5.9 is<br />

similar to the measured LFN bias dependence <strong>in</strong> figure 4.6. While proper design can<br />

help to lower the offset frequency where the 30 dB/decade <strong>and</strong> 20 dB/decade slopes<br />

meet, only Hajimiri’s model [5] addresses how this can be accomplished <strong>and</strong> its ability<br />

to do this is still debated. With these results <strong>in</strong> m<strong>in</strong>d, we now compare <strong>GaN</strong> oscillators<br />

to other material systems for phase noise <strong>and</strong> for other oscillator measurements.<br />

5.5 Comparison to Other <strong>Oscillators</strong><br />

A summary of measurements of several oscillators is found <strong>in</strong> table 5.1. Listed are<br />

the measured frequency of oscillation (or range if the oscillator is tunable), the total<br />

device (or devices) width, the oscillator power at its resonance frequency <strong>in</strong> addition<br />

to second <strong>and</strong> third harmonics, the best efficiency of which the oscillator was capable,<br />

phase noise measurements at 100 kHz <strong>and</strong> 1 MHz offsets, <strong>and</strong> the reference for each<br />

work.<br />

The table is separated <strong>in</strong>to four parts depend<strong>in</strong>g on oscillator type. All oscillators <strong>in</strong><br />

the table are either HEMT or MESFET-based (no BJTs or HBTs) <strong>and</strong> <strong>in</strong>tegrated (no<br />

143


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

Desc. Carrier Device Fund. 2nd 3rd Best Phase <strong>Noise</strong> Ref.<br />

Freq. Width Power Harm. Harm. Eff. 100 kHz,1 MHz<br />

GHz mm dBm dBm dBm % dBc/Hz<br />

<strong>GaN</strong> HEMT <strong>Oscillators</strong><br />

Colpitts 5.3 0.2 20.5 -11.5 – 14.1 -105 -123 [12]<br />

Hartley 9.56 1.5 32.3 – – 16 -87 -115 [16]<br />

VCO 9 ± 0.5 1.5 31.8 10.8 4.8 21 -77 – [14]<br />

Colpitts,<br />

1.1 µm FP<br />

5.02 0.5 30 ∼8 – 24 -104 -132 [11]<br />

GaAs MESFET Oscillator<br />

GaAs<br />

VCO<br />

11.5±0.3 – 9 >20 – – -91 -119 [17]<br />

Differential <strong>Oscillators</strong> (Fixed Frequency), This Work<br />

§ 5.4.1 4.16 0.4 22.9 -19.2


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

<strong>in</strong> its circuit <strong>and</strong> [11] used a FP on the devices <strong>in</strong> the oscillator, provid<strong>in</strong>g astonish-<br />

<strong>in</strong>g improvements <strong>in</strong> phase noise (∼10 dB over the same circuit us<strong>in</strong>g devices with<br />

no FP). The oscillators <strong>in</strong> Si <strong>and</strong> GaAs technologies all used varactors for frequency<br />

tun<strong>in</strong>g. This will degrade the tank Q <strong>and</strong> lead to poorer phase noise performance. As<br />

a safe assumption, assume that the phase noise would be -10 dBc/Hz lower (better)<br />

than what is stated <strong>in</strong> the table. With this adjustment, we see that <strong>GaN</strong> has better or<br />

comparable phase noise at a 1 MHz offset, but by offsets ≤ 100 kHz the phase noise<br />

is worse. Work by Rice [15] agrees with these measurements. The reason for this<br />

difference is because <strong>GaN</strong> oscillators are dom<strong>in</strong>ated by their LFN <strong>and</strong> never show the<br />

20 dB/decade slope from thermal noise. This is better-expla<strong>in</strong>ed by figure 5.10. Here<br />

the phase noise is displayed comparatively for a <strong>GaN</strong> oscillator <strong>and</strong> a typical low-noise<br />

oscillator <strong>in</strong> GaAs or Si. Note that the magnitudes of the phase noise are not the same,<br />

as phase noise is expressed relative to the carrier. The measured absolute power of the<br />

phase noise sideb<strong>and</strong>s for the <strong>GaN</strong> oscillator will be much larger than <strong>in</strong> GaAs or Si.<br />

<strong>GaN</strong> oscillators provide at least an order of magnitude more power than the other<br />

technologies. For the total device width of the <strong>HEMTs</strong>, the oscillators <strong>in</strong> this work<br />

provide an average amount of power compared to the other <strong>GaN</strong> oscillators. The<br />

st<strong>and</strong>out is the FP HEMT oscillator, with 2 W/mm. The l<strong>in</strong>earity of the first oscillator<br />

<strong>in</strong> this work is second to none. The efficiencies of the <strong>GaN</strong> oscillators appears to be<br />

much higher than GaAs <strong>and</strong> Si.<br />

145


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

Figure 5.10: Relative comparison of <strong>GaN</strong> oscillator to a typical oscillator with low<br />

phase noise (figure courtesy of Dr. Robert York).<br />

5.6 Summary<br />

Two <strong>GaN</strong> HEMT-based differential oscillators were presented. The first displayed<br />

very good harmonic suppression but poor phase noise. This oscillator was also the first<br />

<strong>GaN</strong> differential oscillator to appear <strong>in</strong> the literature [2]. The second oscillator showed<br />

excellent phase noise performance at a 1 MHz offset of -132 dBc/Hz. However, be-<br />

cause <strong>GaN</strong> oscillator phase noise appears to be dom<strong>in</strong>ated by very large amounts of<br />

LFN, the noise performance is worse at offsets closer to the carrier. The hoped-for<br />

benefit of better phase noise with more oscillator power is shattered by measurements<br />

that show that more power actually <strong>in</strong>creases the phase noise. This could be expected<br />

because of the HEMT dra<strong>in</strong> LFN bias dependence with Vds. Because of these two<br />

146


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

problems, <strong>GaN</strong> does not appear to have an advantage over GaAs or Si oscillators for<br />

phase noise.<br />

References<br />

[1] H. Xu, “MMICs us<strong>in</strong>g <strong>GaN</strong> <strong>HEMTs</strong> <strong>and</strong> Th<strong>in</strong>-Film BST Capacitors,” Ph.D. dissertation,<br />

University of California, Santa Barbara, 2005.<br />

[2] C. Sanabria, H. Xu, S. Heikman, U. Mishra, <strong>and</strong> R. York, “A <strong>GaN</strong> Differential<br />

Oscillator With Improved Harmonic Performance,” IEEE <strong>Microwave</strong> Components<br />

Lett., vol. 15, pp. 463–465, Jul. 2005.<br />

[3] A. N. Riddle, “Oscillator <strong>Noise</strong>: Theory <strong>and</strong> Characterization,” Ph.D. dissertation,<br />

North Carol<strong>in</strong>a State University, 1986.<br />

[4] W. P. Rob<strong>in</strong>s, Phase <strong>Noise</strong> <strong>in</strong> Signal Sources. London, UK.: Peter Peregr<strong>in</strong>us<br />

Ltd., 1982.<br />

[5] A. Hajimiri <strong>and</strong> T. Lee, The Design of Low <strong>Noise</strong> <strong>Oscillators</strong>. Boston: Kluwer<br />

Academic Publishers, 1999.<br />

[6] S. A. Maas, <strong>Noise</strong> <strong>in</strong> L<strong>in</strong>ear <strong>and</strong> Nonl<strong>in</strong>ear Circuits. Boston: Artech House, 2005.<br />

[7] M. J. Buck<strong>in</strong>gham, <strong>Noise</strong> <strong>in</strong> Electronic Devices <strong>and</strong> Systems. New York: John<br />

Wiley & Sons, 1983.<br />

[8] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, 2nd ed.<br />

New York: Cambridge University Pess, 2004.<br />

[9] D. Leeson, “A Simple Model of Feedback Oscillator <strong>Noise</strong> Spectrum,” Proc. IEEE,<br />

vol. 54, no. 2, pp. 329–330, 1966.<br />

[10] V. Manan <strong>and</strong> S. Long, “A Low Power <strong>and</strong> Low <strong>Noise</strong> p-HEMT Ku B<strong>and</strong> VCO,”<br />

<strong>Microwave</strong> <strong>and</strong> Wireless Components Letters, IEEE [see also IEEE <strong>Microwave</strong><br />

<strong>and</strong> Guided Wave Letters], vol. 16, no. 3, pp. 131–133, 2006.<br />

[11] H. Xu, C. Sanabria, S. Heikman, S. Keller, U. Mishra, <strong>and</strong> R. York, “High Power<br />

<strong>GaN</strong> <strong>Oscillators</strong> Us<strong>in</strong>g Field-Plated HEMT Structure,” <strong>in</strong> <strong>Microwave</strong> Symposium<br />

Digest, 2005 IEEE MTT-S International, 2005, pp. 1345–1348.<br />

[12] H. Xu, C. Sanabria, A. Ch<strong>in</strong>i, S. Keller, U. Mishra, <strong>and</strong> R. A. York, “Low Phase-<br />

<strong>Noise</strong> 5 GHz <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT Oscillator Integrated with BaxSr1−xTiO3 Th<strong>in</strong><br />

Films,” IEEE <strong>Microwave</strong> Theory <strong>and</strong> Tech. Symp., pp. 1509–1512, 2004.<br />

147


CHAPTER 5. GAN HEMT BASED OSCILLATORS<br />

[13] J. B. Shealy, J. A. Smart, <strong>and</strong> J. R. Shealy, “Low-Phase <strong>Noise</strong> <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> FET-<br />

Based Voltage Controlled <strong>Oscillators</strong> (VCOs),” IEEE <strong>Microwave</strong> Components<br />

Lett., vol. 11, no. 6, pp. 244–245, Jun. 2001.<br />

[14] V. Kaper, R. Thompson, T. Prunty, <strong>and</strong> J. R. Shealy, “Signal Generation, Control<br />

<strong>and</strong> Frequency Conversion <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT MMICs,” IEEE <strong>Microwave</strong> Theory<br />

<strong>and</strong> Tech. Symp., pp. 1145–1148, 2004.<br />

[15] P. Rice, R. Sloan, M. Moore, A. R. Barnes, M. J. Uren, N. Malbert, <strong>and</strong> N. Labat,<br />

“A 10 GHz Dielectric Resonator Oscillator Us<strong>in</strong>g <strong>GaN</strong> Technology,” IEEE<br />

<strong>Microwave</strong> Theory <strong>and</strong> Tech. Symp., pp. 1497–1500, 2004.<br />

[16] V. S. Kaper, V. Tilak, H. Kim, A. V. Vertiatchikh, R. M. Thompson, T. R. Prunty,<br />

L. Eastman, <strong>and</strong> J. R. Shealy, “High-Power Monolithic <strong>Al<strong>GaN</strong></strong>/<strong>GaN</strong> HEMT Oscillator,”<br />

IEEE J. Solid-State Circuits, vol. 38, no. 9, pp. 1457–1461, Sept. 2003.<br />

[17] C.-H. Lee, S. Han, B. Mat<strong>in</strong>pour, <strong>and</strong> J. Laskar, “A Low Phase <strong>Noise</strong> X-b<strong>and</strong><br />

MMIC GaAs MESFET VCO,” <strong>Microwave</strong> <strong>and</strong> Guided Wave Letters, IEEE [see<br />

also IEEE <strong>Microwave</strong> <strong>and</strong> Wireless Components Letters], vol. 10, no. 8, pp. 325–<br />

327, 2000.<br />

[18] P. K<strong>in</strong>get, “A Fully Integrated 2.7 V 0.35 m CMOS VCO for 5 GHz Wireless<br />

Applications,” ISSCC Dig. Tech. Papers, p. 226227, 1998.<br />

[19] M. Soyuer, K. Jenk<strong>in</strong>s, J. Burghartz, <strong>and</strong> M. Hulvey, “A 3-V 4-GHz nMOS<br />

Voltage-Controlled Oscillator with Integrated Resonator,” Solid-State Circuits,<br />

IEEE Journal of, vol. 31, no. 12, pp. 2042–2045, 1996.<br />

[20] S.-W. Yoon, E.-C. Park, C.-H. Lee, S. Sim, S.-G. Lee, E. Yoon, J. Laskar, <strong>and</strong><br />

S. Hong, “5 6 GHz-B<strong>and</strong> GaAs MESFET-Based Cross-Coupled Differential Oscillator<br />

MMICs With Low Phase-<strong>Noise</strong> Performance,” IEEE <strong>Microwave</strong> Components<br />

Lett., vol. 11, no. 12, pp. 495–497, Dec. 2001.<br />

148


6<br />

Summary, Conclusions, <strong>and</strong> Future<br />

Directions For <strong>Noise</strong> Studies<br />

6.1 Summary <strong>and</strong> Conclusions<br />

THIS dissertation has looked at several aspects of the noise performance of<br />

<strong>GaN</strong> <strong>HEMTs</strong>: noise figure, low-frequency noise, <strong>and</strong> phase noise. The noise<br />

figure of <strong>GaN</strong> <strong>HEMTs</strong> is comparable to GaAs <strong>HEMTs</strong>. However, the gate leakage<br />

needs to be well-controlled. Source resistance might also be too large a contributor for<br />

small devices. At high bias<strong>in</strong>gs, self-heat<strong>in</strong>g causes the ga<strong>in</strong> to drop, <strong>and</strong> an <strong>in</strong>crease<br />

<strong>in</strong> source resistance quickly degrades the noise figure performance.<br />

A simple model, that <strong>in</strong>cluded device scal<strong>in</strong>g, was put together that not only pre-<br />

dicts the m<strong>in</strong>imum noise figure well, but also the other noise parameters: |Γopt|,� Γopt,<br />

<strong>and</strong> rn. It is useful for h<strong>and</strong> <strong>and</strong> Matlab calculations to predict <strong>and</strong> underst<strong>and</strong> noise<br />

without the need for noise parameter measurements beforeh<strong>and</strong>. It was successfully<br />

used to underst<strong>and</strong> how gate leakage <strong>and</strong> a field plate <strong>in</strong>fluence noise figure. The<br />

model does not work when the ga<strong>in</strong> of the transistor drops, such as at high bias<strong>in</strong>gs<br />

149


CHAPTER 6. SUMMARY, CONCLUSIONS, AND FUTURE DIRECTIONS<br />

or frequencies close to fτ. It was shown that the Pospieszalski <strong>and</strong> correlated noise<br />

models can be successfully applied to <strong>GaN</strong> <strong>HEMTs</strong>, even for noise versus bias.<br />

Summariz<strong>in</strong>g the other noise figure studies, devices on either SiC or sapphire sub-<br />

strates can have the same m<strong>in</strong>imum noise figure as long as the bias is low. Thick-cap<br />

devices show low noise at high bias<strong>in</strong>gs. However, their noise at lower bias<strong>in</strong>gs is<br />

sub-par because of a very large amount of gate leakage.<br />

A low-frequency noise setup that works at bias<strong>in</strong>gs typically needed for <strong>GaN</strong> <strong>HEMTs</strong><br />

was constructed <strong>and</strong> thoroughly expla<strong>in</strong>ed. Unlike many other setups, this allowed<br />

bias dependent studies well <strong>in</strong>to the device saturation region. From this, a strong Vds<br />

dependence for the dra<strong>in</strong> was discovered <strong>and</strong> a scalable, bias dependent, model that<br />

could be entered <strong>in</strong> a circuit simulator was <strong>in</strong>troduced. The gate was found to follow<br />

the low-frequency noise model<strong>in</strong>g for a diode, as could be expected for a Schottky<br />

contact.<br />

Low-frequency noise studies were performed. Passivation was found to improve the<br />

low-frequency noise by an order of magnitude. An unpassivated thick-cap device has<br />

the same low-frequency noise performance as a passivated st<strong>and</strong>ard HEMT. A field-<br />

plate does not appear to help LFN, unlike noise figure. <strong>GaN</strong> HEMT low-frequency<br />

noise is worse than that of GaAs <strong>HEMTs</strong> because the slope of the noise is much larger.<br />

The phase noise of two <strong>GaN</strong> HEMT-based differential oscillators were explored.<br />

The first had poor phase noise, but very good harmonic suppression, <strong>and</strong> was the<br />

150


CHAPTER 6. SUMMARY, CONCLUSIONS, AND FUTURE DIRECTIONS<br />

first <strong>GaN</strong> HEMT differential oscillator <strong>in</strong> the literature. The second oscillator showed<br />

excellent phase noise performance at large offsets, but poor performance at closer<br />

offsets because of the phase noise be<strong>in</strong>g dom<strong>in</strong>ated by the low-frequency noise. The<br />

<strong>GaN</strong> power advantage that was hoped would improve the oscillator signal-to-noise,<br />

<strong>and</strong> thus phase noise, was more than offset by the <strong>in</strong>crease of low-frequency noise<br />

with power. Because of these problems with low-frequency noise, <strong>GaN</strong> oscillators do<br />

not perform better - they do not even meet the performance typically seen by GaAs<br />

<strong>and</strong> Si oscillators.<br />

6.2 Future Paths<br />

Try<strong>in</strong>g to make progress <strong>in</strong> an entire field of study which is not completely under-<br />

stood <strong>and</strong> mak<strong>in</strong>g three difficult-to-use measurement systems work allows room for<br />

improvement. To improve noise figure, gate leakage needs to be controlled, but first<br />

it needs to be understood. This is no doubt the pursuit of many champions of <strong>GaN</strong><br />

research because of its implications for reliability <strong>and</strong> power performance. It would<br />

also be of <strong>in</strong>terest to look at bias dependence of ion-implanted <strong>GaN</strong> <strong>HEMTs</strong> to see<br />

if it improves the curve typically seen for m<strong>in</strong>imum noise figure versus dra<strong>in</strong>-source<br />

current. As thick-cap <strong>HEMTs</strong> do not need passivation, they are an <strong>in</strong>terest<strong>in</strong>g device<br />

with possible <strong>in</strong>dustry applications. Their noise figure performance at large current<br />

bias<strong>in</strong>gs was better than any other measured by the author. A better underst<strong>and</strong><strong>in</strong>g<br />

151


CHAPTER 6. SUMMARY, CONCLUSIONS, AND FUTURE DIRECTIONS<br />

could be pursued.<br />

The work on low-frequency noise studies was only started <strong>and</strong> could be greatly<br />

exp<strong>and</strong>ed upon. In particular, while few researchers have even reported the dra<strong>in</strong> low-<br />

frequency noise Vds dependence, none have attempted to expla<strong>in</strong> it. As this seems to<br />

be directly l<strong>in</strong>ked to the measured phase noise of oscillators, it is strongly suggested<br />

it be studied <strong>in</strong> depth.<br />

There are many ways to try to improve phase noise. The oscillators presented here<br />

were relatively simple. More-sophisticated circuits should be undertaken <strong>and</strong> with<br />

smaller devices than those presented here or <strong>in</strong> other works. The researchers whose<br />

work has been discussed were design<strong>in</strong>g with high power <strong>in</strong> m<strong>in</strong>d. Before these cir-<br />

cuits can be attempted, the circuit model needs to be improved. There are several<br />

shortcom<strong>in</strong>gs. First, the ADS circuit model predicts that m<strong>in</strong>imum noise figure ver-<br />

sus dra<strong>in</strong> source current is flat, which is a glar<strong>in</strong>g problem (see figure 3.7). Dynamic<br />

source <strong>and</strong> dra<strong>in</strong> resistances must be <strong>in</strong>cluded along with self-heat<strong>in</strong>g effects. Gate<br />

leakage is not <strong>in</strong>cluded <strong>in</strong> the model <strong>and</strong> should be added. The attempt at us<strong>in</strong>g<br />

an ADS symbolically-def<strong>in</strong>ed device to generate a voltage-controlled low-frequency<br />

noise current source was not successful, as the simulator appeared to use the DC bias<br />

<strong>in</strong>stead of the vary<strong>in</strong>g large-signal voltage of the oscillation. A custom ADS compo-<br />

nent might be required, which could <strong>in</strong>volve a few hundred l<strong>in</strong>es of code. With these<br />

deficiencies corrected, it might be possible to more closely simulate phase noise <strong>and</strong><br />

152


CHAPTER 6. SUMMARY, CONCLUSIONS, AND FUTURE DIRECTIONS<br />

possibly better underst<strong>and</strong> why a field plate helps the phase noise performance.<br />

153


An ADS project that conta<strong>in</strong>s files used <strong>in</strong> this work can be downloaded at<br />

http://my.ece.ucsb.edu/sanabria/research/ADS<strong>Noise</strong>Template.zap<br />

A<br />

ADS Files<br />

or by contact<strong>in</strong>g the author at sanabria@ece.ucsb.edu. The downloaded file will need<br />

to be unarchived <strong>in</strong> ADS. The project conta<strong>in</strong>s files necessary to:<br />

1. Extract <strong>and</strong> optimize the small-signal parameters, as well as to simulate a smallsignal<br />

model of the transistor for S-parameters<br />

2. Simulate noise figure with either the correlated noise or Pospieszalski model<br />

3. Extract the noise variables for the correlated noise model<br />

Steps expla<strong>in</strong><strong>in</strong>g how to do each of these functions with the project are presented <strong>in</strong><br />

the next three sections. Example files for each are provided <strong>in</strong> the project.<br />

154


APPENDIX A. ADS FILES<br />

A.1 Small-Signal Extraction<br />

To extract the small-signal model, three S-parameter measurements are needed:<br />

those of the device, a shorted-pads version of the device, <strong>and</strong> an open channel version<br />

of the device. For more details, refer to § 2.3. It may also be useful to measure<br />

extr<strong>in</strong>sic parasitic resistances through another means (such as Hall measurements).<br />

The S-parameter measurements will then need to be imported <strong>in</strong>to <strong>in</strong>dividual ADS<br />

data sets.<br />

Once the data is correctly <strong>in</strong> the ADS project, open the data display file<br />

template small signal parameter extraction.dds<br />

The middle of the first page of the dds file should look like figure A.1. This is where<br />

the S-parameters for the device, open, <strong>and</strong> short are entered. Example data sets are<br />

currently entered (Hongtao open, Hongtao short, d021218FC1...). Once these are en-<br />

tered, turn to the next page <strong>in</strong> the data display (its shortcut key comb<strong>in</strong>ation is Alt,<br />

p, e). Plots for the small-signal extr<strong>in</strong>sic capacitances are displayed, along with two<br />

markers per graph. The plots should be relatively flat, except maybe at very low fre-<br />

quencies (less than 1 GHz). Move the markers to a suitable range of values for each<br />

parameter. An average will be calculated for these (<strong>and</strong> the rest of the small-signal<br />

parameters). Cont<strong>in</strong>ue through the next two pages for the parasitic resistance <strong>and</strong> <strong>in</strong>-<br />

ductances. Then advance two pages to the <strong>in</strong>tr<strong>in</strong>sic parameters graphs page. Here<br />

there may be larger variations <strong>in</strong> the parameters. Choose values that make sense, <strong>and</strong><br />

155


APPENDIX A. ADS FILES<br />

Figure A.1: First page of template small signal parameter extraction.dds.<br />

narrow the range close to the target frequency. Advanc<strong>in</strong>g to the next page shows the<br />

calculated small-signal parameters. These can be entered <strong>in</strong>to a small-signal model of<br />

the transistor. Two have already been provided <strong>in</strong> the project:<br />

template HEMT smallsignal <strong>and</strong> Pospieszalski<strong>Noise</strong> model.dsn<br />

156


APPENDIX A. ADS FILES<br />

template HEMT smallsignal <strong>and</strong> Pucel<strong>Noise</strong> model.dsn<br />

Simply substitute your values <strong>in</strong>to one of these <strong>and</strong> save it as a new file. Once the<br />

small-signal parameters have been entered, the sub-circuit conta<strong>in</strong><strong>in</strong>g the network can<br />

be simulated for S-parameters. Open the schematic file<br />

template Sparameters extraction optimization.dsn<br />

Here there is one of the small-signal files <strong>in</strong> a subnetwork, an S-parameter simulation<br />

block, optimization blocks, <strong>and</strong> a Data Access Component for identify<strong>in</strong>g an ADS<br />

data set. This is shown <strong>in</strong> figure A.2. Leave the deactivated components out of the<br />

simulation. Exchange the subnetwork with the circuit to be simulated, <strong>and</strong> simulate.<br />

This will call the previous data display w<strong>in</strong>dow. If ADS asks to change the data set,<br />

answer “yes.” Go to the verification page. Here, S-parameters for the simulation are<br />

compared to the data set used for extraction. The fit should be good, but not excellent.<br />

The parameters can be optimized to the data. In the schematic w<strong>in</strong>dow, enter the<br />

ADS data set <strong>in</strong>to the Data Access Component that was used for the extraction. Ac-<br />

tivate all components <strong>in</strong> the schematic, then simulate. The simulation w<strong>in</strong>dow will<br />

display the f<strong>in</strong>al optimized small-signal parameters. The verification page of the data<br />

display w<strong>in</strong>dow will show S-parameters for both the <strong>in</strong>itial <strong>and</strong> optimized values. En-<br />

ter the optimized values <strong>in</strong> the circuit sub-network. Deactivate the optimization blocks<br />

aga<strong>in</strong>, <strong>and</strong> simulate. The agreement on the verification page of the data display w<strong>in</strong>-<br />

dow of the simulated <strong>and</strong> measured S-parameters should be excellent.<br />

157


APPENDIX A. ADS FILES<br />

Figure A.2: Schematic used for simulat<strong>in</strong>g S-parameters of the small-signal circuit<br />

<strong>and</strong> for optimization.<br />

A.2 <strong>Noise</strong> Figure Simulation<br />

Open the project, <strong>and</strong> then open the circuit schematic file<br />

template noise figure measure.dns<br />

You should now be at the w<strong>in</strong>dow shown <strong>in</strong> figure A.3. The subnetwork<br />

template HEMT smallsignal <strong>and</strong> Pucel<strong>Noise</strong> model<br />

conta<strong>in</strong>s a small-signal model for a HEMT that <strong>in</strong>cludes noise sources <strong>and</strong> noise vari-<br />

158


APPENDIX A. ADS FILES<br />

ables for a Correlated <strong>Noise</strong> (CN) model. How to obta<strong>in</strong> the noise variables from<br />

noise parameter measurements <strong>and</strong> small-signal parameter values is discussed <strong>in</strong> the<br />

next section. This subnetwork can also be substituted with<br />

template HEMT smallsignal <strong>and</strong> Pospieszalski<strong>Noise</strong> model.dsn<br />

which conta<strong>in</strong>s a small-signal model of a HEMT <strong>and</strong> noise sources <strong>and</strong> noise vari-<br />

ables for the Pospieszalski model. The Pospieszalski noise variables are obta<strong>in</strong>ed from<br />

noise parameter measurements <strong>and</strong> small-signal parameters with the Matlab code <strong>in</strong><br />

Figure A.3: Schematic used for simulat<strong>in</strong>g noise parameters.<br />

159


APPENDIX A. ADS FILES<br />

appendix C.<br />

Return<strong>in</strong>g to the circuit schematic, simulat<strong>in</strong>g br<strong>in</strong>gs up the data display<br />

template noise figure viewer.dds<br />

Page 3 of this data display (use the shortcut Alt, p, e to navigate through the pages)<br />

conta<strong>in</strong>s noise <strong>and</strong> ga<strong>in</strong> circles on Smith Charts, the noise parameters, stability factor,<br />

<strong>and</strong> noise figure for a 50 Ω term<strong>in</strong>ation compared to NFm<strong>in</strong> versus frequency. The<br />

other pages of the display also conta<strong>in</strong> useful <strong>in</strong>formation for design.<br />

A.3 Correlated <strong>Noise</strong> Model Extraction<br />

To extract the noise variables for this type of noise model, the small-signal param-<br />

eters need to be determ<strong>in</strong>ed <strong>and</strong> entered <strong>in</strong>to a circuit as expla<strong>in</strong>ed <strong>in</strong> the first section<br />

of this appendix. In particular, have all the small-signal parameters entered <strong>in</strong>to the<br />

variable (VAR) block of the circuit network<br />

template HEMT smallsignal <strong>and</strong> Pucel<strong>Noise</strong> model.dsn<br />

Measured noise parameters are also needed to determ<strong>in</strong>e the noise variables. Once<br />

setup, open the circuit schematic<br />

template parasitic matrix for correlation extraction.dsn<br />

which is shown <strong>in</strong> figure A.4. Here, there are two networks simulated for S-parameters.<br />

They correspond to the <strong>in</strong>tr<strong>in</strong>sic <strong>and</strong> extr<strong>in</strong>sic parameters, respectively. If the small-<br />

signal network listed above was used, its VAR block can just be pasted <strong>in</strong>to this design<br />

160


APPENDIX A. ADS FILES<br />

<strong>and</strong> the previous one deleted.<br />

Simulat<strong>in</strong>g should br<strong>in</strong>g up the data display<br />

template noise correlation extraction.dds<br />

A screen shot of it is shown <strong>in</strong> figure A.5. The noise parameters will need to be entered<br />

<strong>in</strong>to the equations <strong>in</strong>side the Change These box. Once this is done, the Cpd unnorm<br />

box <strong>and</strong> Cext will be the desired results: Cpd unnorm(1,1) is the gate noise <strong>in</strong> units<br />

of A 2 /Hz, Cpd unnorm(2,2) is the dra<strong>in</strong> noise <strong>in</strong> units of A 2 /Hz, Cpd unnorm(1,2)<br />

<strong>and</strong> Cpd unnorm(2,1) are the cross-correlation terms (not used here), <strong>and</strong> Cext is the<br />

correlation between the gate <strong>and</strong> dra<strong>in</strong> noise. Enter<strong>in</strong>g these back <strong>in</strong>to the second VAR<br />

block <strong>in</strong><br />

template HEMT smallsignal <strong>and</strong> Pucel<strong>Noise</strong> model.dsn<br />

will allow for simulation of the noise parameters.<br />

161


APPENDIX A. ADS FILES<br />

Figure A.4: Schematic used for extract<strong>in</strong>g correlated noise model noise variables.<br />

Figure A.5: Data display used for extract<strong>in</strong>g correlated noise model noise variables.<br />

162


B<br />

Matlab Code for <strong>Noise</strong><br />

Parameter Model<strong>in</strong>g<br />

THIS Matlab script computes all four of the noise parameters as developed <strong>in</strong> § 2.6,<br />

displays them <strong>in</strong> different formats, <strong>and</strong> plots m<strong>in</strong>imum noise figure versus frequency.<br />

This should facilitate the model’s use for other researchers. A copy of the file<br />

can be obta<strong>in</strong>ed by contact<strong>in</strong>g the author at sanabria@ece.ucsb.edu.<br />

% NF.m<br />

% NF <strong>and</strong> other noise parameters <strong>in</strong>clud<strong>in</strong>g gate leakage, but not Cgd.<br />

% Chris Sanabria, 12/03/04<br />

clear all;<br />

close all;<br />

% Variables (change)<br />

163


APPENDIX B. MATLAB CODE FOR NOISE PARAMETER MODELING<br />

Igs = 6e-6; % Gate leakage<br />

rge = 3.03; % Gate resistance<br />

rs = 6;<br />

gmi = 0.033;<br />

ri = 8;<br />

cgsi = 0.21e-12;<br />

rds = 1000; % Dra<strong>in</strong>-source resistance. Only used for a ga<strong>in</strong> calculation<br />

% <strong>and</strong> not for noise prediciton.<br />

G = 2/3; % Gamma, use 2/3<br />

f = [5e9 10e9]; % Operat<strong>in</strong>g frequency or frequencies<br />

Zo = 50; % Reference impedance for calculat<strong>in</strong>g reflection coefficient<br />

Ta = 290; % Kelv<strong>in</strong>. This is the <strong>in</strong>put temp., NOT the channel temp.!<br />

% Constants<br />

q = 1.6022e-019;<br />

k = 1.3807e-023;<br />

w = 2*pi*f;<br />

% Intermediates that may change<br />

r<strong>in</strong> = rge + ri + rs;<br />

gm = gmi./(1+gmi.*rs);<br />

cgs = cgsi./(1+gmi.*rs);<br />

% Intermediate variables (no chang<strong>in</strong>g)<br />

wt = gm./cgs; % That’s right, no 2pi.<br />

a = gm.*G.*(w./wt).ˆ2;<br />

b = q.*Igs./(2.*k.*Ta);<br />

xgs = -1./(2.*pi.*f.*cgs);<br />

% Optimum <strong>in</strong>put impedance (reflection coefficient) for noise<br />

Xopt = 1./(w.*cgs).*a./(a+b);<br />

Ropt = sqrt(r<strong>in</strong>./(a+b) + a./(a+b).*r<strong>in</strong>.ˆ2 + a.*b./((a+b).ˆ2.*(w.ˆ2.*cgs.ˆ2)));<br />

Zopt = Ropt + j.*Xopt;<br />

Gamopt = (Zopt-Zo)./(Zopt+Zo);<br />

% M<strong>in</strong>imum <strong>Noise</strong> Figure<br />

164


APPENDIX B. MATLAB CODE FOR NOISE PARAMETER MODELING<br />

rg = Ropt;<br />

xg = Xopt;<br />

Fm<strong>in</strong> = 1 + r<strong>in</strong>./rg + b./rg.*(rg.ˆ2+xg.ˆ2) + a./rg.*(abs(r<strong>in</strong>+rg+j.*(xg - 1./(w.*cgs)))).ˆ2;<br />

% <strong>Noise</strong> resistance<br />

% First calculate NF at Zo, then use NFm<strong>in</strong> <strong>and</strong> NF @ Zo to get rn<br />

rgzo = real(Zo);<br />

xgzo = imag(Zo);<br />

Fzo = 1 + r<strong>in</strong>./rgzo + b./rgzo.*(rgzo.ˆ2 + xgzo.ˆ2) + a./rgzo.*(abs(r<strong>in</strong>+rgzo+j.*(xgzo -<br />

1./(w.*cgs)))).ˆ2;<br />

Rn = Zo.*(Fzo - Fm<strong>in</strong>).*(abs(1 + Gamopt)).ˆ2./(4.*abs(Gamopt)); % noise resistance<br />

(not normalized)<br />

% Ga<strong>in</strong> av @ NFm<strong>in</strong>. Will not be accurate because Cgd is not <strong>in</strong>cluded<br />

Gav = abs(-(rds./2).*(wt./w).*j./((rg+r<strong>in</strong>) + j.*(xgs + xg))).ˆ2;<br />

% Display outputs<br />

disp([’Fm<strong>in</strong> = ’ num2str(Fm<strong>in</strong>)])<br />

disp(’ ’)<br />

NFm<strong>in</strong> = 10*log10(Fm<strong>in</strong>);<br />

disp([’NFm<strong>in</strong> (dB) = ’ num2str(NFm<strong>in</strong>)])<br />

% disp([’|Zopt| = ’ num2str(abs(Zopt))])<br />

% disp([’|Zopt| = ’ num2str(abs(Zopt))])<br />

% disp(’ ’)<br />

% disp([’phase Zopt = ’ num2str(angle(Zopt)/pi*180)])<br />

% disp(’ ’)<br />

disp([’r n = ’ num2str(Rn./Zo)])<br />

disp([’|Gamma opt| = ’ num2str(abs(Gamopt))])<br />

disp([’phase Gamma opt = ’ num2str(angle(Gamopt)/pi*180)])<br />

disp([’Ga<strong>in</strong> av [dB] = ’ num2str(10.*log10(Gav))])<br />

% Some plott<strong>in</strong>g<br />

figure;<br />

plot(f./1e9,NFm<strong>in</strong>,’r’)<br />

xlabel(’Frequency [GHz]’)<br />

ylabel(’NF {m<strong>in</strong>} [dB]’)<br />

165


C<br />

Matlab Code for Pospieszalski <strong>Noise</strong><br />

Parameter Model<strong>in</strong>g<br />

THIS Matlab script calculates the two noise temperatures for the Pospieszalski<br />

noise model as discussed <strong>in</strong> § 2.5.3. Because this <strong>in</strong>volves solv<strong>in</strong>g a pair of<br />

quadratic equations, it returns two sets of noise temperatures. The temperatures that<br />

are negative <strong>in</strong> value are non-physical <strong>and</strong> should not be used.<br />

% noise temps.m<br />

% Uses Pospieszalski’s model to get the two noise temperatures. You need to<br />

% have taken noise figure measurements <strong>and</strong> have extracted a small signal<br />

% model. Because the solv<strong>in</strong>g is for a pair of quadratic equations, there<br />

% are two sets of solutions. Ignore the negative temperature pair.<br />

% Chris Sanabria, March 2003<br />

clc;<br />

format compact<br />

%Input these:<br />

freq = 10e9; % Hz<br />

NFm<strong>in</strong> = 2.03; % dB<br />

rn = .755; % NOT ohms, the normalized value<br />

gam opt= 0.609*exp(j*54.4/180*pi); % Complex Reflection Coefficient<br />

ft = 22.1e9; % In Hz. Pospieszalski def<strong>in</strong>es this as gm/(Cgs*2*pi), but the<br />

% author f<strong>in</strong>ds that the measured ft can work better.<br />

Cgs = .258e-12; % Farads<br />

rgs = 7.46; % Same as Ri<br />

166


APPENDIX C. MATLAB CODE FOR POSPIESZALSKI NOISE PARAMETER MODELING<br />

rds = 1155; % Ohms<br />

gm = .032; %S<br />

To = 290; % Kelv<strong>in</strong><br />

% Intermediate Equations<br />

gds = 1/rds;<br />

Rn = rn*50;<br />

gn = 1/Rn;<br />

Zopt = (1+gam opt)/(1-gam opt)*50;<br />

Xopt = imag(Zopt);<br />

Ropt = real(Zopt);<br />

Tm<strong>in</strong> = (10ˆ(NFm<strong>in</strong>/10)-1)*To;<br />

N = Ropt/Rn;<br />

% Solve the equations<br />

[Td Tg] = solve(’Roptˆ2 = (ft/freq)ˆ2*rgs/gds*Tg/Td+rgsˆ2’,...<br />

’Tm<strong>in</strong> = 2*freq/ft*sqrt(gds*rgs*Tg*Td+(freq/ft)ˆ2*rgsˆ2*gds...<br />

ˆ2*Tdˆ2) + 2*(freq/ft)ˆ2*rgs*gds*Td’,’Td’,’Tg’);<br />

% Values are currently <strong>in</strong> Kelv<strong>in</strong><br />

disp(’Values <strong>in</strong> Kelv<strong>in</strong>’);<br />

Td = eval(Td)<br />

Tg = eval(Tg)<br />

disp(’ ’);<br />

disp(’ ’);<br />

% Values are now <strong>in</strong> degrees C<br />

disp(’Values <strong>in</strong> Celcius’);<br />

Td C = Td - 273.15<br />

Tg C = Tg - 273.15<br />

% Check the validty<br />

disp(’ ’)<br />

disp(’ ’)<br />

disp(’Validity Check’)<br />

167


APPENDIX C. MATLAB CODE FOR POSPIESZALSKI NOISE PARAMETER MODELING<br />

check = 4*N*To/Tm<strong>in</strong>;<br />

(1

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!