10.06.2016 Views

eldo_user

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

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

Optimization<br />

Robust Optimization Using Corners<br />

* DRAIN CHARAC W=25U L=25U AT VBS=0.0<br />

.INCLUDE d25x25.dat<br />

.DC DATA=DATA_D25X25 ! DATA DRIVEN ANALYSIS<br />

.OBJECTIVE DC LABEL=FIT_IDRAIN I(V101) ! DOUBLE DC SWEEP OPTIMIZATION<br />

+ GOAL=DATA_D25X25(IDS) TYPVAL=1E-4<br />

* DRAIN CHARAC W=25U L=2.5U AT VBS=0.0<br />

.ALTER DRAIN 25X2P5 AT VBS=0<br />

M25X25D1 101 102 0 103 NMOS W=25U L=2.5U NRS=0.12 NRD=0.12<br />

.INCLUDE d25x2p5.dat<br />

.DC DATA=DATA_D25X2P5<br />

.OBJECTIVE DC LABEL=FIT_IDRAIN I(V101)<br />

+ GOAL=DATA_D25X2P5(IDS) TYPVAL=1E-3<br />

The parameter TYPAL has been used to globally rescale the objectives (the extracted measures<br />

and the associated target values), such that the low current data is not dominated by others data.<br />

Optimization Results<br />

The optimization was performed on a 32-bit Linux machine and the results generated during the<br />

optimization process are shown in Table 13-11:<br />

Table 13-11. Nmos Example Results<br />

Minimum Error Global Error Maximum Error<br />

Length L=25U 0.0 1.1×10 −18 3.5×10 −9<br />

Length L=2.5U 0.0 6.4×10 −17 3.0×10 −8<br />

These values are the result of a synthetic problem. The circuit parameters were fixed, the data<br />

was then generated, and the circuit was optimized after performing a perturbation of the initial<br />

point. This is why the global error is so small. The residuals are almost zero. This is not the case<br />

with realistic data where the physical measurements involve some variability in the current<br />

curves.<br />

Related Topics<br />

Optimization Options<br />

Inner and Outer Sweep Parameters<br />

Robust Optimization Using Corners<br />

Netlist file: aop_optim.cir<br />

Eldo® User's Manual, 15.3 677

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!