Optoelectronics with Carbon Nanotubes

(a) (b)
Figure V-10. Full-width at half maximum of electroluminescence spectra as a function
of (a) applied power and (b) light emission intensity. The device was operated in the
diode (red squares), resistor (blue circles) and single-gate FET (black triangles) modes
by changing top-gate voltages. The inset in (b) shows the example of a typical
Gaussian fitting with data taken with the device operating in the global gate FET mode
(VTG1/VTG2 = +5V/+5V) at the drain-source bias of 12 V. The data correspond to
Figure V-9 (c). The widths were extracted by fitting the Gaussian function on each
spectrum. Since the spectral peak is near the edge of the detection range and widths
depend sensitively on the peak position, the average peak position was determined first
and used as a fixed peak position (no statistically significant peak shift was observed)
before the fittings were performed.
We attribute the small and rapid rise in width we see at a low power (up to ~50 W/m) to
the multiple-tube effect. While Adam et al. found that only 4 % of the largest diameters
contribute to emission in carbon-nanotube network FETs, our parallel-tube film FETs, with
distinctly separate channels, are statistically more likely to involve a somewhat wider range of
diameters that participate in conduction. We can think of a parallel-tube film as a collection of
many networks, each with a limited number of tubes. Depending on the degree of bundling, the
conducting path in any given channel can involve smaller-diameter tubes in the emitting region
than the largest diameter tubes in the sample. To find the distribution of conductive paths would
require a statistical analysis of devices made with different widths from the same film (since the
film thickness varies), which is beyond the scope of this study. However, the very rapid rise in
the width seen at low electrical power can be attributed to this effect. At the lowest bias/power,
only the paths with largest diameter tubes can contribute to conduction. As the source-drain bias
102