In the late 1970's solid-state microwave IC technology really became feasible, due primarily to the GaAs MeSFET. In general, one needs two types of transistors in a system: low-noise devices in the receiver, and high-power devices in the transmitter. TI was (and continues to be) very successful in developing the high-power devices, and so I worked on understanding the power-limiting mechanisms and finding ways to circumvent them.
There are two power-limiting effects: the current limit and the breakdown voltage. The current limit was well-understood; it is due to the maximum velocity with which electrons can move through the crystal. The nature of the breakdown voltage was more obscure at that time. The situation was confused by several factors: Many of those working on GaAs MeSFETs had worked on Gunn diodes in the late 1960's and early 1970's and some had developed a strong suspicion that funny things were occuring at the anode of those devices, which was transferred to the drain of the MeSFET. These suspicions were apparently comfirmed by the observation of light-emission from the drain edge in a MeSFET: T. Mimura et al., "Visible light emission from GaAs field effect transistor," Proceedings of the IEEE 65, 1407-8  and R. Yamamoto, A. Higshisaka, and F. Hasegawa, "Light emission and burnout characteristics of GaAs power MeSFETs, IEEE Transactions on Electron Devices ED-25, 567-73 . The picture was also consistent with the notion proposed by M. S. Shur that the breakdown occurred in the high-field region of a stationary Gunn domain in the FET channel L.F. Eastman, S. Tiwari, and M.S. Shur, "Design criteria for GaAs MESFETs related to stationary high field domains," Solid-State Electronics 23, 383-9 .
We entered this study with a firm appreciation of the fact that the only behavior of the device that matters is its behavior in a microwave circuit. In such a case, the device is constrained to operate along a load line, as shown below. (Well, at least near the load line. If the load is properly matched, the conduction current should follow the load line, but nonlinearities in the reactive elements of the device will cause slight deviations.) Now, the experiments in which light emission was observed employed dc bias with the gate shorted to the source. Consequently, they explored the upper right-hand (high voltage, high current) corner of the characteristic curves. Note that the load line keeps the device away from this region, and that breakdown really occurs in the lower right corner (high voltage, low current).
I developed a two-dimensional device simulation code capable that could handle complex geometries, in particular recessed gate structures, which were known to produce higher-power devices. We ran this simulator with knowledge of the load-line constraints, with the results shown below.
It was quite clear that the breakdown should be a purely electrostatic effect, occuring in the nearly singular electric field at the drain-side edge of the gate. My colleagues Hua-Quen Tserng and Paul Saunier set up an experiment to optically examine a device under high-power rf operation and observe the light emission. As we expected, it was the edge of the gate that lit up (below). H. Q. Tserng, W. R. Frensley, and P. Saunier, "Light Emission of GaAs Power MESFETs Under RF Drive," IEEE Electron Device Letters EDL-1 20-1 .
Given the insight into what was actually occuring, and a simple model of the charge distribution in the FET at breakdown, I was able to develop an analytic model which demonstrated that the beakdown voltage was inversely proportional to the channel doping times the channel thickness.