Steve C. Cripps
IMAGE LICENSED BY INGRAM PUBLISHING
Now and again one sees, or even invents, something that could be termed a solution without a problem (SWAP). Curiously though, sometimes these SWAPs turn out to be not only useful after all but maybe even transformational. This probably applies to several of the familiar accoutrements of daily life in the 21st century: the electric car, the Internet, the personal computer, and more. Closer to the home of microwave technology, I remember the term SWAP being applied to microwave monolithic integrated circuits (MMICs) in their evolving era of the 1980s; it seemed that the amount of gallium arsenide required to implement the passive elements of a functional MMIC would always be much more expensive compared to the existing, ubiquitous chip-and-wire hybrid approach. The rest of the MMIC story, as they say, is indeed history.
However, not all such inventions lead to such a happy ending, and the path of “progress” is strewn with SWAPs that get quickly written out of the script. I rather think many of us have experienced the SWAP effect early in our careers and can still be heard, on occasions, boring the pants off our younger colleagues as we wax lyrical about them. That is exactly what I plan to do—so, with no further apologies, I present the avalanche diode and, most importantly, its rather mysterious and short-lived evolution, the “anomalous avalanche mode,” or TRAPATT, an unlikely acronym from “trapped plasma and avalanche-triggered transit” device.
Dedicated readers of this column, should there be any left, may remember that I did broach this subject some years ago [1], although this was something of a more formal recognition of the academic approach to the subject, most notably promoted by Carroll [2]. Indeed, this particular reference (Carroll, not Cripps—later my own Ph.D. advisor) was itself unorthodox inasmuch as it focused on the analysis of the TRAPATT circuit, rather than the device physics of the diode itself.
This was the pre-GaAsFET era, when microwave semiconductor research was all about two-terminal devices in the form of “transferred electron” or “Gunn” diodes, which were not actually diodes, and avalanche diodes, which were. The most familiar brand of microwave avalanche diode was the IMPATT (impact avalanche and transit time), which was able to produce oscillations at gigahertz frequencies and has, marginally, survived to the present day. Both the Gunn and the IMPATT were fundamentally oscillators in themselves; the basic semiconductor chip itself would oscillate under suitable bias conditions. Indeed, J.B. Gunn, the inventor of the Gunn diode, famously commented that the most important observation he had ever written down was the word “noisy,” in his notebook, when measuring the I–V characteristics of gallium arsenide samples. The IMPATT has a somewhat similar history, inasmuch as it was predicted as a theoretical concept well before the proposed structure could be fabricated [3], but oscillations were later observed experimentally and, fortuitously, using diodes that had not originally been fabricated for this purpose.
This serendipitous aspect of avalanche diode oscillation was certainly an important aspect of my own work on TRAPATTs. Two main features differentiated the TRAPATT from the more well-known IMPATT: the TRAPATT appeared to deliver much higher powers, one or even two orders of magnitude higher than either Gunn or IMPATT diodes of similar physical size, and, somewhat conveniently, showed comparably quite high efficiencies, in some cases approaching 50% at power levels approaching a kilowatt. The RF waveforms were characterized by very short current pulses, fewer than 100 ps, and a spikey voltage that collapsed rapidly after peaking well above the breakdown level. As such, most TRAPATT results were obtained using pulsed bias supplies running at low duty cycles between 0.1% and 1%. Here was the SWAP: the TRAPATT could deliver large powers, levels unheard of for a semiconductor device at the time and still “healthy” today, with very small, cheap structures—but only for a microsecond or so. The obvious application was radar, but the rather noisy spectral properties of the TRAPATT resulted in a fairly rapid decline of interest in this application, the radar system industry being more interested in the possibilities being strongly touted by MMIC developers for phased-array radar systems.
One of the interesting aspects of any SWAP narrative is to consider, or speculate about, what the impact of today’s technology would be. The TRAPATT era was pretty much confined to the 1970s, when solid-state microwave power technology was struggling to generate even a watt or two of power at gigahertz frequencies. Having myself built TRAPATT oscillators using cheap wire-ended computer diodes, in which the tiny die itself is essentially suspended in inside a small glass package but that could still generate 50 W of power for a few hundred nanoseconds before it fried, I have always wondered what could be achieved using modern fabrication techniques. For example, if the duty cycle could be increased and the RF pulse length substantially increased, other applications may start to come within range. However, to get any interest at all, it has always seemed that I will have to reinvent the TRAPATT myself, as one of the very few remaining TRAPATT practitioners still around and with access to the necessary equipment. Before I describe such an effort (yes . . . it’s coming.!), it would be worth backtracking a little to the essentially still unresolved questions of how the TRAPATT really works and, even more mysteriously, how it starts.
Pretty much all of published attempts to explain TRAPATT oscillation focus on the device itself and, in particular, the physics of avalanche breakdown. There is much talk of the Read equation [3], avalanche shock fronts, and trapped plasmas. The role of the passive circuit somehow gets secondary status; it was given little more than a handwave. I prefer to reverse that process to consider a typical TRAPATT circuit and show, with my own handwave, how it is at least conceivable that large signal oscillation could be sustained by a diode driven repeatedly and quickly into its breakdown regime.
Figure 1 shows an idealized set of TRAPATT diode waveforms, as predicted and—rather less frequently—actually measured. The key aspect of avalanche breakdown is the dynamic behavior when viewed on picosecond timescales. The current waveform is a sequence of very short pulses, which are generated due to the voltage, which sweeps to a peak value well above the avalanche breakdown level and then collapses due to the impedance of the external passive circuit. The key feature is that the avalanche current multiplication continues while the voltage is above the breakdown level, even though it has passed a peak value. In retrospect, looking at the idealized waveforms in Figure 1, it is obvious that the voltage closely resembles the derivative of the current, with suitable scaling and offset. This is certainly a handwave, but there certainly seems to be the possibility of continuous oscillation once the process has been initiated if the external circuit can act essentially as a differentiator—the trick being to generate enough charge in each cycle for the circuit to produce the necessary overvoltage kick that keeps the process going.
Figure 1. Idealized TRAPATT waveforms. LPF: low-pass filter.
However, how such an oscillation gets started has remained something of a mystery that was never fully solved. Gigahertz-bandwidth oscilloscopes in the 1960s era were entirely of the sequential sampling type, and, although TRAPATT waveforms were reported, looking remarkably similar to those in Figure 1, such an instrument was not capable of displaying the very noisy oscillation buildup period.
Most reported TRAPATT circuits had the basic form shown in Figure 2. The diode is placed at the end of a length of transmission line, which initially is considered to be terminated with a short circuit. A key observation is that the TRAPATT oscillation frequency is always slightly higher than the half-wavelength of the line, so that, at the oscillation frequency, the line length is slightly shorter than a half-wavelength. I will not repeat the analysis of Carroll [2], having already done so in [1], which shows that, if the shortfall is small enough, and the frequency analysis is restricted to harmonics of the fundamental only, such a circuit behaves as a negative inductance, thus providing the required differential voltage. Needless to say, and in the serendipitous spirit of the TRAPATT, this circuit was demonstrated well before the theory was published, but looking at the “Evans” TRAPATT circuit on a modern simulator certainly upholds the “differentiator” function and suggests that various other circuit topologies may well work.
Figure 2. A TRAPATT oscillator circuit.
If we assume that the diode delivers a short repetitive pulse into this impedance, being inductive, the voltage across the diode will be a scaled derivative of the current, which is offset by the dc supply, as indicated in Figure 1. Therefore, if the dc supply is set to a level just below the diode breakdown, the voltage will swing the diode briefly into its breakdown region, and carrier multiplication will escalate rapidly and only then to be suppressed when the voltage collapses due to the differential action of the circuit. Therefore, continuous oscillation is conceivable; a few tweaks in the various parameters that characterize the avalanche breakdown and, especially, how the generated charge is extracted from the junction after the voltage collapses are necessary to convert a handwave into reality—but sometimes, it does all appear to work .
As such, I certainly do not dismiss the focus on the device physics in the vast majority of papers and articles published on TRAPATTS in the 1970s, although, with this long in retrospect, it is clear that, in devising suitable diode doping profiles, it was clearly quite easy to come up with structures that had the “necessary” properties, while experimental evidence suggested that “sufficient” properties could also be demonstrated using diodes that were never intended to work at gigahertz frequencies at all, never mind to generate tens or hundreds of watts of power. Custom or otherwise, TRAPATT oscillators that yielded tens of watts at efficiencies approaching 50% were reported at frequencies up to X-band and hundreds of watts were reported at lower gigahertz frequencies.
The first indication of the serendipitous TRAPATT was in a short paper [4] that described a cheap commercial diode, the Fairchild FD300, as working in a TRAPATT circuit. Admittedly, the impressive power of 68 W was obtained at a somewhat low frequency, by today’s standards, of 630 MHz, not to mention a very short pulse and long duty cycle, but it did rather set a cat among the pigeons as far as the TRAPATT theories were concerned. It also opened up a door for doing some research into TRAPATTs without necessarily having access to a semiconductor fab (as they were not then known) to manufacture custom diodes. I was therefore somewhat fortunate, depending on one’s views on SWAPs, to find this open door as I started my Ph.D. work.
Curiously, one reason why playing the TRAPATT game—essentially, testing any diode that came to hand for its potential TRAPATT properties—was not a widespread activity was not so much due to the design and construction of the oscillator circuit, which were remarkably simple but to the difficulty of finding a suitable generator for the bias pulses. At that time, generating reasonably sharp 1-µs pulses of up to 200 V at currents of 1 or 2 A was well outside the capability of regular lab pulse generators, as it still is. Commercial products were available, such as a classical product from Velonex, but more Heath Robinson contraptions involving a long length of coaxial cable and a mercury-wetted relay were quite commonly employed.
At the time, I did, in fact, have access to a Velonex pulser, which was full of ultrahigh-frequency vacuum tetrodes and could deliver 1-kV pulses with a risetime of about 10 ns. This was considered important inasmuch as no one seemed to have much of a theory as to how the TRAPATT oscillation started. Being very much a large signal effect, it was not possible to see how it could build up gradually from the noise floor, and one possibility was that the initial kick of the bias pulse may get things started. This is certainly an area where a half century of development in power electronics can make a huge difference; such bias pulses can now be generated quite easily using a high-side PMOS switch, and it was an awareness of this that made me speculate about whether I could repeat some of the TRAPATT results reported in the 1970s. This was only in part a nostalgic exercise; recent work on the interactions of microwaves with biological samples have shown some interesting results such a “electroporation” of cells [6], and the use of short pulses is a way of excluding local heating effects.
The 1S44 diode had already been established as one that readily performed as a TRAPATT oscillator at low gigahertz frequencies, albeit with somewhat lower power and efficiency than the custom diodes being reported in the literature. It was generally described by the then-manufacturer as a “core driver” diode, used in ferrite core memory banks. I also managed, with statutory serendipity, to discover the 1S952, also a core driver diode, which had higher breakdown and improved TRAPATT properties [5]. Fifty years on, a quick search on eBay did not reveal any 1S952s, but I did obtain a batch of 1S44s (Figure 3). Whether these were “original” or a modern equivalent I had no idea, but the project appeared to have been launched, and I had to take a YouTube lesson on the design of a high-side switch, which quickly resulted in a viable circuit that would handle 200 V and a couple of amps while delivering a pulse of a few hundred nanoseconds.
Figure 3. 1S44 diodes.
Although a TRAPATT circuit appears very simple in concept, being essentially a length of transmission line terminated in a short circuit, there are a couple of extra issues to consider. The waveforms clearly contain substantial higher harmonic components, and the line has to be able to support these without much loss. It is also necessary to consider how to extract power from the oscillator. This was achieved by replacing the short with a low-pass filter that reflected harmonics but introduced an antiphase voltage component at the fundamental [2].
Therefore, armed with this rather ragtag collection of memories and handwaving theories, I cobbled up the test circuit shown in Figure 4. It is, simply, a microstrip line terminated at one end by the diode, and the other end is a low-pass filter, which was rather crudely designed to match the diode impedance at a predicted oscillation frequency around 2.4 GHz. Setting the pulsed bias supply to give about a 0.5-µs pulse at very low duty cycle (1-kHz pulse-repetition frequency) and with a curious mixture of nostalgia and anticipation, I gradually turned up the voltage to the point where breakdown commenced. Sure enough, and, admittedly, following some hours of circuit tweaks and diode burnouts, I obtained the bias voltage waveform shown in Figure 5, which displays a classical signature of oscillation whereby, after an oscillation buildup period, the mean diode voltage drops to a level substantially below the breakdown voltage. This indicates that there is an RF oscillation, which swings the voltage above and well below the breakdown level, thus generating sharp pulses of current, probably, in this case, in the region of a 10-A peak. The second trace in Figure 5 shows the detected 2.4-GHz RF output, which, after some judicious tuning and bias adjustments, indicated a pulse power of about 30 W.
Figure 4. The TRAPATT evaluation circuit board.
Figure 5. The TRAPATT voltage bias (lower trace, 50 V/div) and detected RF (upper trace, approximately 20 W/div; horizontal: 80 ns/div).
One has to be cautious indulging in technical nostalgia, but this still remains, 50 years on, the most remarkable result I have personally experienced. Take a small, cheap computer switching diode, drop it into a simple circuit, and get 30 W of microwave power. Drinks on me tonight.
There are caveats. As indicated in Figure 5, the oscillation takes many RF cycles to become established, and this very noisy start-up was one of the reasons the TRAPATT had such a short attention span. This made it difficult to display anything approximating an RF waveform. Sampling oscilloscopes available at that time typically had sampling rates no higher than 100 kHz, so, for a bias pulse length of less than a microsecond, each point on the RF waveform had to be sampled from successive bias pulses. Therefore, in devising a suitable triggering scheme, it was important only to trigger the sampling process once the oscillation had stabilized, and this required some trickery [7]. The design of suitable voltage and current probes was also challenging, and this is a subject I have addressed in a previous “Microwave Bytes” article [8].
All of that said, I cannot resist the temptation to show my best effort in capturing the RF voltage, shown in Figure 6. The calibration of the probe has to be at best approximate and is certainly questionable, but the general form of the voltage does appear to confirm TRAPATT oscillation, noting that the basic reactive waveform in Figure 1 now has an additional fundamental sinusoidal component due to the power extraction.
Figure 6. The measured TRAPATT diode RF voltage waveform. (See the text regarding probe calibration.)
So that’s about it—an illustrated history of a SWAP that never went anywhere further. “Big deal,” I hear the “gallium nitride (GaN) generation” saying. “Just 30 W and only for a couple of hundred nanoseconds?!” The key point, however, is that the active device here is a tiny, simple little silicon diode, probably costing a few cents if it were in production today. The short pulse length is entirely a function of the package, which provides essentially no heat sinking at all. For some decades, I have felt that modern microwave semiconductor fabrication could surely come up with a TRAPATT diode capable of delivering hundreds of watts for tens of microseconds. More speculatively, I wonder whether some emerging industrial microwave applications really need the pristine signal generated by a GaN transistor, which is required for telecom applications but seems something of an overkill when only raw microwave power is required. The TRAPATT does open a door that reveals the avalanche breakdown effect as an alternative simple and cheap way of generating large amounts of microwave power, limited only by thermal considerations.
Maybe the “Avalatron” could yet replace the magnetron .
[1] S. C. Cripps, “Trapwave Inc. [Microwave Bytes] ,” IEEE Microw. Mag., vol. 9, no. 4, pp. 46–51, Aug. 2008, doi: 10.1109/MMM.2008.924789.
[2] J. E. Carroll, “An analytic theory for the Evans circuit for avalanche diodes,” IEEE Trans. Microw. Theory Techn., vol. 18, no. 11, pp. 977–979, Nov. 1970, doi: 10.1109/TMTT.1970.1127382.
[3] W. T. Read, “A proposed high-frequency, negative-resistance diode,” Bell Syst. Tech. J., vol. 37, no. 2, pp. 401–446, Mar. 1958, doi: 10.1002/j.1538-7305.1958.tb01527.x.
[4] R. J. Chaffin et al., “A poor man’s TRAPATT oscillator,” Proc. IEEE, vol. 58, no. 1, pp. 173–174, Jan. 1970, doi: 10.1109/PROC.1970.7579.
[5] C. Oxley et al., “Design and performance of I-band (8-10-Ghz) TRAPATT diodes and amplifiers,” IEEE Trans. Microw. Theory Techn., vol. 27, no. 5, pp. 463–471, May 1979, doi: 10.1109/TMTT.1979.1129650.
[6] S. C. Cripps et al., “A theoretical and experimental study of the antiparallel TRAPATT diode oscillator circuit,” in Proc. 3rd Eur. Microw. Conf., Brussels, Belgium, 1973, pp. 1–4.
[7] N. A. Slaymaker and J. E. Carroll, “Improved sampling technique for the observation of high harmonics in TRAPATT waveforms,” Electron. Lett., vol. 7, no. 18, pp. 554–555, Sep. 1971, doi: 10.1049/el:19710374.
[8] S. C. Cripps, “Probing times [Microwave Bytes] ,” IEEE Microw. Mag., vol. 10, no. 1, pp. 28–34, Feb. 2009, doi: 10.1109/MMM.2008.930689.
Digital Object Identifier 10.1109/MMM.2023.3294879