Field notes from European Microwave Week 2026, London.
Maury Microwave's booth at EuMW 2026 had two load-pull stations side by side. One showed a harmonic load-pull setup ready to sweep the fundamental, second and third harmonic impedances. The other showed results labelled with a word that matters a great deal for GaN amplifiers: "DoublePulse". Two gain curves that should have described the same transistor clearly did not agree, and that disagreement is the point of the demo.
Two power sweeps, one load
The viewer had two data sources selected: "DoublePulsePowerSweep" and "SinglePulsePowerSweep". The left plot showed gain versus output power for both, at one load termination marked on a 50 Ω Smith chart.
- One curve starts at about 22 dB gain at low drive, stays roughly flat, rises slightly to a peak near 25 dB, and collapses beyond the top of the sweep.
- The other starts much lower, around 17–18 dB and noisy, rises steadily with drive by about 7 dB, joins the first curve near its peak, and then collapses with it.
The plot does not say which curve is which, and I did not confirm it at the booth. The engineering message does not depend on that: the same device, at the same load, shows very different small-signal gain depending on its pulse history.
Why pulse history matters for GaN
GaN HEMTs suffer from charge trapping. When the drain sees a high voltage, typically during the high-power part of an RF cycle or a previous pulse, electrons are captured in traps in the buffer and at the surface. They release slowly, over microseconds to milliseconds and beyond. While trapped, they reduce the channel current, so gain and output power are lower than in the "clean" state.
That makes the measurement depend on what happened just before:
- A single-pulse measurement from a quiet bias point shows the device in a relatively untrapped state.
- A double-pulse measurement applies a first pulse that conditions the traps, then measures in a second pulse shortly afterwards. That is closer to what a radar amplifier sees in a pulse train, or a communications PA sees after a high-power symbol.
A gain curve that rises steadily with drive, as one curve here does, is the typical signature of gain that is suppressed at low drive by a trapped state and recovers as the RF swing increases. The practical lesson is that a load-pull optimum found with one pulse scheme is not automatically valid under another. Gain, output power and efficiency contours can all move.
The status panel showed a gain compression variable of about 2.93 dB, with a slider up to 5.85 dB, a reminder that "P3dB" and similar figures are read from exactly these curves and inherit their dependence on pulse history.
Harmonic load pull: three impedances at once
The second station showed a "Single Tone Load Pull" setup at 8 GHz with an IF bandwidth of 30 Hz. The fundamental source, fundamental load, 2nd harmonic load and 3rd harmonic load were all set to "Sweep", each with its own passive tuner. On the load Smith chart:
- a dense grid of blue fundamental load points sat around the centre;
- a few orange points marked second harmonic loads near the edge of the chart;
- purple points marked third harmonic loads, also near the edge.
Harmonic loads near the edge of the Smith chart are deliberate. High-efficiency modes such as class F and inverse class F rely on presenting near-short or near-open impedances at the harmonics to shape the drain voltage and current waveforms. To explore those modes, the tuners have to reach high reflection at 16 and 24 GHz, while simultaneously providing the right fundamental impedance at 8 GHz.
What to check before trusting load-pull contours
- Pulse scheme: CW, single pulse or double pulse, with pulse width, duty cycle and gap. For GaN, contours are only valid for the scheme they were measured under.
- Bias history: quiescent bias point and settling time between points.
- Tuner reach at the harmonics: the maximum reflection actually achieved at 2f₀ and 3f₀, after fixture losses.
- Fixture de-embedding: reference planes at the device, not at the tuner.
- Measurement order: sweeping across the Smith chart in a different sequence can change results if the device has memory.
Takeaway
Load pull is often presented as a map: find the best impedance and design the matching network for it. For GaN, the map depends on how you drove the device to get there. Maury's two curves from the same load, one measured after a conditioning pulse and one without, made that visible on a single screen. The harmonic load-pull setup next to it showed the other half of modern PA design: the fundamental impedance is only one of three that matter.
Notes: Observations and interpretations are my own, based on what was running at the booth during the show. Values are read approximately from the plot grids in my photos. The mapping of curves to datasets, the device under test and the pulse parameters were not confirmed with the exhibitor.