GaN PA at 28GHz - thermal runaway in phased arrays?

Running a 28GHz GaN PA in a 64-element phased array, and thermal coupling between elements is shifting the bias point. Has anyone characterized thermal runaway risk in dense GaN arrays? What derating do you apply?

Dense 28 GHz GaN arrays do see thermal coupling walk the bias — “runaway” is less often a sudden thermal latch and more often a slow shift of Idq / gain / phase that eats your calibration and PAE.

What teams usually characterize:
1. Element-to-element thermal resistance (IR or on-die sensors) with neighbors driven at your real duty cycle, not single-element CW.
2. Bias topology: per-element vs shared gates/drains; shared supplies can amplify a hot element’s current hogging.
3. Derating: many production arrays hold a junction or case margin vs datasheet Tmax (often several tens of °C) *and* a max ΔT between neighbors so the beamforming LUT stays valid.
4. Control loop: temperature-compensated gate bias or periodic Idq recal, plus a hard current/thermal foldback before you rely on “the GaN will take it.”

If thermal coupling is already moving bias in your 64-el fixture, instrument one center and one edge element through a long soak at max duty and plot Idq vs time with neighbors on/off — that curve tells you whether you need better heat spreading, lower duty, or active bias. What packaging (bare die / QFN / AiP) and what duty cycle are you running?

— rfforum 编辑部 / Editorial Team