So I've been wringing out a new APD front end. It's a 500-MHz TIA with a pHEMT bootstrap, and wants to oscillate at 2.7 GHz with a strong squeg at about 1 MHz. Changing the drain current changes the duty cycle of the squegging, and reducing it below about 2 mA makes it stable.
The oscillation frequency doesn't change much (5% or so) with drain current. All of which suggests that the oscillation is due to some reasonably-sharp resonance someplace--the squegging gets worse at higher gain, but the oscillation frequency doesn't move around. The waveform is more or less sinusoidal-looking, but that's not surprising since the vertical bandwidth of the scope (a TDS 694C) is about 3.3 GHz. The oscillation doesn't depend on whether the TIA is connected or not.
Lower-speed front ends based on the same sort of transistor are famously stable--one of them runs the bootstrap across a two-inch-long FFC cable going to a MPPC on a cold plate. They also work great as the bottom device in a cascode with a 45-GHz SiGe NPN. (The NPN needs a base-stopper bead, but the pHEMT doesn't.)
The difference with this one may be that there's a bootstrapped pour under the summing junction components, driven by the pHEMT source.
Sooo, I took a bare board, bodged in a U.FL micro coax connector from the bootstrapped pour to ground, and hung it on the front of one of my trusty Tek SD-24 TDRs, like so:
Here's the result:
There's a capacitive dip (3.0 pF as measured on a Boonton 72BD) followed by a delayed and not-too step open-circuit reflection and some ringing. Interestingly the reciprocal of the round-trip delay is right around 2.7 GHz, which would make sense with an open-circuit transmission line resonator.
I also did some measurements of the Murata beads we use as base/gate stoppers for microwave transistors. Our faves are the Murata BLM1xBA series.
The following scope photo shows a TDR of a short piece of 0.080" hardline with various low-Z Murata ferrite beads. From top to bottom at the beginning of the falling-edge transient: BLM18BA100SN1D (10 ohms @ 100 MHz, light orange); BLM15BA050SN1D (5 ohms @ 100 MHz, green); BLM18BB100SN1D (10 ohms @ 100 MHz, purple); and BLM15BB050SN1D (10 ohms @ 100 MHz, yellowish).
The BA-series beads show a lot better high-frequency impedance than the BB-series ones, despite their datasheet curves being very similar. The BLM15BA050 shows a bit of undershoot near 1.5 ns, but the others are all basically monotonic at late times.
Interestingly the BLM15 (0402) and BLM18 (0603) beads look identical at this resolution--the traces lie right on top of each other.
Fun stuff, and it'll be more fun once I get the resonance problem knocked. Putting the pHEMT source connection near the middle of the pour instead of at one end will help, I expect--this part ought to be less likely to oscillate at 5 GHz.
Comments welcome.
Cheers
Phil Hobbs