Optics & Photonics — Ultrasensitive Instrument Design — Analog Electronics

pHEMT probe

The application is in a molecular probe experiment, where we're trying to measure currents of 1 nA or so in a 100-MHz bandwidth. A bit of a recreational impossibility, possibly, but cool if it turns out you can do it. (Yes, I know that 1 nA in 10 ns is 62 electrons.)

The real thing will have the pHEMT chip wire-bonded directly to the DUT, to minimize input capacitance.

The cascode design is to eliminate Miller capacitance and allow higher voltage gain, in order to swamp the second-stage noise.

The pHEMTs have drain impedances that are low and nonlinear, whereas the BFP650 has effectively infinite Early voltage, which helps the stage linearity considerably, and allows single-stage voltage gains of 30 to 50 with no problems.

The measured 1/f corner frequency of the pHEMT is around 10 MHz.

• Board dimensions are 25 x 50 mm.
• pHEMT Q1 is actually a Skyworks SKY65050, which is nice and stable but has a bit more input capacitance than I'd like--about 0.8 pF. I'm trying out some other possibilities, because we really need another factor of 2 here.
• There's a 10-ohm bead bodged in right at the base of Q2;
• The zero-ohm input resistor R11 is actually a 1-uF cap;
• The collector load R2 is actually 200 ohms;
• Supplies are about +-10 V, so VDD is about 6.4 V (the LM329's voltage minus a Vbe drop);
• Bias is adjusted for about 1.9 V_DS on the pHEMT and about 3.5 V at the output, i.e. just under 15 mA bias current.

	Noise results This is the measured noise of one prototype SKY65050 amplifier as a function of the shunt resistance on the input. Normally you want to keep the termination exactly constant and change the noise voltage, in order to prevent shifts in the overall gain from screwing up the measurement, but after three tries at making an adequate jig for that, I gave up and just stacked feedthrough terminations on the input.
	These curves show that the noise is pretty low, and the gain looks pretty constant with source resistance. The gain is one of the unknowns, however, so to get a noise measurement, I combined pairs of these output noise curves, so as to cancel out the gain and the known resistor noise. The following plot shows the results for various pairs with the ATF34143, plus fit curves for it and for the SKY65050. The noise plots tell a consistent story, which is encouraging. It's even more encouraging that the story is less than 0.4 nV noise and 0.96 pF input capacitance.
	Input Capacitance Measurements The input capacitance is computed by a sort of ju-jitsu: Using a 25-ohm source, change the input coupling capacitor from 1 nF to 1 pF, measure how much the output drops, and compute the input capacitance from the voltage divider ratio. This works because the input conductance is very small below about 1 GHz.