Biasing with variable supply voltage

I'm designing a simple radiation detector using a PIN photodiode to pick up gamma radiation. When a gamma particle strikes the photodiode, it'll produce a tiny pulse that the rest of the detector needs to amplify. Something like Maxim App Note 2236 (http://www.maximintegrated.com/app-notes/index.mvp/id/2236) but using discrete components.

The kicker is that it uses supercaps for power storage, which means the supply voltage will vary from about 4.5V down to about 2V, and I need to keep the quiescent current for the whole thing as small as possible, ideally around 1 uA.

I've been trying to figure out how to bias the transistors. To minimize quiescent current but remain sensitive they should be just at the threshold of turning on, but as they draw power from the supercaps the supply voltage will gradually decrease, so a traditional voltage divider would probably result in reduced sensitivity as the bias point drops with supply voltage. The zener diodes I looked at all had leakage currents on the order of 7 uA which is over budget for the design.

Is there some sort of self-biasing arrangement that would use negative feedback to keep the transistor at or just below conduction? Any other ideas? Either MOSFETs or BJTs would be acceptable in the design.

I'm mostly a digital and embedded software guy and am gradually trying to re-learn analog design, so suggestions and pointers would be much appreciated. Thanks in advance!

I looked at Maxum's schematic, there are no transistors (external). If you want help you could be more clear on what your are referring to when you ask. What transistors are you talking about.

Also it isn't as simple as Q-current. The current draw is coming out of the caps which is energy so even if you did design a circuit to minimize Q-current, if it is a linear circuit regulator you have to make sure that the linear circuit doesn't dissipate energy just trying t regulate the current to a minimum. Simply you don't want to spend a dollar trying to save a dime (use 50mA regulating 2uA).

Be more specific.

Sorry, I will try to be more specific.

You are correct, the Maxim design uses op-amps as its amplifiers rather than transistors. My goal would be to instead use transistors as, essentially, class B amplifiers to provide the necessary voltage gain, though I don't care that much about keeping them in the linear region as long as they provide the necessary gain. The last stage will probably be a monostable multivibrator rigged to light an LED for about 2 ms, so some distortion would be acceptable.

As for regulators, the goal is not to have one: I'm trying to design the circuit so that the supercap provides the supply voltage directly without regulation. I'd like to run a solar cell through a silicon diode to charge the cap and then have the circuit draw power directly from the cap at whatever voltage the cap can provide, down to around 2 volts. The two volt number is an arbitrarily chosen design point; the actual minimum voltage will depend on the details fo the circuit design. My goal is at least 24 hours in standby mode once the cap's charged.

As a result of using the cap's output voltage directly, supply voltage will decrease during operation. The decreasing supply voltage creates a design challenge because most biasing techniques like the normal voltage divider approach rely on having a known fixed voltage, or they use a zener diode to enforce one. The former approach would require a regulator which, as you point out, would draw too much current. A buck/boost chip or energy scavenger might be more efficient, but it would probably introduce too much noise. And a zener diode would probably still be pushing it on current requirements.

Which led me to the question I posted above. Is there another way to bias the transistors? Possibly one that uses feedback to put the DC bias point right at the cutoff voltage?