I'm just finishing up an interesting job for a group in the Department of Chemistry at Notre Dame. It's a laser noise canceller that works down in the 10 nA-50 uA range, about 100 times lower current than the original model.
Noise canceller performance depends pretty well completely on BJT diff pairs and cascodes performing as Ebers and Moll predict, so that the fluctuations see exactly the same gains as the DC.
The two main sources of error are log nonconformance, which makes the diff pair's large- and small-signal current division ratios different, and beta error (1/h_FE - 1/h_fe), which causes the same problem in the base/collector split of the emitter current.
I built a tester that works over an emitter current range of 100 pA to
100 mA, and seems to do a good job of measuring those two parameters directly. (You have to be careful about junction temperature in the log conformance measurement, of course.)The vellum got a little bit munched in the process, but the scan is still pretty readable and has some points of interest. (It even has a 555.)
Building it would have taken half as much time if I'd used a bigger box!
Early results indicate that modern RF transistors are _much_ better than old ones at low I_C. Both BFT25As and the Infineon HFA3xxx parts have betas of well over 20 at 100 pA I_C.
Cheers
Phil Hobbs