Interesting behaviour of BJT

I haven't come across such an odd phenomenon in a while, so I thought I'd share.

I've made a circuit to generate a linear ramp waveform from a 5V clock. A BJT current mirror sources a constant current into a capacitor, and and the capacitor is periodically discharged through a diode, which is driven by the clock. The clock has a high duty cycle such that its low time is just 25ns. The schematic is shown below, with accurate component values.


Here's the ramp waveform at 2MHz. Basically expected, except the recovery of the diode seems a bit slow.


I tried lowering the clock frequency a bit. As expected, the ramp reaches a higher max voltage before being brought low. Also it looks as if the diode recovery is getting slower...


Lower the frequency a bit more to 1.56MHz, and now the max voltage takes the BJT towards saturation. But the recovery from the pulse is even more distorted now....


Lower the frequency a bit more to 1.54MHz, and suddenly I get this:


Strangely, if I increase the frequency to 1.7MHz, the waveform is still messed up:


As the frequency is further dialed up to 1.9MHz, the ramp is starting to show up a little bit again...


And at 2MHz, it suddenly snaps back to the original ramp waveform:



I've pretty much ruled out the diode as the culprit (tried different devices with no effect). It seems to be that the BJT does not like being brought out of saturation quickly. When this happens, it will conduct more strongly than normal for a hundred or so nanoseconds, causing it to snap back into saturation. This gives the circuit a sort of hysteretic behaviour.

I know BJTs have storage time, which limits limits how fast it can exit saturation when the gate current is switched off, but I never heard of something similar happening when operated when the base current is fixed while the collector is driven. But I assume it's the same phenomenon? The BJT pair I used doesn't list anything about storage time, but it is specified for use in high speed current mirrors and diff amps. So how would I pick a replacement?

I've verified that a baker clamp prevents the circuit from snapping back into saturation, but the distortion after each reset pulse is still there, which is annoying. I think finding another BJT is the only way to get rid of that....

Are you really operating the current mirror with transistors in inversion or is it a drawing fault?

- - - Updated - - -

Apparently not, otherwise the current mirror won't have current gain near 1.

- - - Updated - - -

The unexpected point is that you can apply 5V Vce but the transistor is apparently staying in saturation.

I would try a current mirror made of really fast transistors, e.g. BFT92. Matched GHz pairs are rather exotic, I only know HFA3135.

Oops, yes they were drawn backwards, fixed below:


A faster pair would work, but now I've also encountered the issue that the matched pair is susceptible to thermal gradients across the PCB, which is a problem because they're on one side of a 100W power supply. I can literally tune the amplitude of the ramp up or down by bringing the tip of my soldering iron near each side of the package. This would be great fun if it weren't a serious project.

Since I'm already stuck with the PCB, I will have to hack some other current bias method which can operate at high frequency and is fairly temperature independent... like a resistor from a higher bias voltage.

You can greatly improve the thermal stability by placing identical resistors in each emitter.
The larger the better, but then you lose some voltage head room.
Much faster transistors would also be a big step forward if you really need 2Mhz.

Yes, emitter degeneration would help, and also maybe help desaturation recovery. It's a bit tricky though, since I'm working with a SOT363 package...

What I still can't figure out is the cause of the increased current after the reset pulse. It's significant even when the BJT isn't brought near saturation, like in the first two figures. It's as if the high dv/dt causes the beta beta of the output BJT to increase momentarily. Any physics gurus have any insight on that?