A note on quasi-saturation:
I ran a 2D BJT simulator for a hypothetical device, which I can only assume to be representative rather than a specific device which would mean knowing a lot of process info.
At low collector voltages and high base bias (actually 1...2V and 0.6 ...0.8V) it shows that the potential drop across the epi layer reduces the potential on the base in the vicinity of the BC junction. Rather than increasing hole injection into the collector, (this was an NPN device) the hole density does increase, but that translates into additional emitter-base current.
This seems to confirm what the literature says - quasi-saturation is a gain reduction mechanism, caused by the epitaxial layer resistance. The extra base current will affect the ft as well.
It also implies that measuring the output characteristics with a fixed Vbe rather than Ib *will show less effect, but the reduced gain will of course increase distortion.
As Bob C says, near clipping is going to be increasing distortion anyway as things tend to get non-linear, and as Mark Johnson indicated, for small signal stages cascode implementation can largely eliminate the problem, if you can afford the extra voltage drops.
The current levels and potentials did not seem to me to get into the Kirk effect regime. Carrier velocities did not quite reach saturation levels, though got close. But that may be different with a different description of the transistor (e.g. lower epi doping).
For the interested - I included a simple implementation for mobility (doping and field effect) and band-gap narrowing, and solved the electrostatic and quasi-Fermi levels for electrons and holes in a Newton-Raphson iterative approach, simultaneously. Predicted gain of the device was perhaps on the high side (~490) but not unheard of for a BC547 type device, though no means of calibrating the model without a full device description. Even detailed electrical analysis of a device won't pin down everything that is needed without geometrical data too. So it can only offer qualitative insight.
* if this can be accomplished under constant temperature conditions!
I ran a 2D BJT simulator for a hypothetical device, which I can only assume to be representative rather than a specific device which would mean knowing a lot of process info.
At low collector voltages and high base bias (actually 1...2V and 0.6 ...0.8V) it shows that the potential drop across the epi layer reduces the potential on the base in the vicinity of the BC junction. Rather than increasing hole injection into the collector, (this was an NPN device) the hole density does increase, but that translates into additional emitter-base current.
This seems to confirm what the literature says - quasi-saturation is a gain reduction mechanism, caused by the epitaxial layer resistance. The extra base current will affect the ft as well.
It also implies that measuring the output characteristics with a fixed Vbe rather than Ib *will show less effect, but the reduced gain will of course increase distortion.
As Bob C says, near clipping is going to be increasing distortion anyway as things tend to get non-linear, and as Mark Johnson indicated, for small signal stages cascode implementation can largely eliminate the problem, if you can afford the extra voltage drops.
The current levels and potentials did not seem to me to get into the Kirk effect regime. Carrier velocities did not quite reach saturation levels, though got close. But that may be different with a different description of the transistor (e.g. lower epi doping).
For the interested - I included a simple implementation for mobility (doping and field effect) and band-gap narrowing, and solved the electrostatic and quasi-Fermi levels for electrons and holes in a Newton-Raphson iterative approach, simultaneously. Predicted gain of the device was perhaps on the high side (~490) but not unheard of for a BC547 type device, though no means of calibrating the model without a full device description. Even detailed electrical analysis of a device won't pin down everything that is needed without geometrical data too. So it can only offer qualitative insight.
* if this can be accomplished under constant temperature conditions!
Last edited:
I have modified Bob's SPICE models for 2SC3503 and 2SA1381 to include quasi-saturation effects based on the measurements earlier in the thread and in simulating my amp there is indeed the expected effect on distortion when those devices are used for the VAS but it only happens when as the output is approaching to within 2V of clipping. I also use them for the diamond buffer and pre-drivers, used there it degraded gain margin by up to 5 dB when the output swings close to maximum, depending on the amount of current being sourced by the output stage. I'd say that's pretty significant.
Another question that I have, is whether there is also a similar massive Ft droop in the quasi-saturation region that we see with the high-speed NPN power devices like the MJL4281. Interestingly, the MJL3281A doesn't suffer as greatly from this effect. Perhaps the reason is that it is a lower voltage, slightly lower speed device. If that theory holds, there could be a similar issue with 2SC3503 as it is a high voltage, high speed device. I would love to see Ft vs current traces for it at 5V Vce. And kudos to OnSemi for including curves for both 5V and 10V for their power devices. Nobody else seems to.
Another question that I have, is whether there is also a similar massive Ft droop in the quasi-saturation region that we see with the high-speed NPN power devices like the MJL4281. Interestingly, the MJL3281A doesn't suffer as greatly from this effect. Perhaps the reason is that it is a lower voltage, slightly lower speed device. If that theory holds, there could be a similar issue with 2SC3503 as it is a high voltage, high speed device. I would love to see Ft vs current traces for it at 5V Vce. And kudos to OnSemi for including curves for both 5V and 10V for their power devices. Nobody else seems to.
Last edited:
The curve of the silver face looks good. I hope the complement is also good. I'm going to get 546B/556B from NXP, made in china, I'm hoping they are as good as the ones you've measured. There could be a difference between older Philips ones and NXP, does yours look like the attached one? @stormsonic
Thanks,
Alex
Attachments
Thanks.
This is such an excellent thread. The other one by Mark is excellent as well:
https://www.diyaudio.com/community/...-early-voltage-va-of-some-pnps-at-8ma.298404/
My Tek 454A has sockets for all transistors! Maybe I will make my amp protoype like that.
BTW, I am looking into making a simple curve tracer to use with the scope. I will use a fixed base current (adjustable), and a triangle waveform for the collector voltage and X axis of the scope in X-Y mode, this way I will have a return trace that may help spot transistors with thermal memory problems.
This is such an excellent thread. The other one by Mark is excellent as well:
https://www.diyaudio.com/community/...-early-voltage-va-of-some-pnps-at-8ma.298404/
My Tek 454A has sockets for all transistors! Maybe I will make my amp protoype like that.
BTW, I am looking into making a simple curve tracer to use with the scope. I will use a fixed base current (adjustable), and a triangle waveform for the collector voltage and X axis of the scope in X-Y mode, this way I will have a return trace that may help spot transistors with thermal memory problems.