Best method to temp compensate cascaded BJTs?

[hlab]

Hello everyone. I'm currently working on designing a geiger counter that only uses discrete components as a challenge, and I would like to get some pointers on the best avenue for Vbe temperature compensation. My design is moderately complex and the high voltage regulation loop involves multiple bipolars which are cascaded, so I can't afford to put a widlar or wilson current mirror topology to properly compensate each stage. I have been toying around with using NTC thermistor networks to achieve this, and I wrote a python script that finds the optimal resistor values. This method definitely improves the temperature dependency, but it's my understanding that the exact tempco of a bipolar can't be exactly matched with this type of network due do the thermistor exponential curve being different. There is probably a way to further refine this curve by incorporating multiple NTC thermistors, but this seems too complicated for my liking. Are there any alternate ways of achieving this that I'm not aware of? I'd prefer to get the voltage regulation stable within +/- 5% over the -40C to +65C range.

Normally, I would just use zeners instead of Vbe as my reference source, but there are no commonly available zeners that are viable for the ~3.5V system supply. For the HV regulation, I could probably use a string of zeners there, but I'm unsure of the static accuracy at such low currents, since I'd like to keep the HV load to a minimum.

Any ideas? Thanks!

[David Hess]

Differential pairs temperature compensate themselves and were commonly used in high performance regulators before operational amplifiers became available.

For a current source or sink, an old style red LED has about the same temperature coefficient as a Vbe junction, so can be used as a 2 volt zener diode when forward biased to make a temperature compensated current source or sink.

There are probably some discrete Tektronix designs from the late 1960s and early 1970s, before operational amplifiers became available, which could give you ideas for regulating high voltages.

[langwadt]

Differential pairs temperature compensate themselves and were commonly used in high performance regulators before operational amplifiers became available.

For a current source or sink, an old style red LED has about the same temperature coefficient as a Vbe junction, so can be used as a 2 volt zener diode when forward biased to make a temperature compensated current source or sink.

There are probably some discrete Tektronix designs from the late 1960s and early 1970s, before operational amplifiers became available, which could give you ideas for regulating high voltages.

https://www.dropbox.com/s/d4ntmq7fdzah69a/LED_Isrc_data.JPG?raw=1

[hlab]

Never considered LEDs, although now that makes sense. Is the Vf tempco consistent for AlGaInP? I'd like to avoid having to manually bin the tempco.

[David Hess]

Never considered LEDs, although now that makes sense. Is the Vf tempco consistent for AlGaInP? I'd like to avoid having to manually bin the tempco.

The temperature coefficient is consistent, but I am not sure which red LED technology matched best.  There is considerable variation in processes which produce red LEDs.

The alternative would be to use a low voltage bandgap reference, 1.2 or 2.5 volts, and then a pair of the same transistor with one configured as a diode to match the Vbe temperature coefficient, but a red LED is pretty good.  I have seen it done both ways.

[hlab]

Alright well I did some more playing around, and found some math bugs with my original python script. But it appears that I can get less than 3% total deviation over the -40C - +65C range with the values determined with my new calculator. I think I'd rather go this route instead of experimenting with a bunch of different red diodes. Here's a link to it in case this comes in handy for someone else.

https://godbolt.org/z/Toz6c8nG4

[moffy]

Never considered LEDs, although now that makes sense. Is the Vf tempco consistent for AlGaInP? I'd like to avoid having to manually bin the tempco.

The temperature coefficient is consistent, but I am not sure which red LED technology matched best.  There is considerable variation in processes which produce red LEDs.

Here is some data on LEDs voltage temperature coefficient from: https://proaudiodesignforum.com/images/pdf/Junction_Temperature_LED_Tempco.pdf