Matching BJTs to replace discrete transistor pair IC

[bd139]

Got an interesting little power supply board from the 1970s in the junk box. It was extracted from some industrial kit. Fundamentally it’s a comparator with amplifier and pass transistor driver that drives a 2n3055. In the interests of learning, I like to reverse engineer and fix these sorts of things.

Well the differential amplifier working as the voltage comparator is knackered. It’s an unmarked NPN dual BJT based on the fact that one half looks like a BJT and the other half is short.

Turns out dual transistors in the original TO99 package are expensive. The current requirements are pretty trivial for this so I’m intending to repacling it with two discrete transistors, selected from a large bag of Motorola 2n2222a’s.

What do I need to do to ensure a reasonable match and thermal stability between both transistors?

[cat87]

Well, in general to match a BJT you'd need a curve tracer. Matching Betas is only part of the story. Their characteristic curves should also match as well as the VBEs.

However, there are some other options you can look at. There's the "multiple matched BJTs in an IC" possibility such as the MMPQ3904 and NMT2222. "Matched" is a strong word for some of today's ICs, but it should get you closer to what you want than making your own  curve tracer.

Now, if you do want to build yourself a tracer, you can look up some of W2AEWs recent videos. where he makes basically what is the backbone of a curve tracer. There's also one from ELV, that is all analog. and it goes up to 100V Vce. If you want, PM me and I'll give you the schematics for this

[Zero999]

Matched pairs can be purchased relatively cheaply but they're all SMT packages. A small adaptor board could be made by gluing the device to some perf' board, soldering wires to the board and connecting them to the device with #28 gauge tinned copper wire.

http://www.mouser.com/ds/2/302/BCM847BV_BS_DS-188878.pdf
https://www.mouser.co.uk/ProductDetail/Nexperia/BCM847BS115/?qs=sGAEpiMZZMvKM5ialpXrmjQi8YuBF0T0

[bd139]

Thanks both - much appreciated.

SMD is the chosen solution. Had no idea that sort of thing existed. The power dissipation of these is very low so should be fine. BCM847DS it is, as that's got good enough matching and is easy enough to handle. I'll glue it to the top of the board and run solid wire to the pins.

Will order some from Farnell. £0.159 each is much better than £10 a pop for a TO99 equivalent!

[Kleinstein]

For a differential amplifier it is usually VBE matching that matters. To find matching ones, just measure VBE of something like 5 units for a suitable test current (this might be the diode test of a DMM, with base connected to collector).

Besides matching, thermal coupling could be important. There are cheap SMD type double transistors. There might be space for a small adapter board. For tHT parts one can use 2 TO92 parts with a piece of shrink tubing / glue for close coupling. Coupling TO 5 cases is more tricky as the case is collector, so better get some 2N3904.

[bd139]

Thanks for that. I've gone with the dual packaged SMD part for now. It's not going to arrive until Monday so I'll dig out my 2n3904 bag to play with in the meantime

[orolo]

To thermal match the TO-92 transistors, you can try the Isis and Omega clips. I have several BCM847DS in the hoard, but never got the time to use them. Please, share your experiences with the part as diff amp  .

[bd139]

Interesting read - thanks

I've seen those clips used by Philips in their oscilloscopes before as well.

[David Hess]

Manually matching a pair of TO-92 parts will usually be sufficient in non-precision applications especially since most surface mount duals are separate dies mounted in one package and many are not matched at all.

If better temperature tracking is required than affixing the TO-92 flats together would provide, then I might bolt a pair of TO-225 KSC3503s together with an insulator to get a much lower junction to junction thermal resistance.  S or W clips would be nice but even Bob Pease had to have someone fabricate them from scratch.

[bd139]

Good plan.

What I've decided to do is stick two 2n3904's together (thermal paste then heatshrink them) and hit the pair with a hairdryer (through a hole in some bubble wrap bag for insulation) and see what happens to the output voltage. Then I'll do the same with the BCM847 pair.

Eventually I'll probably have design an entire replacement power supply from scratch

[danadak]

Use PSOC to sweep one of its internal current sources, drive base,
and use internal A/D to measure Vbe, Vc, collect a few DUT tests,
and have PSOC code tell you which ones are best matched.

Note beta can be calculated from measurements, additionally you could
use one of the VDACs and control Vce to add to capability. Many possibilities.


Regards, Dana.

[bd139]

Interesting idea.

What's a good entry point to PSoC ecosystem out of interest? I've had my eye on that for a couple of years.

[orolo]

A bit of solar padding and a touch of the soldering iron. Not a very fine work, but took me five minutes. The transistors fit tightly. If two sacrificial transistors are used while soldering, I think it can be made much more regular.

[bd139]

Looks good

[danadak]

The IDE is free -

http://www.cypress.com/products/psoc-creator-integrated-design-environment-ide

Board, high end part on it, basic -

http://www.cypress.com/documentation/development-kitsboards/cy8ckit-059-psoc-5lp-prototyping-kit-onboard-programmer-and?source=search&cat=software_tools

Using this gives you almost all the high end capabilities, then you can drop
back to low end for final design if needed and capabilities match up.


Low end, but Arduino foot print -

http://www.cypress.com/documentation/development-kitsboards/cy8ckit-042-psoc-4-pioneer-kit?source=search&cat=software_tools

http://www.cypress.com/documentation/development-kitsboards/psoc-4-cy8ckit-049-4xxx-prototyping-kits?source=search&cat=software_tools



Regards, Dana.

[bd139]

Thanks for the links.  IDE looks like custom Visual Studio.

Have just ordered a 5LP kit. Very surprised at how little they cost!

Local supplier for me: http://uk.rs-online.com/web/p/processor-microcontroller-development-kits/1244192/

[Zero999]

Manually matching a pair of TO-92 parts will usually be sufficient in non-precision applications especially since most surface mount duals are separate dies mounted in one package and many are not matched at all.

If better temperature tracking is required than affixing the TO-92 flats together would provide, then I might bolt a pair of TO-225 KSC3503s together with an insulator to get a much lower junction to junction thermal resistance.  S or W clips would be nice but even Bob Pease had to have someone fabricate them from scratch.

The BCM847DS is fairly well matched, because the dies are from the same wafer. V_BE 2mV or better, ratio of betas 0.9 or better. The thermal coupling will also be better than two discrete transistor.
http://www.mouser.com/ds/2/302/BCM847BV_BS_DS-188878.pdf

[David Hess]

The BCM847DS is fairly well matched, because the dies are from the same wafer. V_BE 2mV or better, ratio of betas 0.9 or better.

The problem is with dual parts which give no specifications for matching so you have to check the datasheet carefully.

[Zero999]

The BCM847DS is fairly well matched, because the dies are from the same wafer. V_BE 2mV or better, ratio of betas 0.9 or better.

The problem is with dual parts which give no specifications for matching so you have to check the datasheet carefully.

I agree. That's why I suggested a part with the matching specified. I wouldn't recommend using a part with no matching specifications given, although wouldn't be surprised if the matching is fairly good, as it's highly likely the dies will have come from the same wafer.

[bd139]

I just found a CA3096 and a CA3028 in the IC box to add to the things to try.

[jaromir]

Another option - LM3046 is NRND, but still available online for acceptable price http://www.mouser.sk/Search/Refine.aspx?Keyword=lm3046

[SeanB]

Simplest is to use that transistor tester in the cheapie multimeter, and simply take a bag of transistors and bin them for gain, being sure to handle all the devices as little as possible so they do not heat up, and choose 2 in the middle of the spread. Generally good enough for this application, and in general the manufacturer of the dual devices mostly used 2 chips from the die next to each other in the assembly, with the premise that being next to each other on the wafer they probably were pretty closely matched anyway.

A lot of equipment manufacturers would simply buy a lot of transistors, put them all in a jig that held them at a constant temperature, and get a few parameters out of them after a heat cycle to even them out, then simply put them in bins selected for the parameters desired. That is why you often see in older HP and TEK, along with Phillips, the transistors with a paint mark on the case, with the parts that were gain matched, typically differential DC amplifiers, all using a same colour transistor per stage, and with all being the same base type. Often they were selected for either gain within a narrow range, low leakage or for voltage rating, out of the big pool of raw devices.