Curve Tracer

I'd like to see pulsed avalanche behavior of mosfets, preferably to destruction.

C-V curves would be great too.

Stuff that's not on the data sheet.

The thing about an analog CRT scope is that it isn't quantitative and forgets instantly. A digital scope would fix that. A curve tracer should be able to trigger the scope and do short single-shot events.

I've had quite good thermal-matching results with T0-220 power transistors used in low-current circuits and mounted on an aluminium bar. The change of junction temperature with ambient temperature is slow due to the thermal mass of the bar and its small surface area. The temperature difference between the junctions is small, even when they are carrying different currents, because of the large chip area and the low thermal resistnace between the junctions and the bar.

I've also bound two 'D'-packaged transistors together, flat sides touching, with thermal compound between them. The ZTX series are good in this respect because they have large junction areas but very small 'D' packages.

With diodes in a logarithming circuit, I use two pairs of 1N4148s (to average-out any discrepancies) mounted vertically on a board in 'diagonal' formation . (Viewed from above the 'North' and 'South' form one pair and the 'East' and 'West' form the second pair) The pairs are joined at their top ends; their bottom ends go through holes in the board in a square pattern.

A heatshrink sleeve full of heatsink compound is then shrunk over them so that the ccompoud penetrates and fills the voids between them. This results in a solid 4-wired 'component' with closely-thermally-matched diodes.

The CA3046 / LM3046 were cheap until they stopped selling them. Fortunately I have several tubes of them left.

What makes you think those are monolithic? I think they are separate chips, but measured to have similar parameters, like the BCM846BS.

The thermal coupling between the chips will be poor, so they will no longer be matched if the dissipation is not the same between them. You could cascode a current mirror to fix that, but if you are trying to make an exponentiator (as used in analogue synth VCOs) then you are stuffed, because you need to operate the two transistors at different currents, that being the whole point of the circuit.

You will know if they are monolithic because it will have a pin called "substrate" or a note saying one of the pins is the substrate, and there will be a spec pointing out that the voltage between the two devices must be kept below some lowish value.

They are two similar chips, not monolithic. Thermals will be awful.

You would know if they were monolithic if they did have a substrate pin.

The fact that they haven't got one isn't proof that they aren't monolithic. A stronger argument is that they haven't put any limits on device-to-device voltages.

My reason for thinking that they were monolithic was the 2mV worst case and the 1mVB typical difference in Vbe at 2mA.

Monolithic does seem to offer the cheapest route to get that.

Prove it.

They may be two separate close-to-identical chips. There isn't room in the package to mount them far apart, and the chip to chip thermal resistance can't be large, and has to be much smaller than the package to ambient thermal resistance, which is 328C/Watt.

Thermals won't be awful. Somebody who doesn't know about Wilson current mirrors isn't going to be a particularly reliable source of information about that kind of subject.

Interdigitated monolithic is hard to beat for thermal matching but side-by-side devices on the same subtrate aren't going to be any better than devices on separate substrates if the substrates are mounted back-to-back.

Renesas Still seems to make the much faster HFA3096

Phil Hobbs doesn't like them - the base resistance seems to be a bit high for his kind of work.

I seem to remember someone here has an xray machine which could answer the question?

It’s a FAQ that we’ve gone through many times, including my doing a bit of math on the datasheet for the BCV61 current mirror that used its thermal runaway spec to estimate the die-to-die thermal resistance.

Turns out to be about the same as the die-to-ambient, 300-500 K/W.

They really aren’t monolithic.

Cheers

Phil Hobbs There

Thermal runaway is all about die-to-ambient. How does it tell you anything about the die-to-die thermal resistance?

That doesn't necessarily follow.

Right. Those dual-chip things are not much better thermally then two SOT-23s mounted close on a board.

I make current mirrors now and then, with an opamp and a mosfet. Or, sometimes, just an opamp. That's way more accurate than the transistor versions.

Hey, Dr Hobbs, what is the attraction that PhDs seem to have for current mirrors?

Send me that On part and I'll x-ray it.

On 2025-02-07 10:52, john larkin wrote: > On Fri, 7 Feb 2025 14:18:48 -0000 (UTC), Phil Hobbs> snipped-for-privacy@electrooptical.net wrote: >

The basic reason is to be able to use current-mode circuitry.

I use current mirrors in some designs, e.g. my best laser noise canceller, where I need replicas of the currents in the main diff pair to cancel the effects of R_ee' and R_bb'.

My fave is the switcheroo Wilson, which is a double-cascode design (two transistors per side), where the diode-connected transistors are in opposite corners, like this:

O OUT O IN | | | | | V *------* V | | | \| | |/ |----*----| V| Q1 Q2 |V | | *------* | | Q3 | Q4 | \| | |/ |----*----| V| |V | | | | GND GND

This actually has three distinct feedback loops (two transdiodes and the big loop), and cancels beta error to leading order.

Since all of the transistors are operating at the same collector current, you can apply a couple of small emitter resistors and a dual op amp to null out the V_BE offsets. The dissipation in Q2-Q4 is generally small enough that the thermals don't hurt much. (The V_BE error of Q1, which potentially dissipates much more, gets taken out in the main loop--it appears between the collector of Q4 and the emitter of Q2.)

Larger amounts of emitter degeneration will also equalize the currents in a restricted current range, at the price of trashing the bandwidth (nobody talks about that).

There are a few devices, AFAIK only SiGe, that have high enough Early voltages that you can equalize the dissipation in the two sides by cascoding them and dorking one of the V_CEs so that I_C*V_CE is the same.

That's magic--you can basically reproduce the performance of a 40 GHz monolithic diff pair using two separate transistors.

(All of which takes several extra parts, but you can do some pretty stunning things, for sufficiently permissive definitions of 'stunning'.)

For lower performance things, a bunch of outfits will make semicustom ICs on an old-fashioned planar NPN process (300 MHz-ish) for a few $k of NRE. I've often thought of making a laser noise canceller ASIC, but the performance isn't good enough--betas of 50, no fast PNPs, and so on.

Cheers

Phil Hobbs

And lot slower and more expensive,

Bob Widlar didn't have a Ph.D. Neither did Jim Thompson. Both used the a lot.

I know about them and I'm glad they still make them, but they are not cheap and are single-source.

Anyone know anything about this company?

They will be quite a bit better. You might come close if you put your SOT-23 devices close together and covered them with a blob of thermally conductive elastomer, but two devices in the same package will end up quite a bit closer together, and the packaging material will probably have quite a bit better thermal conductivity than any material you could risk blobbing onto a board.

And lot slower and more expensive,

Bob Widlar didn't have a Ph.D. Neither did Jim Thompson. Both used them a lot. Barry Gilbert did have a Ph.D. and he was even more enthusiastic.

Interesting, in Latvia. They are agnostic, or maybe random, about using commas or decimal points as digit separators.

Kinda grim on Street View, typical Russia/former satellite sort of look.

Australia, America and the UK use the full stop as a digital separator. The Dutch and the Germans use a comma. I don't know what the Russians do. I haven't worked there.

No surprise there.