accidental differential pairs

Yikes, NST30010MXV6T1G is matched to 2 mV and costs 22 cents! Looks like two die in the package.

According to my experience, 10mV ballpark.

Vladimir Vassilevsky DSP and Mixed Signal Designs

Hmmm, that would be a problem. But that cute little Onsemi pair guarantees 2 mv, good enough.

Here's the idea: use an analog multiply to compute the power dissipation of a mosfet, and then simulate the realtime junction temperature, as a fast, smart protection shutdown.

I've done this in software before, simulating realtime junction temp, but one might not always have a processor handy to do the math.

From a year+ ago thread initiated by Harry D:

Message-ID: Subject: Analog Squarer {Y=X^2} 1 Quardrant From: Harry D

I responded...

Then elaborated...

My 40 year old automotive chip design tricks are coming back to me. Here's how to do it without an OpAmp; and the added transistors don't have to match those in the MAT-04....

They are PNP arrays available from Analog Devices, but cheaper ones can be found from THAT Corp. ...Jim Thompson

You may also have a look at NXP's PMP5201, which is also matched to 2 mV (max) and same pricing (12 to 46 cents on Digihey depending on quantity).

It's also a two die/package component, which is not the ultimate solution with regard to temperature dependency, but one can hardly complain at this price. In the smallest package it probably isn't much of an issue anyway.

Regards, Klaus

Doesn't putting in small emitter resistors improve that?

If your bandwidths work out, make an oscillator whose frequency is proportional to one variable and whose on-time is proportional to another. Then LPF the output.

It only works for one quadrant, but you only need one quadrant.

(I've never tried this -- thought about it, but never actually put pencil to paper, much less solder to components).

It screws with the logarithmic equation, which is how the multiplication comes about in the first place. ...Jim Thompson

"Off a reel" is no guarantee that the devices are from the same wafer, which is about the only way that they would "match" without testing

If you must use discrete's, take a big batch and bin according to Vbe.

...Jim Thompson

The ON data sheet has extensive and somewhat confusing specs of cross-die thermal coupling. I'm guessing that heating one die raises its temp by about 150 K/W above the other one.

Here's an NEC dual-chip transistor

The separate-chip duals track ambient well but any local heating couples poorly between chips. So I should keep the power dissipation down on my diffamp.

That's similar to a classic PWM multiplier... make one signal into a PWM and use that to analog chop the other, then lowpass. They were used in things like analog flowmeters, and got impressive (like 0.02%) accuracy. But it is fairly complex compared to my little thing, and I don't need high accuracy for a mosfet shutdown.

D'oh. I would have known that, if I'd only bothered to think.

(;-) ...Jim Thompson

s 2 mv,

of a

mart

one

if maxim doesn't scare you too much, max4210/max4211 ?

-Lasse

I don't get how your circuit multiplies voltage and current. Normally I would expect to see exp(ln(voltage) + ln(current)).

Is this meant to operate only in switched mode, so Vd is either HI or LO? Is there some load connected to Vd that is not shown?

Bob

poorly

Yup. You don't need any speed, and anyway since the RC time constant has to be similar to the dominant thermal TC of the MOSFET die, keeping the current down helps reduce the capacitor size.

Cheers

Phil Hobbs

I spent a couple of weeks at Kent Instruments in Luton in 1974 putting just such a system together, and didn't have much trouble getting 0.1% accuracy . Since then Analog Devices took to selling highly linear voltage-to-frequenc y converters - presumably for the slow A/D converter market - which stay hi ghly linear even at very low oscillation rates, which is nice when you have to cope with a large dynamic range.

My switches were a pair of saturating transistors in sequence, with the sec ond transistor inverted for minimum off-set voltage - slow as a wet week, b ut adequate for industrial control. Somebody else's idea, and I got stuck w ith showing that it worked.

such a system together, and didn't have much trouble getting 0.1% accuracy.

converters - presumably for the slow A/D converter market - which stay highly linear even at very low oscillation rates, which is nice when you have to cope with a large dynamic range.

transistor inverted for minimum off-set voltage - slow as a wet week, but adequate for industrial control. Somebody else's idea, and I got stuck with showing that it worked.

does that bother you when some one else idea works?

Jamie

poorly

Plus, I can run them at the zero-incremental-dissipation bias point, the old scope vertical amp trick.