2mV opamp

Would that be a silicon carbide resistor? Oven temperatures don't really work well with silicon, but SiC is another semiconductor entirely.

You won't find one. There are thousands of opamps that fit the bill. Most (P/Ns) will and it's probably easier to make a list of those that won't.

Some manufacturer's parts may. Other than CMOS parts, 2mv is a *lot* these days.

happens. ;-)

Cheaper, maybe but interchangeable thermistors are available matched to +/-0.2C and +/-0.1C

Linear Technology just advertised a universal temperature measuring chip that was claimed to offer higher precision, but I suspect only after calibration.

Back when I was interested, thermistors were the only option if you weren't going to calibrate. Making a well-stirred ice/water bath isn't that difficult, but production prefers to avoid it.

If not a proper three-terminal integrated circuit temperature sensor - the LM35 has been around for a while and the AD590 for even longer. Neither of them is all that precise until the individual example has been calibrated.

And George Herold reported an 'unmatched' batch of transistors had about that same scatter. You can't usefully check this kind of parameter on a hot CPU chip, nor use that much precision in cooling-fan drive decisions.

The CPU sensor is a diode-connected BJT. By running it alternately at

can be achieved without adjustment. Cheap and effective. You can do the same thing with a 2N4401 etc.

With about 10:1 ratio in currents you get about 1/10 the sensitivity of a diode junction, but that's still plenty (200uV/K).

Best regards, Spehro Pefhany

Since it's an "AC" signal (square wave), you can easily gain it up as well. ...Jim Thompson

There were no junctions involved, AFAIUI it was a bulk silicon

self-cleaning cycle.

As you can see here:

Currently they're using PTC resistive sensors with a fairly low tempco- certainly not a thermistor.

It's within about 5% of a Pt1000 DIN, but not bang-on.

Some kind of nickel or platinum RTD most likely. The HVAC guys have some crummy base-metal RTDs they use as well... when you have enough volume you can shave pennies like that and eschew standardization.

Best regards, Spehro Pefhany

that > same scatter.

Sadly, a narrow scatter about an unspecified Vbe isn't of much practical us e. You've still got to test all the parts to make sure that they are from t he same batch.

that

You can check the Vbe of a poorly specified sensing diode on at a well-defi ned temperature on an unpowered CPU chip. The fact that the chip is going t o get hotter when the CPU is powered up is precisely what you want to measu re.

The precision you want in making cooling fan drive decisions is related to your enthusiasm for pushing the chip close to it's rated limits. the more p recisely you know the temperature of the chip (or at least the measuring di ode's corner of it) the closer you can get to the absolute maximum rating o f the chip.

And I was arguing with the claim that silicon temperature sensors are no le ss accurate than thermistors, not making any claims about the accuracy need ed in a CPU protection sensor.

It's VERY useful!

A diced wafer of tiny transistors ought to have the benefits of doping uniformity and geometry matching both. An 8-inch wafer dices up into twelve thousand MMBT101 transistors, so you can get a reel of a thousand transistors, and probably all match.

To wait for an assembly line to fill one of the matched-set bins with a similar quantity of tested thermistors, would be quite a production bottleneck! If the thermistors are manufactured to 2% range, there's gotta be a sorting step with sixty bins behind the scenes, to get a 0.1C match guarantee. It could take 50k units produced and tested before 1k of 'em match to 0.1 degree C.

Either way, you have to test all the units you use. The bin-array way, though, you have to test 49k other units as well...

Slowman is clueless, as usual. I'm not going to tell him what fundamental property he missed... just let him stew in his own excrement >:-} ...Jim Thompson

Thinfilm platinum and nickel RTDs are very good. They are laser trimmed, not bin selected.

You can get surface-mount 0805 and 1206 platinum RTDs. Zetex makes a neat SOT-23 nickel RTD; pin 3 is the isolated thermal flag. I used one in my mosfet SOAR protection circuit, the one that runs a realtime analog simulation of junction temperature.

icon

d to > > > > +/-0.2C and +/-0.1C

out > > > that same scatter.

l use. You've still got to test all the parts to make sure that they are fr om the same batch.

lve

You are out of your mind. The diodes on a processor chip are fine temperatu re sensors, but you have to calibrate them. The diodes in each of the proce ssor chips on the same wafer will presumably have very similar Vbe's, but y ou won't get 12,000 processor chips from an 8-inch wafer, and the people wh o set up the temperature control scheme for the processor chip aren't going to be able to rely on getting many processor chips from the same batch.

tta > be a sorting step with sixty bins

ts

Unless someone has worked out a scheme to trim them.

We aren't manufacturers testing the products coming off a production line, but customers buying parts. Who cares how the manufacturers get the product - the only interesting questions are whether we can afford them and whethe r they'll keep on selling them.

licon

ed to +/-0.2C and +/-0.1C

bout that > same scatter.

al use. You've still got to test all the parts to make sure that they are f rom the same batch.

elve

stors,

similar quantity of

rmistors

bins

its produced

though,

Platinum and nickel resistive sensors can be quite good. You end up with a much smaller volts per degree signal than you get with thermistors or silic on diodes, to the point where adventitious thermocouple voltages can get to be a problem, so you might need AC excitation and synchronous demodulation extract a useful signal, but John Larkin leaves that kind of detail to his minions.

Jim-out-of-touch-with-reality-Thompson hasn't got a clue what the discussion is about or the "fundamental property" that he thinks I've missed.

Maybe he'll expose his cluelessness by telling us what he thought that it might have been.

if one of the masks is slightly skewed you'll get two independant variables in your set

One can take two very closely "matched" diodes, place them in parallel, and STILL, one will end up in thermal runaway, and die, then the other.

There is no guarantee that ANY two die cut elements would be closely matched simply because they were originally cut from the same wafer.

You exhibit flawed logic.

Real world output response on each one is required. IOW individual calibration.

Doping levels are going to be much the same. Obviously, the actual dimensio ns of the diode on the die make a difference - for very small diodes even t he tolerance on the lithography could make a difference between notionally identically dimensioned parts.

George Herold had reported that transistors from the same batch had come in within a few tenths of a degree Celcius/Kelvin, which ties up with experie nce I've had with parts in production, where final test would complain abou t fitting the same value of select-on-test resistor for months on end, and demand a mod to fix on that value, until the next batch of parts came along .

You can't recognise flawed logic when you churn it out, and don't do any be tter on other peoples reasoning.

That's pretty much what I've been saying, for this particular situation. Qu ite why you think I'm saying anything else has to be put down to your capac ity to be AlwaysWrong.

If one of the masks was slightly skewed, most of the parts on the wafer wouldn't work.

On the Cambridge Instruments electron beam lithography machine, wafer position was set by two-axis laser interferometry.

l,

ions of the diode on the die make a difference - for very small diodes even the tolerance on the lithography could make a difference between notionall y identically dimensioned parts.

in within a few tenths of a degree Celcius/Kelvin, which ties up with exper ience I've had with parts in production, where final test would complain ab out fitting the same value of select-on-test resistor for months on end, an d demand a mod to fix on that value, until the next batch of parts came alo ng.

Hmm I'm not sure what you guys are talking about. I looked at the body diode in a FET recently.

I also use diode connected transistors as temp sensors. Single point calibration... but only after one of the batch was measured over the entire temperature range.

Here's a link to that.

For absolute calibration out of the box the 1:10 current difference scheme that Spehro refereed to is good to ~1% (3 degree C) with a diode connected transistor and assuming ideality factor = 1 (That's for the TIP31 that I used above... other transistors may be better or worse.)

George H.

better on other peoples reasoning.

Quite why you think I'm saying anything else has to be put down to your cap acity to be AlwaysWrong.