more AlN insulators

Get a Peltier junction (perhaps from Marlow) and a couple of thin platinum resistance thermometers on alumina

seems to be around the right size.

Sandwich your TO-200 AlN insulators between a (finned?)heat sink and the Pe ltier junction and measure the temperature of the top of the Peltier juncti on and at the bottom of the heat sink.

You will probably need second heat sink to go on the other side of the Pelt ier junction, and it would make sense to measure its temperature too (thoug h nowhere near as accurately). The heat driven through the Peltier junction and your insulator by a given amount of current can be calculated fairly a ccurately.

See appendix A to my 1996 paper.

Sloman A.W., Buggs P., Molloy J., and Stewart D. ?A microcontroller

-based driver to stabilise the temperature of an optical stage to 1mK in th e range 4C to 38C, using a Peltier heat pump and a thermistor sensor? ? Measurement Science and Technology, 7 1653-64 (1996)

Jim Willian's application notes on the Linear Technology Peltier driver sho uld have included something similar, and I upbraided him for not doing so, but he used the Stephen Hawking defense, which is that every additional equ ation in a text halves the number of readers. This is nonsense for applicat ion notes, but he saw himself as an author rather than a writer of applicat ion notes.

A big thin film power resistor, or a power transistor could be used to gene rate the heat flow, but you don't know which way the heat is flowing, which makes it harder to be precise.

Nah, it's easier than that. Two thermoelectric gizmos, same model, on either side of the thermometer plate/insulator/second-thermometer-plate sandwich. Clamp it at ambient temperature, with some heatsink goo, and reverse the thermoelectric power every minute for a few hours, paying attention to the slightly-out-of-phase thermal response.

Doesn't need to be anything fancy, you can calibrate with a chunk of quartz glass or just leave the sample out. It's only temperature AC fluctuations that matter, so thermistors or thermocouples are as good as fancy stuff.

It can run overnight, whie you sleep, if your data=handling skills are up to the analysis. Only critical thing is, the cycle time has to be long compared to the time-delay due to diffusion of heat through the sample (adobe walls eighteen inch thick take a week between cycle reversals).

Thermocouples are vile. The advantage of thin film platinum sensors on alumina is that they are thin and flat (which thermistors aren't).

There are surface mount thermistors which are pretty flat

but 0.8mm isn't all that thin, in context.

offers 0.45 mm - okay 0.55m high, worst case.

The problem - which I didn't think hard enough about when I put my post together - is that you really want to put your temperature sensors on the top and bottom faces of the alumina insulator.

That isn't a problem on the heat-sink side - you can mill a slot into the heat sink and bury the sensor on one side in that.

If you used a TO-220 transistor as the heat source on the other side, you might be able to mill a slot into the transistor package.

Peltier junctions really wouldn't accommodate this, but if you used two small Peltier junctions driven in parallel and mounted the other side sensor between them you could probably do as well.

You really do want to keep the thermal resistance of the heat-sink goop out of the measurement. Having an air-gap above the temperature sensors is probably enough to stop it being a problem.

You guys might do some math. AlN conducts heat almost as well as aluminum. The TO-220 insulator should have a side-to-side theta around

I'm thinking about something like this:

The heat flux per unit area has to be high to get a measurable delta-T across AlN or BeO.

I just received some nice 0.25" copper rod stock and some copper sheets from Amazon, and we have the Caddock TO-220 resistors, so I'll try to build this next week. Amazon has everything.

I should be able to dump, say, 15 watts into about 0.05 square inches, about 300 watts per square inch. That will make about 5 degrees C drop across my AlN insulator. I can zero things by measuring "nothing", just a smear of silicone grease.

Everything will have to be pretty flat.

The heat conduction will be so high, and the masses so low, that the thermal time constant will be seconds.

I need to measure some gap-pads too, thetas around 5 W/m-K. Good AlN should be around 170. The same rig should work for both. Gotta account for a little fringing, of course.

The gap-pad theta will be a radical function of compression.

But, in the AC method, only the statistical accumulation of temperature differences matter, and that gives you two extra decimal places (at the cost of takes-a-day test cycles). It's not about big signal, it's about big signal-to-noise, with phase-sensitive detection.

Absolute temperature discrimination and accuracy are irrelevant, unless your ambient room temperature has some kind of sub-hertz interfering signal (mine doesn't). If you stabilize the thermometer d(output)/dT to a couple of decimal places, the thermal resistance accuracy (after calibration with a known sample) is one percent. Two thermistors and bridge measure amplifier will cover any small-signal issue with meter resolution.

Aging of heatsink goo and warping of materials is a bigger error than such a test accuracy. So, too, is clamp pressure.

I use to-220 pnp's TIP32c as temp sensors. Diode connected the four different production dates I measured from Fairchild* all had the same V(t) curve, and I do a single point calibration. I could send you two. Or just use any to-220 diode, fet? Then grease and SS bolt.. with heater and heatsink on the ends.

My issue with thermal T measurement is that it's always off by about a factor of two or more. I wave my hands about surface resistance and pressure.

George H.

Right, thermal couples are nice for differential. As you say delta T will be small.

George H.

Sounds a little, well, compulsive to me. I can build this thing and take the measurements in a couple of hours. But run the numbers and I'll consider it.

LM35 comes in TO-220. We still have some in stock, from an old project. They are nice for measuring heat sink temps.

And people claim they know the temperature of an entire planet, to a fraction of a degree C, over hundreds of years. Of course, they don't.

I have some US-made and Chinese AlN insulators and I'd like to compare them. And I have some gap-pad stuff that is very soft and compressable, claimed 8 w/m-K. I really need to experiment with that, theta vs compression. That could be done at lower heat flux, so could be a simpler rig, but the fancy one would work too.

inum resistance thermometers on alumina

te sandwich.

So what? You have to put heat through it to get a temperature difference, t hen you have to measure the temperature difference - which clearly isn't go ing to be big.

You can measure temperature differences down to microdegrees with thermisto rs and platinum resistance sensors.

That set-up measures the thermal resistance across the device under test an d both the gaps between it and your lumps of copper.

You could at least drill wells in the copper and stick your temperature sen sors on the opposite faces of the device under test and leave an air-gap be tween them and the copper.

Sadly, it doesn't seem to sell clear thinking.

Sadly, those two "smears" of silicone grease have their own thermal resista nces, which you will be measuring, and adding to the thermal resistance of the AlN insulator. You will need to know that extra resistance when you use the parts, but if it is more than the thermal resistance of the AlN (as is quite likely) it rather wrecks the argument for spending money on fancy Al N insulators.

There's also the problem that no two smears of silicone grease are of exact ly the same thickness. Thermopads are at least consistent (if not all that thermally conductive).

And you are going to quantify how flat everything actually is?

rtz

ons

up to

pared

ls

What I was slow to remember was that Peltier junctions are also Seebeck jun ctions.

Use them most of the time to set up the heat flux, then turn off the drive current for a few milliseconds, and they will tell you the temperature diff erence between each face.

This doesn't deal with thermal resistance across the gap between the surfac e of the Peltier junction and the AlN insulating pad, but John doesn't see m to care about that anyway.

With a pair of Peltier junctions you could drive heat one way for a few sec onds - enough to get the Peltiers and the AlN insulator to their steady sta te - then reverse for the same number of seconds.

The backing thermal masses - decent heat sinks - would get warmer because o f the Joule heating in the Peltier junctions, but they'd get equally warm ( with a rather longer time constant because they would be bigger, and with a rather longer time constant because their thermal resistance to ambient ai r would be bigger) and wouldn't bias the measurements.

A bridge measurement (AC) on a capacitor can tell you its electrical resistance, with a small signal input. ESR 'tester' gizmos put a large signal in, and capture one peak to get the resistance.

The problem is, the bridge gets accurate results, while the excitation with a spike just gets the right order of magnitude (good enough for fault detection).

So, the idea is to make a thermal bridge with the insulator in one leg (and ratio the data with recorded test data of a known-thermal-conductivity sample which was the calibration, the other leg). Repeatability of thermal current generated by TEC modules should be plenty precise enough, it'll take a calibration easily.

I'd consider a zero-resistance sample to be thin alumnum foil with thermal grease, and known-resistance to be quartz with grease (high enough thermal resistivity to get a good nonzero indication). That sets the calibration curve. You can use any gain short of saturation on the thermal-difference-signal (another bridge measurement, potentially, using thermistors) with some confidence that it'll only have a narrow sensing frequency range in which it contributes any noise.

The transient-pulse-of-heat lasting a second or so has a 1 Hz bandwidth of noise on the small temperature measured. The .01 Hz signal of the AC method will, after a thousand seconds, have only a millihertz of noise bandwidth sampled. Overnight (16 hours, circa 50,000 seconds) the noise nanovolts per root hertz is a factor of a few hundred less than the transient-pulse measurement.

It's things like this that are the reason for AC measurement of milliohm resistors; the small signal is easy to get above the noise either by hitting it with kiloamps, or tickling with milliamps for a while.

John Larkin really hasn't got his head around the implications of deducing temperature from isotope ratios on ice cores. You do get some diffusion wit hin ice cores, so the temperatures do get averaged over periods approaching a hundred years or so, but that's good enough for most purposes.

The people that make the claim do know what they are talking about, and Joh n Larkin hasn't got a clue.

He seems to be equally clueless about what he needs to do to measure therma l resistance accurately, and just as unwilling to listen to anybody else's opinion.

Why not just grease?

If I know the watts going into my heater, and the resulting delta-t, I don't need any other calibration. A TEC produces an unknown (and very small) heat flow, so would need calibration. And signal averaging.

Sure, at the end of the day I assume you want to know how much heat you can get out of a TO-220 pac, so you might as well measure that from the start.

George H.

In order to compare with a sample inserted in the clamp, in series with two greased surfaces, it's better to have... two greased surfaces in all the calibration samples.

It'd also work to just have a few different thicknesses, but aluminum foil is (in this application) pretty close to nothing.

But, the 'watts going into the heater' will shed heat to things like convection currents caused by its temperature rise... so it's not necessarily as linear as (in the small-signal thermal drive) something that only wobbles a few millikelvin. The calibration is easy enough (just another determination), and it means you get ratios of conductivity with multiple materials, which I'd consider ideal.

If something warps out of good contact with heat, that's not strictly a thermal conductivity issue, and a temperature-elevated test will show that kind of nonlinear effect, too (which is good), but only if it's not the test heater plate that warps).

I'd think of using a constant-current drive to the TECs, in series. Don't know how they age, but a yearly recalibration wouldn't be hard. It isn't necessary to know the heat/cool power flow to get the results, calibration takes care of that.

ote:

te:

al

The trouble with lashing something up in a few hours is that it can take da ys to work out that measurements are fatally flawed, or worse - misleading.

with

and

the

d by > > TEC modules should be plenty precise enough, it'll take a calibrat ion

mal

Because both are well known not to be a zero thermal resistances, and neith er promises to be a reproducible low resistance. Thermopads don't offer as low a thermal resistance, but at least you can peel them off and put them b ack on and expect to get the same thermal resistance

Every heater has two sides. If you don't know what's going out the other si de, you don't know what is going in the direction which is generating your heat rise.

TEC's do have two sides, but you do have a better handle on the heat flows that you set up.

TEC's produce significant heat flows - my 1996 paper shows one pushing 9W o f heat into our thermostatted block, and pulling 3.5W out.

The heat transferred at each face is calculable from the manufacturers data - calibration of an individual junction would make the result of the calcu lation more accurate, but the time taken to melt a known mass of ice is a p retty accurate measure of heat output, and doesn't take long to do.

Making a layer of ice and weighing it would be easier, but the temperature at junction under the layer of ice would be less predictable.

The result - for each face - isn't going to be much less accurate than the total heat flow coming out of both faces of a power resistor (depending on what tolerance power resistor you go for) and the capacity to produce posit ive and negative heat flows is handy, if you are willing to invest enough t o get a signal you can process and average.

Right. What really matters is how it performs in the real application. Take a power transistor and make a little block with a hole for the thermocouple, and a hole for the clamping screw. Clamp the transistor to a heat sink with thermal paste, apply power and appropriate gate voltage, and compute power in and delta-T between the transistor block and the heat sink.

Then, put whatever items you want to test under the transistor and repeat. This is as close as you can ever get to the real conditions of cooling a power transistor, which I assume is what you are using these for.

Jon

That should be small, but is easily measured; just leave a gap between the heater probe and the DUT. The thermal conductivity ratio between AlN and air is about 7000:1.

Measuring watts, and measuring degrees C, sounds like a logical way to calculate watts per degree C.

I'll be cooling a TO-247 Cree SiC fet, and some Coilcraft inductors. I can probably use the TO-220 insulators for both. Gap-pads are good enough to cool the less crazy parts of the board.

The bottom line is cooling the parts, but I'd like to measure the AlN thetas for future reference.