more AlN insulators

You're being silly now. Measure volts and amps into the heater and multiply.

Like, a thermocouple meter?

is TWO expenses and two error

Hang on to that resume for a while, OK?

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

If the air is kind enough to sit still. You can almost force this with flu ffy insulation - silica wool or the like that is mostly air with enough sol id content to slow down convection, though it can't stop it.

ed

of

And work out how much of the heater power that is coming out is not going t hrough the AlN insulator.

Even Tulane should have taught you to think about that. The back of a heat generator is usually cooled by convention, which is non-linear - Newton's L aw of Cooling says that heat loss is proportional to the square of the temp erature, which is rarely accurately true, but close enough for some applica tions.

Fluffy insulation to slow down the convection helps, but isn't perfect.

With an evacuated space, you only have to worry about radiation, which incr eases as the fourth power of the temperature of the radiating surface.

Thermocouples are horrible. Calibrated - interchangeable thermistors - are way better, and platinum resistors are better still, though much less sensi tive.

nst a known material, means the calibration is the single step from a preci se measure to an accurate one.

budget.

If you want a job with John Larkin, you have to flatter him, rather than po int out that he's being as shallow,sloppy and superficial as ever.

--
Bill Sloman, Sydney

That is indeed a logical method and conceptually the simplest. But how do you separate out the thermal resistance of the interfaces from the bulk thermal resistance of the insulator under test? Simply removing the DUT also removes one of the interfaces.

There are a lot of ASTM standards on measuring conductivity and a lot of commercial instruments on the market which make measurements using those standard methods, as well as labs that will measure your sample for a fee. If you just wanted a single measurement a lab would be the way to go. Anyone wanting to build an accurate instrument would obviously take a look at prior art. But for your quick and dirty relative comparison measurement I think you are on the right track, and any concerns about air convection losses are easily eliminated with a bit of styrofoam or other insulation around the hot end of your tester. If you could get samples of the AlN in two different thicknesses you could even extract a bulk conductivity measure. Stacking introduces an additional different interface, not quite the same as a thicker material, but possibly an interesting data point. If I were doing it I might use copper blocks the same size as the sample, solder or braze fine thermocouple wires into small grooves milled into the faces of the blocks (no need for a junction between wires as long as they both contact the block), lap the faces flat to a microinch or so, and calibrate the thermocouples.

Thermocouples are easily good to better than 0.1 F over the range of operating temperatures of semiconductors if they are calibrated, they only have poor accuracy if you insist on interchangeability between wire production lots without any calibration and/or use lousy instrumentation. Calibration standards good to better than 0.1 F absolute accuracy have been available for well over a century, starting around 1905 in this country from what is now NIST, earlier in Europe, I have no idea why some people think temperature measurement is difficult. The temperatures of the phase transitions used to define our temperature measurement scales have not changed, nor do I expect them to change anytime soon :-). Water is an obvious choice, my other favorite is Tin which freezes at 231.9681 C, 449.5426 F, a very convenient full scale cal point for my usual uses and readily available as a temp cal standard with heater and thermowell for around 2k last time I looked.

Two thermal interfaces (as, thin copper foil with grease on both sides) would fix that problem. But I only need ballpark measurements, specifically to compare AlN insulators from two different sources.

I could easily measure the theta of the test probe to ambient, but I expect that to be high compared to the AlN.

If you could get

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

The benefit of the AC method is, in part, that it looks at thermocouple accuracy differentially. If a 400C range is covered to 2C accuracy, that's pretty good, better than 1%; the curve, though, is smooth, both the standard curve, AND the real (calibration) curve, too, so we can expand the difference in Taylor-series fashion, deltaT = const1 + const2*T * ...

Now consider a small-temperature oscillation, T never drifting more than a single degree (less than 1% of the 400C curve range). That deltaT calibration offset can be truncated after two terms, and the first term (constant offset) cancels (we're doing AC excitation). The second term, const2 * T, must be less than or equal to one 0.01 per C, or it would exceed the total allowed calibration-versus-standard-curve 2C deviation. So, we get better than

1% accuracy in the AC determination of thermal conductivity even with the uncalibrated thermocouple.

Numbers don't tell the whole story; there's some calculus required to figure out what the equations really mean. Thermocouple accuracy is a removable error even when you don't calibrate it out.

Who's being silly? That gives you total heat output, NOT the heat flow through the surface of interest. It measures A powert, not THE power of interest. Everyone with a teapot should know that ALL the surfaces shed heat. The TEC isn't like a teapot, so it does have an advantage here.

The two paths for heat to get out of the power resistor are through the AlN and through air; radiation would be insignificant.

The AlN conducts heat about 7000 times better than air. If I care about heater-to-ambient theta, that's easiy measured too.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

But you won;t bother ...

AlN conducts heat about 7000 times better than still air. Real air convects, and the bigger the heat difference the faster the convection currents, and the thinner the boundary layer.

If you put silica wool behind the resistor, the convention currents are slower and reliably laminar, but they are still there.

Power resistors are rated to operate with case temperatures a few hundred degrees above ambient, so you should think about what you are doing, even if you are dedicated to quick and dirty solutions.

--
Bill Sloman, Sydney

That's odd; my mental image doesn't have a resistor in air against a surface, it has a resistor connected to wires, clamped against a surface. Where does the clamping force come from? Why do you think the wires don't get hot and/or aren't conductive?

A TO-220 can dissipate a watt or so in free air, if your intended power is a hundred times that, and the cooling is so efficient that it doesn't rise to the one-watt temperature, then air can be negligible. As for the screw through the hole into the cold side of the heatsink, maybe not so negligible. One hopes thin electrical wiring can be a minor heat loss, but I've been known to use a hulking 30A 5V supply for heaters.

I posted this above; maybe you missed it.

The heater is a Caddock TO-220 power resistor. A couple feet of twisted #30 wire-wrap wire won't conduct significant heat. I can always run the rig in free air to quantify its heat loss, which I expect will be ballpark 1% of the conduction into the AlN.

Something like 20 watts into 0.05 square inches of AlN should make a useful delta-T. I suppose I could also measure thermal resistance along the long axis of the insulator, instead of side to side, but I want to characterize gap-pads too.

One thing that's interesting about SED is how many people get serious about elaborate, expensive, hard ways to do things... things that often just can't work quantitatively. I've known a lot of engineers who do that, and enjoy it until the project fails.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

True, but with two thermocouples from the same wire lot (cut one in half if necessary) and a single good meter switched between them you also get good differential accuracy even without calibrating.

There are some nice commercial instruments that use the AC method using a probe with a single element heater-RTD, which puts some watt-seconds into the sample and heater combined thermal mass then measures the turned off heater resistance while the sample and heater cool. They need and come with calibration samples. But JL will be done with the simple method long before he could build one of these! I expect to see measurements posted tomorrow :-).

If I use two tc's, I can set my heater power to zero and note any temp differential. But there are fundamental reasons that two type K thermocouples will match very well near room temp.

Maybe not. I'm furiously scribbling schematics of something else just now, trying to keep ahead of my layout guy.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Any delta-T, no matter how small, can be useful. Gain is cheap.

Another thing you can learn: some people are serious about measurement errors, and do elaborate things to understand and avoid 'em. Those are the folks who can tell you the temperature averages of past centuries, know their accuracies... and get hired on at NIST.

Yeah, the large hadron collider is an elaborate way to weigh the Higgs. But, no one knows a better one. The project didn't fail.

A TEC is maybe $20, more expensive than the transistor-packaged resistor, but the back-and-forth pumping of heat nulls a number of errors that you are minimizing (without calculating their effects). A LabJack or similar data collecting brick, maybe some drive power to the TEC, and a good thermal characterization is complete. Is that elaborate? Or expensive? It's a lashup most of us could build from things in the junk drawer...

The convection contribution is seriously nonlinear, and even thermal conduction is far from perfectly linear, especially in nonmetals. See the nice curves on P. 2 of

So for heat sink measurements, near enough is good enough.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC / Hobbs ElectroOptics 
Optics, Electro-optics, Photonics, Analog Electronics 
Briarcliff Manor NY 10510 

http://electrooptical.net 
http://hobbs-eo.com

I can run my probe in free air, at the same temp my AlN measurement was done, and see how much power that takes.

and even thermal

Nice. Transistor insulators (BeO, alumina) lose thermal conductivity above room temp; AlN probably does too. My baseplate will be water cooled, so that effect should be minor.

Gotta get Paulo to machine some copper for me.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com