I'm pondering self-heating errors in an LM335Z temperature sensor. The National data sheet gives the junction-to-ambient (still air) thermal resistance as 202 degC/W and the junction-to-case as 170 degC/W. A graph of the junction-to-air thermal resistance vs air velocity is also given which starts out at 202 degC/W for still air and drops down to about 70 degC/W at high velocities.
Can someone explain to me how the junction-to-air thermal resistance can be less than the junction-to-case value ?
It is certainly curious. However, note that the junction to case thermal reisitance is what you have to add onto the thermal resistance of a heat-sink to calculate the final junction temperature.
TO-92 packages are notoriously difficult to couple to TO-92 clip-on heat sinks - the shape of the plastic package isn't well-defined, and if you squeeze your metal heat-sink hard enough to force close contact over an extended area of the package, you break the package - so I guess that 170 degC/W includes a lot of case-to-heat-sink thermal resistance.
Fast-moving turbulent air might conceivably do better. The fact that the physically bigger and solid TO-92 packages does better than the empty metal TO-46 can at high air-velocities is persuasive.
An experiment might be in order - try sticking an LM335Z into a 6.35mm (0.25") hole drilled into a decent sized chunk of aluminium, after initially filling the hole with zinc-oxide loaded silicone grease, and see what sort of thermal resistance you get there,
Almost all the heat goes out the leads. Copper has a thermal conductivity of ~400 W/m/K, vs. more like 0.1 for plastic.
IC temperature sensors are just as crappy as can be for air temperature sensing, and not much better for anything else except possibly board temperature.
That makes a lot more sense than my speculations. It doesn't explain why the TO-92 package does better than the TO-46 at high air velocities
- presumably the rather wider thermal path through the plastic compensates (to some extent) for the high specific conductivity of copper, and since both of them are a great deal more thermal conductive than air (even a thin boundary layer, and the boundary layer at 1000 feet per minute - 11.4mph - is still a couple of millimetre thick) the differences in their thermal conductivities probably won't matter as much as all that.
For boundary layer thicknesses this paper looks as if it might be useful
formatting link
Everything is pretty crappy for air temperature measurement - the one time I did it seriously, I wound 8 metres of very thin platinum wire (
60 micron diameter) around hexagonal squirrel cage, giving me a resistance of 270R. The wire wasn't entirely stress-free, so it wouldn't have done for NBS, but it worked pretty well.
The LM35 is actually a pretty good temperature sensor - admittedly ten times noiseir than a platinum or thermistor resistive sensor, but it doesn't dissipate much power and it is very easy to use.
The TO-92 package actually does worse than the TO-46 at high air velocities - I wasn't reading the data sheet carefully enough. Curiously, the TO-92 package always has the shortest time constant, even though the thermal resistances cross over at about 800 feet per minute, which imples that the effective thermal mass of the TO-92 package declines with increasing air-flow rate, and at a faster rate than that of the TO-46 package.
Should we invoke the thermal mass of the quasi-static layer of air in the laminar part of the boundary layer? I'd be surprised if there is enough mass there to do any good ... Or do we have to figue that there is a signnificant thermal gradient inside the LM35 package at high air flows, thus decreaisning the thermal mass in a way that you wouldn't see with a hermitcally sealed TO-46 metal can package.
I guess I could believe that the J-C measurement invoved something along the lines of the thermal-grease filled hole Bill suggested while the J-A values consist primarily of leads-to-air effects.
I was thinking of gluing or clamping the case to the object I want to measure the temperature of...maybe if I solder the leads to a tiny PC board and bond it down too it might work better.
Farnell has a bunch of thermally conductive glues and encapsulants.
The cheapest - and it isn't cheap - is a conductive rubber jointing compound (order code 130-485) for which they claim 2W/(m.K). I've used something similar and it worked well.
More expensive is the self-shimming thermal conductive adhesive (order code 537-020) which only offers 0.815 W/(m.K), in a layer not thinner than 0.15mm.
Even more expensive is the thermally conductive slicone potting compound, which offers
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.