dpak fets as heaters (2023 Update)

What have you got against resistors?

Or is it the handy metal tab as a more point heat source?

There are plenty of metal cased resistors after all and they don't tend to short out if something goes wrong...heck there is nichrome wire for that matter.

John :-#)#

Power goes as I^2, which wrecks the control loop. Fet power is linear on current.

The fets conduct heat through the board to the block nicely too.

Fets pick-and-place, no hardware, no tapped holes, no wires.

The fets are cheap!

I'll use copper pours to spread the dpak heat contact area. Gap-pads on the back.

Too much hand labor.

Actually, we've had the metal case resistors short, cycling at higher power.

On a sunny day (Mon, 23 Aug 2021 19:47:46 -0700) it happened snipped-for-privacy@highlandsniptechnology.com wrote in snipped-for-privacy@4ax.com:

Did you copy that from my tir-pic experiment ;-)?

TLC274 driving IRLZ34 on hotplate..

So long as you're happy with <Tsmax, and can avoid it in transient conditions. The plated-through board won't like anything over 130C for long, either.

RL

I want to heat the oven block to about 35C, which might need maybe 10 watts. I figure that each fet might have to dissipate a couple of watts, and each dpak tab will rise to about 40C. We might push the power up a bit at startup, to reduce warmup time, but that's just code.

Fets make great heaters.

Here's a small tweak:

Of course, we could just use an opamp per fet and dump the cascode idea. That would increase the cost of the project about 60 PPM.

That's 1/2 of a temperature regulator, even in a lab environment.

RL

Watch out that the slow conduction through the gap pad doesn't trash your control bandwidth. With a big chunk of metal like that you might be in the thermal mass limit, but maybe not--aluminum is surprisingly good.

We attach boards to cold plates using an exposed ENIG bottom-side pour with many vias to the ground plane, with a little bit of good-quality thermal paste and lots of small screws.

The thermistors go on the bottom, hanging five off the edge of the cold plate, with one side soldered to the big pour and the other side connected with a longish skinny trace.

The resulting delay is under 100 milliseconds (which is as well as we can measure it). That really helps the control BW, which in turn helps the thermal forcing rejection at low frequency.

No additional hand work, muss, fuss, or bother.

Cheers

Phil Hobbs

I use a soft 3G material that's about 5 w/m-K compressed. Under 0.5 k/w for a square inch; the vias are worse. The mosfet-to-block time constant will be milliseconds (thanks for the number!) and the giant oven block tau will be maybe half an hour.

I may as well leave the solder mask. Each of the six fets will only dissipate about 3 watts.

I'm planning on a small pcb on the bottom of my eom mounting plate. It will have three thermistor Wheatstone bridges and gap-pads. We'll use a 24-bit ADC. Gross overkill.

We're machining a prototype oven block and The Brat is laying out the controller board, which we can drive from a LabJack and Python, which setup the customer can play with.

They want sub-mK stability over hours. A Mach-Zender modulator splits the light, modulates the delays in the two paths, and combines, with

180 degree optical phase shift making darkness. There might be 20,000 or so wavelengths in each leg and we want cancellation to a small fraction of a wave. Temperature probably matters.

Okay, so you maybe don't care much unless you want to use local feedback at the heater. But thermal diffusion in polymers is very very slow--the diffusivity kappa is

alpha kappa = -------- rho*c_P

where as usual alpha is the thermal conductivity, rho is the density, and c_P is the specific heat of the material.

In ordinary plastic (alpha ~ 0.1 W/m/K), high c_P, the diffusivity can be 5000 times less than in aluminum. For a given bandwidth, the allowable thickness goes like sqrt(kappa)--i.e. 1 mm polymer is as slow as 70 mm of aluminum.

Your stuff has much higher thermal conductivity, but from a bandwidth POV a 5-mm pad will be something like 50 mm of aluminum.

Probably not a huge item if it's only 500 microinch or so.

Yup.

Bandwidth really matters, because you need lots of loop gain to get good rejection at low frequency.

For a spec like that I'd definitely use a proportional-only local feedback loop around each heater as well as the main loop. It doesn't have to be that that fast--a 100-Hz update rate should be lots--but your main loop will see these lovely fast temperature actuators, so it ought to be able to go much faster.

You can control gradients better that way too.

Cheers

Phil Hobbs

Why bother? You have heat pipes, just use a resistor on each, to heat the main block; a gap pad and heat spreader under the target device , with a thermometer or so, will presumably have no large heat sources, so will remain isothermal.

Since you WANT the gap pad to attenuate heat gradients, it doesn't need to be anything high-performance; greased paper would work fine. Mostly, power FETs are poorly characterized for non-switching use, so the resistor presumed nonlinearity is not a useful decision criterion. Just PWM the things; why not? Millikelvin switch noise is unlikely.

There will be fixed 24 volts across each dpak mosfet. And its current is set by a closed-loop current controller, set by a 16-bit DAC.

That's linear as hell.

I am late into the discussion, but why not just a PWM control of a FET turned on fully

Seems you need a loop control of the temperature and that loop is sloooooow

You can buy this off the shelf instead of building it

He wants to pretend to be original. I published that back in 1996.

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 the range 4C to 38C, using a Peltier heat pump and a thermistor sensor” Measurement Science and Technology, 7 1653-64 (1996)

It doesn't have to be. Using a fast sensor and putting it close to the heat source can help.

Getting the right sensor in the right place can be tricky with a bought-in controller. During the development of the 1996 gadget we did use a bought-in controller. It wasn't cheap and it was bulky. The auto-tuning was conservative. Setting the numbers using the Ziegler J and Nichols N 1942 Trans. ASME 64 759 procedure gave us much faster settling.

On a sunny day (Sun, 29 Aug 2021 09:41:57 +0200) it happened Klaus Kragelund snipped-for-privacy@hotmail.com wrote in snipped-for-privacy@nntp.aioe.org>:

Cannot speak for John, but in tri_pic I get PWM from a PIC micro and filter it and then do the current control of the power FET to avoid high current spikes (disturbing the rest of the circuit). Also saves a power resistor, now the FET itself heats the hotplate. Slow? A few ms... Outside temp variations are already much slower than that. And it worked for about 8 years to within 2 parts in 500, data available

scroll down for circuit and data.

Try: cat tritium_decay_experiment_one_year_data_14_5_2013_to_14_5_2014.txt | awk 'BEGIN {FS=" "} {print $9}' | sort -n

Took an hour or so to figure out the control so no significant temperature overshoot on powerup.

What would get hot?

I don't want any PWM noise anyhow.

The oven block thermal tau might be a half hour. Gotta test that.

The customer wants sub-millikelvin stability. I'll use a thermistor Wheatstone bridge and a 24-bit ADC and a software PID loop and 16-bit DACs to set the heater currents. Gross overkill.

But you should be able minimise it enough by careful design and compact layout. We did back in 1993.

We ended up with 414+/-2sec - not quite seven minutes - but the thermal resistance through the Peltier junction was a problem, and our light path was through open air, rather than an optical fibre

That's roughly what we used in 1993. We were worried about pulse-width modulation noise, but put lots of filtering right up against the point where it got generated, and it didn't mess up up the temperature sensing.

We got +/-1mK stability. The pulse-width modulation only had 10bit resolution, but that did turn out to be enough. I wrote it up and it got published in 1996.

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 the range 4C to 38C, using a Peltier heat pump and a thermistor sensor” Measurement Science and Technology, 7 1653-64 (1996).

Google scholar says that it has 24 citations, but two of them are mine.

John Larkin should find it much easier to insulate his electro-optic modulator, and he's not using Peltier junctions where the exhaust side has to have a low thermal resistance to ambient, so he should be able to do well - perhaps not as well as Larsen did in 1968 (Larsen N T 1968 Rev. Sci. Instrum. 39 1–12) or Priel did in 1978 (Priel Z 1978 J. Phys. E: Sci. Instrum. 11 27–30) - but both relied on well-stirred, well insulated water-baths.

Priel stirred his water-bath hard enough that his stirring paddle was also his heater. Niether Priel nor Larsen used 20-bit bit A/D converters - they weren't around back then - and they probably wouldn't have been good enough to get the +/-50uK that Larsen managed. or the +/3.5uK that Priel got.

For the kind of fixed temperature application that John has to deal with they don't seem to offer much, but they aren't expensive.