thermistors in parallel

A perforated plastic baffle will drop in between the fans and the boards. It will balance the air flow pretty well.

You shouldn't get an average across the parallel circuit, the voltage will be that of the hottest (lowest voltage) diode which will take the current away from all the other diodes. (That's why you restrict the current to around 1mA to prevent self-heating effects.).

I expect they will work out a lot cheaper than thermistors and it is easy to find alternative types in the event of supply difficulties. Future availability of equivalent parts is almost guaranteed.

You could use even more than that if you wanted a really big signal or an average of the temperature over the hot area. I made a solar panel pump controller with 7 diodes per side in a differential comparator bridge, it could easily respond to a 1degree C temperature difference.

We're all lead-free, but we couldn't let the PCB get anywhere close to solder melting temps.

lørdag den 6. maj 2023 kl. 21.52.56 UTC+2 skrev John Larkin:

weird, I just tried it and it works here

Found it. There were three strange "blank" lines at the end of the .asc that it didn't like.

That's sneaky, using the diodes as their own mux.

Mostly, but the diode conduction curve isn't a brick wall. A 10 deg c difference will be about 20 mv, and that's not enough for the hot diode to steal all the current from the other 3 or 7 diodes.

I don't know that I could trust the diode temp curve. The thermistor is guaranteed and costs 4 cents.

Three cool boards will stop the hottest from getting its fan on at the same temperature as four hot boards.

Could diodes be used with the thermisters, so that only the hottest one counts?

lørdag den 6. maj 2023 kl. 22.55.40 UTC+2 skrev John Larkin:

could work with NTCs too with opencollector/threestate outputs

Each board will have a bit that tells the controller "I'm too hot." If any is too hot, jog the fans up. If none is complaining, jog down.

The case now is to extend that to 8 zones of a single board.

If I analog mux and actually measure the individual temps, the FPGA can consolidate 8 measurements into that one bit to pass to the fan controller, basically set the too-hot bit based on the highest temp of the eight.

The paralleled thermistors would mostly work. The hottest will dominate the weighted mean temperature.

Something like that would work.

In the early '90s, I built a diffraction-based sensor as an online monitor for post-exposure bake in semiconductor lithography that worked like that. It had seven 1 x 3 inch solar cells arranged in a cone (like a poker hand, but 360 degrees).

All the cathodes were connected to a TIA together, and the anodes were grounded one at a time, using a zero-power PAL with tri-state outputs. (A PALCE22V10Z, iirc.) It had to be 3-state, because a logic high would have forward-biased the solar cells. The photocurrent was pretty small, so even a 1990 CMOS output resistance was negligible.

A normal PAL might not have worked, even in 3-state, because their low output levels weren't that well controlled.

Lasse's suggestion is a bit different because the selected diode is forward-biased, so a totem-pole output would be OK as long as it goes to

Dunno about FPGAs--do their outputs really go to ground like that?

Cheers

Phil Hobbs

That's interesting. The Trion is pretty low power so its ground paddle is probably close to actual ground. But we could measure it.

You would probably get away with it if you added a trimmer to make the normal operating conditions equal resistance on each thermistor line.

Otherwise you will end up too sensitive to the channel with the lowest resistance at ambient temperature.

Diodes dV/dT in series might be more reproducible.

'Tis 't'other way round.

A good thermistor has a 3-4%/K tempco and ~0.4K interchangeability. (Some are better than that.)

The diode tempco is about -2 mV/K, so to be competitive, the diode forward voltage at 1 mA would have to be consistent to better than 0.8 mV unit-to-unit, which is a bit hard to believe.

Cheers

Phil Hobbs

We very rarely use trimmers; people have to set them. The only case in the past 15 years is to adjust the gain on a GHz-class o/e converter that has no microprocessor.

The thermistors are 2% resistance accuracy at 25c.

Or LM45s, but they have too many quirks.

I guess I'll just brute-force mux 8 thermistors into an ADC and let code take over.

here's the plot of the Artix7 LVCMOS33 pull down from the ibis file (min/max/typ)

I wonder how real that is. The grounds around the chip aren't really zero ohms to pcb ground.

søndag den 7. maj 2023 kl. 18.48.56 UTC+2 skrev John Larkin:

the slope around zero is roughly 25 Ohm

But ground may not be ground.

It would be interesting to characterize logic low vs chip current (for various pins!) but that would be a project.

Most FPGAs have programmable drive strength too. Bigger project.

I can imagine making a DAC from gpio pins and some resistors, either a ladder or just delta-sigma. We do delta-sigma but with an outboard tiny-logic cmos buffer.

mandag den 8. maj 2023 kl. 18.44.25 UTC+2 skrev John Larkin:

with +/-50mV specs for voltage the ground must be pretty stiff

in the IBIS the ground clamps in an input looks like ~1 Ohm

it is all in the IBIS, pulldown with max drive strength looks like ~9 Ohm

you could use tiny-logic cmos buffers for an R2R DAC too

the big advantage of delta-sigma (when you don't need speed) is that it's only one pin so it'll always be linear