SSD Temperature Measurement

[WarthogARJ]

Hi fellow inmates,
It goes without saying that TPU is the best review site overall for SSD's.
And especially for thermal throttling effects.

Two related questions, if you please:
(1) Which FLIR system are you using?

I'm thinking of getting a thermal camera, and not sure if I'll buy one ready made, or make one.
That's only cost effective for the higher end though.
It's not for SSD's, or even electronic components per se, but sure if I have one I'll use it for that.

From the specs, and a few papers I've seen, I think the FLIR One Pro with the FLIR Lepton 3.5 sensor is a pretty good lower end system.
Pretty good sensor size (160x120 pixels), and acceptable radiometric accuracy (+/- 3 C/5%) and Spatial Sensitivity (12um/pixel).
Your shots look good, and I was wondering if you'd got them with this system.

Otherwise, the FLIR's with the much bigger sensor (464x348), the FLIR Lepton 3.0 or maybe 3.5, are very expensive as a ready-made unit ($+/-$9k??).
Better to buy the sensor on its own and then build the rest I think.
Or perhaps the higher resolution ON Semiconductor sensor (AVNET 1820HS)

(2) You've found that there's sometimes quite a big difference from the SSD's own sensors, and the FLIR.
And sometimes it's abiut the same.
Any ideas on where the SSD on-board sensor is physically located?
It's hard to get much info on the controllers because they seem to restrict the datasheets to NDA-access only.
Bit stupid, the competition would know exactly what they are.

Thanks,
Alan

 

[W1zzard]

I moved the thread to comments and feedback, but kept a permanent redirect

It goes without saying that TPU is the best review site overall for SSD's.

Thanks! Next test system which I'm currently working on will be even better.

Which FLIR system are you using?

FLIR E8, 320x240, the JPEGs are native res, just ran through Photoshop to reduce file size a bit

Any ideas on where the SSD on-board sensor is physically located?

Depends on the controller. Some report temp of the flash, some report their own temp, some have two sensors. Also they usually aren't calibrated that well, as I'm sure you've seen in our reviews.

I had big plans for FLIR imaging but ended up using it much less than I wanted back in the day. The user interest is incredibly low, it's not worth the investment.

 

[WarthogARJ]

I moved the thread to comments and feedback, but kept a permanent redirect


Thanks! Next test system which I'm currently working on will be even better.


FLIR E8, 320x240, the JPEGs are native res, just ran through Photoshop to reduce file size a bit


Depends on the controller. Some report temp of the flash, some report their own temp, some have two sensors. Also they usually aren't calibrated that well, as I'm sure you've seen in our reviews.

I had big plans for FLIR imaging but ended up using it much less than I wanted back in the day. The user interest is incredibly low, it's not worth the investment.

OK, thanks. I haven't used any of the FLIR type yet, but I think in terms of specs vs $$, the One Pro is best.
As much because it saves money by connecting to a smart phone for a lot of the electronics/display etc

You can buy the FLIR sensors themselves that are used in the more expensive models, at least I saw a price for the FLIR Lepton 3.5 used in the one Pro on Digikey for $199.
So it's not worth it to make your own one Pro, since it costs maybe $300 or so.

I didn't see your FLIR E8 listed, but the similar E95 with 464 x 348 goes for close to $10k I think.
I didn't try to find its sensor, because there's a much better one made by ON Semiconductor, the Avnet AR1820HS.
It's got a 4912x3648 sensor!
With a lot more control over it, in terms of being a real "camera".

I didn't see any being sold as singles, and the only prices quoted are for min order of 200.
But they do sell an Evaluation Board for $340 on Mouser.
So that's BETTER for a DIY job, because it's nicely arranged for you.
Evaluation Board

I haven't confirmed if it comes with the sensor, but seems a bit silly to sell it without.
If that's the case, I suspect you can ask them to add a sensor for the unit price of $50.
But the photo shows a sensor with it.

So that means for <$500 you can make your own FLIR with 4912 x 3648 resolution.....

One more point, I was thinking that a better way to test for thermal issues would be to artifically warm up the SSD, say with a controlled hot air stream, and see when it starts to cut out.
That's just to trigger the sensor.

And then you would know the temperature.

So then when you do the thermal throttle test, you know WHEN to expect things to happen, and can plan the test accordingly.

You'd want to do it somewhat gradually, to allow the sensor and the package surrounding it to heat up together.
But a hot air re-work station would be the right tool for the job.
You could even put a thermocouple at the end of your blast of hot air to see what it was on your target.
Most multimeters can plug one in (Fluke etc).

 

[WarthogARJ]

Correction.
I'm wrong about being able to use the ON Semiconductor Avnet AR1820HS sensor for a DIY thermal camera.
It will make a very good low light imaging camera, but it does'nt have the right type o sensor to give the precise measuremwnts you need in an FLIR system.

The type of sensor you need for an FLIR, is called an "uncooled microbolometer", and at the moment, the highest sensor resolution is 640 x 480 pixels wiuth a 17um pixel size (in the sensor, not the display).
And due to the infra-red wavelength size you are detecting, you cannot go below about a nominal 10nm per pixel to measure temperatures at ambient.
Which is bigger than for visible light sensors, because they work with smaller wavelengths.

This is based on Wien's Displacement Law, where Max Wavelength (um) = 2898/T (Kelvin).
And for the value you get, you'd need to add a bit more to account for the rest of the area needed for each sensor.
So in practice, most FLIR sensors use a 17um pixel size.

The highest resolution FLIR sensor i've found that you can buy on its own is the FLIR Lepton 3.5, at 160 x 120 pixels for about $200.
I've not seen the higher resolution sensors sold on their own yet.

Teledyne DALSO seems to be the best commercial sensor to date, but I've only seen it in ready-made cameras.

So if you want a higher res camera, the most cost effective solution is pretty much to buy the FLIR One Pro with the FLIR Lepton 3.5 (160 x 120 pixels).
If you can find a good deal on the camera (say $350), you won't save much by making your own by buying sensor off Digikey, Mouser etc: $200 for sensor plus some extra for an enclosure, Arduino etc and LCD.

And if you want a lower end but very affordable one, best is to buy a sensor breakout board based on the Melexis MLX90640 (32 x 24 pixels) at around $50.
You could buy the rest and stay underr $100.

Going higher res than 160 x 120 needs deep pockets at the moment, I'm afraid.
But this area is advancing fast, so in a year you might well find a high res sensor being sold on its own.

 

[W1zzard]

I vaguely remember that some good sensors popped up a while ago for the automotive market, did you look into that?

Your bigger issue will be calibration, but considering most readers are mostly after the pretty pictures and not super-exact data, this could be a viable approach

 

[WarthogARJ]

Thanks, i'll see if the automotive aector ones are any better.
I was interested in the application, and was researching it, and can see the funamental difference between a standard low light sensor, usung a CMOS, and one specifically designed for quantitative thermal measurements.

My interest is not restricted to computer components, it's just the place where I saw the use of the newer designs from FLIR.
I'd only used spot type before, or cooled sensors, and in a research lab.
I want it for various reasons, but not if I have to pay $10k!

The original use for these systems was military, but now they've managed to to make the sensors quite a bit cheaper, and increased the resolution, it's opened up new markets.
For automotive you can use them for anti-human and animal warning, and for medical, you can use for a fast skin temperature scan: say for Covid.
And related.

But the problem is that they use the Planck assumption of a black body to equate surface temperatue with radiation, and at near ambient temperatures, it's both a bigger max wavelength (that then means you need bigger pixels to capture it), as well as the energy in the wave is quite low at these longer wavelengths.
So for human's/animals and electronics under 100 C, you are a bit constrained.

And for calibration, I think if you do some initial benchmark/calibration with a thermocouple, you should be able to get the right emissivity.

 

[WarthogARJ]

Was thinking more on this topic, and how I would be using an FLIR camera to review SSD's.
I think it would be useful to see how a given SSD handles a given load, either Read, Write or Mixed.

In terms of both TOTAL power used, TEMPERATURE of key components (controller, DRAM and NAND), and an overall assessment of the change in ENTROPY, as defined by Shannon (of Computer Science fame).

If you initially assume no power loss due to electronic component inefficiences, then you could assign the data written, or read, or both to an SSD through its connection to the CPU, as a CHANGE in entropy.
As in, if you had a 1.0 GB set of data, composed of some sequence of bits, either 0 or 1, and you then store it on an SSD as the same set of information, in some ordered sequence, so that you can then extract the SAME information again.

There are computer science papers and texts that look into this, but keeing it simple for sake of this discussion, that you take this 1.0 GB sequence of bits that's on the "ouside of your system", and you take it in, and store it, causing the existing system (the SSD) to change from its starting state, to that of storing the info.
Which is a CHANGE in its entropy.
In a very simplistic way, if you had 1.0 GB of all "1's", and you wrote it on NAND flash that had been "0's" before, then that would be the CHANGE in entropy of the SSD.

If you had to initially pass through a controller's cache, and also a DRAM cache, and that requires them both to change entropy levels, then that needs to be taken into account too.
And is why it's the controller that gets hotter than the NAND, because you are forcing a small, dense collect of memory to keep changing in entropy, and it cannot release the exess energy fast enough, so its temperature rises.

And my point is that neglecting UNITS at the moment, you could compare a large amount of data vs a smaller one.
And also include the RATE.

And in terms of information thermodynamics, you could then look at an energy balance (measure electricity consumption) as well as do a thermal balance (in terms of temperature increase, and heat loss).
By measuring the controller, DRAM and NAND to the best of your ability.
The more information, the better your energy and heat balances.

Sure, be a bit involved to do the initial calculations, but can see how a simple model works before getting too complex.
And then you can see how efficient a given SSD, is in terms of handling the data.

At a very simplistic level, if you take a data file of uncompressed data, and compress it by storing it, you have altered its entropy.
And THAT involves ENERGY.
Either a net RELEASE, or a net CONSUMPTION, depending on how efficient your system is.

The same as you see from the Carnot cycle, when you are compressing gases, and then causing them to undergo a change in state (as a chemical reaction).

Maybe this level is too complex to get into for most readers, but just looking at power consumed by an SSD to process data at a given rate, and seeing the rise in temperature is useful to look at.