thermal baggie things

off

One

PCB

ambient.

about

better,

Of course not. But the *easiest* way to cool the chips is to cool the board.

Also your board is not isothermal or

The PCB is nearly isothermal, as seen by scanning around with the FLIR. It's a 6 layer board with three copper pour planes.

The series thermal model is pretty good. FPGA-to board is in series with board-to-box. Any sneak paths, like FPGA dirct to box, are small.

Of course I thought of various other ways to do it, including pin-fin sinks directly on the FPGA and ADC. Tried them, too.

and a

required

up.

A 3D thermal system, with unknown factors? That would be spreadsheet fantasy, which itself is endemic. Companies have died because something stupid was typed into a spreadsheet and people believed it... because it was on a spreadsheet.

How would you enter the PCB geometry into a spreadsheet?

What's wrong with using experiments, with quantitative results, to do engineering? Engineers measure real-life components and systems all the time.

If an experimental method produces solid numbers and usable results, but is emotionally unappealing to you, do you call it bad engineering?

2. >>>

power

Note the sentence starting with "OK, so let's use the PCB as a heat sink". The implication is that John L. is unwilling to try or use my previously marginal suggestions for moving the heat away using the top of the BGA chip. I gave up on the idea of going through the board in my first rant at the beginning of this thread. However, John L. is driving this exercise his way, and I think we need to see what could be done with going through the PCB before giving up and doing it "the right way(tm)". I looked at the situation, found the air gap, and suspected I could improve on it's heat transfer. I'll admit that it's a marginal proposition, because at that distance, there's almost 100% heat transfer by radiation to the PCB. If the heat source were uniform across the chip, filling the gap would be a waste of time. However, because it seem to be concentrated at the die, enlarging the heat affected zone by filling the air gap will move more heat away from the hot spot, and hopefully improve the limited ability of the PCB do conduct the heat away. I really don't know what it will do without measurements and tests.

True. Neither is Fluorinert or any of the other recommended fillers. However, they're all better than air.

It will move some of the heat away from the center of the BGA chip, towards the output pins, which will then conduct the heat to the PCB. It's not a major gain, but it might be enough to buy a few more degrees.

I like the socket idea.

Again, that would be very true if the temperature across the BGA chip were uniform and each dogbone carried the same amount of heat. That's not the case, since the center is considerably warmer than the edges. Any help moving the heat towards the edges should be an improvement.

Ok. I'll amend that to "The copper sandwiched inside FR4 PCB is a much better heat conductor than air".

I'm guessing as these were not explained in the Altera literature. My guess(tm) is that: Jc means from the die to the top of the package, while: Jb means from the die to the base or bottom of the package. The die substrate has to be thermally bonded to one or the other. Kinda looks like it's to the base or bottom.

I would have like to see the color to temp legend, which the FLIR device will display. If there's a large difference in temperature across the chip, then I'm on the right track. If it's fairly small, this is a waste-o-time.

Please forgive my vague answer. I have done this before, on a blade server card that used an Intel Atom CPU soldered to the motherboard. My instructions were make it run cooler, but don't change anything. Among various experiments, I built a 3 sided dam from window caulk around the CPU, and filled the air gap with silicon grease at about

However, I got similar results by changing the heatsink style and adding a cardboard plenum, to direct the cooling air to where it's needed, so the air gap filling fix was never fully tested. Since I stupidly took on this job with only a verbal agreement, from a manager who apparently didn't have the authority to pay me, I bailed out when it was obvious I wasn't going to get paid.

Yep. Lowering the die temp is the ultimate goal of this exercise.

off

It would also be true if the system could be considered adiabatic (no heat goes in or out of the package) where the components are heating each other inside the box through convection (air circulation) and radiation. With the FPGA dissipating only 1.5 out of 6 watts total, the other components could easily contribute much of the FPGA heating.

As I previously mumbled, I don't think it's the best way. It's just the direction John L. is current going. I'm just trying to be helpful.

What I like to do is make some admitedly impractical changes to the thermal design, and measure the results. These will obviously not be optimized or even practical. However, they will tend to point in the direction which will yield the best possible improvement. Much as I would like to calculate my way through things like this first, the lack of data and a thermal design model suggest that experimentation might yield faster results.

Good idea. However, I don't think it's really necessary to invert the PCB. Using the sliding cover as a heat sink will probably be adequate with a stiffer cover and something under the PCB to balance the compression forces. However, I still like the BGA socket idea.

I'm 75 miles away from the problem. John L. can do his own measuring.

Yet another idea: Take a square plastic box, that's a little larger than the BGA chip, and attach it to the PCB top side with electronics grade RTV. Make sure the bottom is sealed to the PCB. Fill the entire square box with aluminum oxide dust. Shake to settle. With a little care, you can probably sneak some dust under the FPGA chip. Drop in a plastic cover onto the square box and RTV it in place. Now, we have a conformal fitting monster heat sink block. With a little imagination, we could mix in some thin glue, remove the square plastic box form, and turn it into a masonry construct. (No, I haven't tried this yet).

still

temp

the

above

cooling off

too

with

ambient. One

ambient. The PCB

ambient.

that about

is better,

are

analysis

IR

numbers, and a

thing like

required

set up.

do

temperature. The

have made

The narcissist in LARKin strikes again.

?-)

to

time.

is

I don't call it engineering at all. I call it reactive management and tech hacking.

?-/

Once upon a time, about 30 years ago, I had to deal with a synthesized marine radio that had a microphonics problem. That's where banging on the case would modulate the VCO, which could be heard over the air and in the receiver. There was also ocassional feedback from the loudspeaker at full volume. Lots of RTV, rubber board suspensions, and shock mounted modules, reduced the microphonics to a tolerable level, but didn't eliminate it.

In some situations, the radio was mounting in the vessel cockpit where it was exposed to direct sunlight. That heated up the aluminum case sufficiently to cause the radio to quit. Someone declared that a fan should be installed for such "tropical" installations. When it was tried, all we heard out of both the transmitter and receiver was modulated fan noise.

While everyone in engineering (and marketing) were producing endless solutions that were either impractical, useless, or both, my boss decided that perhaps it might be a good idea to determine what was failing in the radio when it was super hot. (Yes, you really could cook an egg on the cover. I tried it). Several minor issues were found, but the big one nobody guessed. It was the temperature difference between the top of the radio and the bottom that caused it to fail. Sections of the radio were on top, while other parts were on the bottom. Some sections were expected to track each other. That worked fine in the environmental oven, where the entire radio was at a uniform temperature. That didn't work so well in the hot sun, where there was about a 25C temp difference between the top and bottom of the radio. Better and more stable components were substituted and the fan magically disappeared.

I don't call it engineering at all. I call it reactive management and tech hacking.

?-/ _____________________________

As R.A. Feynman would say - "do the experiment".

"One experiment is worth a thousand expert opinions."

- Werner Von Braun

and the corillary is

"One experiment is worth a million amateur opinions."

- John Larkin

After all, nearly all science is reactionary to experiments.

Anything worth doing is also worth overdoing. A little thermal goo works well, so a massive overdose of goo should work even better.

The way it should be done is with fine silica (quartz) powder, a little grease, and plenty of volatile solvent. Smear some onto a heat sink, wait for the solvent to evaporate, and scrape the dust off the surface. The silica powder and grease that landed in the cracks has little exposure to the air, leaving the solvent in solution. That should fill all the voids.

I like to run memtest86 and memtest86+ on overclocked machines. About half of them will show RAM errors within about an hour. There are also some CPU stress tests that will send an overclocked machine into thermal shutdown.

Yep. You also can't lower it below ambient (air temp inside the box), which some calculations seem to show.

I've used olive oil for thermal transfer. It gets trapped in the voids in the heat sink and doesn't leak (after the initial compression). It has a certain organic flavor and appeals to the new age computing crowd. Only 0.17 W/m-K.

It's the same. See the first "myth": Unfortunately, that's not quite correct for large diameter pipes but nobody attaches an automobile radiator to a water cooled PC.

It can be a problem if pressure was high enough to produce a flow rate that creates turbulence and wall friction. At the pressure of a typical aquarium pump, neither is going to be a problem.

Give up and go nuclear:

These days, I'm doing AIR (Absurd Impossible Ridiculous) designs.

Oops. Make that aluminum oxide (alumina) powder.

One

PCB

ambient.

about

better,

Here is where the engineering comes in... to know what is "easiest", you have to have a target temperature for the die. You also need to know something about the current die temperature or have a good model for the thermal resistances and the starting temperature. You have played around with the camera to measure some temps that I haven't seen related to a die temperature. I haven't seen any sort of a model for the design. I haven't seen anything where you have actually measured the die temperature. You have found that you can lower the case temperature a few degrees with the thermal pads.

Did that meet your goals? If so, there was a lot you never shared with us.

Looking at the image you posted I don't see that. It looks like the board is rather hot near the two chips and cools rapidly away from there. Maybe you didn't post the best shots of this. You also didn't indicate actual temperatures on the board, so maybe everything in the image is within a couple of degrees which would make the image not very useful.

But that doesn't mean they can't be greatly improved. If you still have not met your goal for die temperature (once you figure out what it is) you can flip the board to facilitate heat sinking the chips directly to the case. This is likely a zero cost change to the case just requiring a hunk of metal to connect the chips and case. No need to worry about electrical impacts and since your pads work well, a couple of very thin ones would provide a very nice connection between the various parts.

They wouldn't do a lot since the bulk of the thermal resistance in that path is the relatively still air with low convection. But any thermal connection direct to the chip would make this a much better path.

and a

required

up.

typed

Yes, you can't make stupid calculations if you don't make any calculations at all.

You don't need a fancy mesh model of the entire board to get some sort of idea of how well various cooling methods would work. The starting point would be to actually measure the die temperature to see if you really have a problem in the first place!

The

made

Nothing is wrong with experiments, but you still don't know the basics of your problem or you at least have not shared them with us. What useful numbers have you produced? Deltas of x degrees doesn't tell you if your die is within the proper range or not since you don't know where you started and you don't know where you are now. That is why this isn't engineering.

What made your customer suspect the cooling in the first place? If the chip is still above 50°C it will feel pretty hot when touched. The customer may still be concerned. What exactly is the criteria you are using to say you have done the job?

I guess I don't have customers who want me to kiss butt or whatever this is called. If they have a concern, they express it to me in engineering terms and we decide on the goal to fix it. Then it is up to me to figure out how to achieve that goal. I don't get to play around with stuff until someone "feels" good about it. Maybe I'm jealous. :^p

Enough said. I got no reason to bust your chops. I just don't get it is all.

Oh, you never replied to the suggestion of fixing the problem with a heat spreader on the chip. It might not lower the actual die temperature, but you don't know what that is anyway, so who cares? What it would do is remove the hot spot on the case so your customer would likely think the thermal problem is solved... end of what you claim to be the problem, a noisy customer!

te:

ce

com

Yeah, a clearance hole in the PCB. With a nice ground plane on the bottom you could tie that to the case with standoffs, of course that's hard to do as a retro-fit.

George H.

ed text -

ambient. One

PCB

ambient.

about

better,

Customer wants 85C Tj. I am assuming that the hot-spot IR temp is a reasonable approximation of die temp.

You also need to

I've measured it and posted it.

or have a good model

The hot-spot IR temp drops 15C when I insert the Bergquist pad.

a 6

and a

like

required

up.

typed

Actual die temp is hard to measure. The customer has taken over the FPGA design and won't help me there. IT is a reasonable surrogate, especially when measuring deltas.

ambient. One

nt. The PCB

o ambient.

that about

is better,

board.

nable

h us.

the

. It's a 6

't

bly

th

ve

sinks

mbers, and a

ing like

he required

to set up.

antasy,

was typed

dsheet.

design

asuring

some FPGAs have a temperature sensing diode, all you need is a resistor and a voltmeter and of course that you can get to the pins

I seem to remember an analog devices app note on using an esd diode at temperature sensor

-Lasse

plane

2. >>>>

power

The real problem with John's approach is that he doesn't even have a goal other than to make the customer stop complaining. That reminds me of a story I heard many, many years ago when we were working on an IRAD write up for a DOD contractor (I am concentrating it greatly). Seems there was confusion between the "purpose" and the "goal" of the research. The story has a similar researcher asking superiors what the difference is and at each step he is told, no, that is your purpose and this is your goal making it loftier at each step. Finally he gives his presentation and starts by saying, "My goal is to get to heaven". That reminds me of John's design goal.

So how much will replacing the air with any of this other stuff improve the conduction relative to the parallel path through the BGA balls? I expect very little. Of course this is just my opinion. More important is to start with some idea of what the current conditions are and how much you need to change the current conditions.

Why bother talking about what might be a few degrees until you know if you need a few degrees or a lot?

"An improvement" is not an engineering approach really.

Still very irrelevant.

These things have to be measured in some way. Typically the way they are measured tells you how to apply them to a real situation like this. If they measured the heat conduction to just the top of the package, how did they do that without board conduction? Likewise the other measurements. Once you understand how they measured it you will understand how to construct a very simple model from these numbers and then can get some insight into which conduction paths have more room for improvement.

I agree. Getting a spot temp reading of the case is of little value and could have been done with a $70 IR thermometer.

Lol! In other words, change physics?

I'm impressed with these numbers actually. I would not have expected that. For the filler to work at all, first there has to be a significant temperature difference across the balls. Adding a thermal superconductor across the gap won't decrease the temp by more than the current drop across the balls. I wouldn't have expected it to be this much.

Yeah, if there was a heatsink there should be lots of room for improvement there. They are usually not so great and the airflow designs are often poor.

Assuming it was needed in the first place.

off

Something isn't right about terming any part of this "adiabatic". This problem is *all* about the heat flow and adiabatic means no heat escapes...

but yes, the other parts could easily be raising the FPGA temp and most certainly are. It is not a point problem, it is a system problem.

I know John is a funny fish. I recall others saying he often hides parts of the problems he discusses so others can't really know what is going on. It sure looks like the case here. It is absurd to begin this effort without a die temperature.

Why is that so great or even better than a heat sink?

Great, so what is your die temperature before and after the great experiment?

ambient. One

The PCB

ambient.

about

better,

So your goal was 85°C and you started at 65°C... problem solved and you didn't need to do *anything*!!!

Which is not bad considering you were 20°C below your target before you started.

I'm sorry, I didn't hear that reply...?

a 6

So do you have images that show the uniformity of board temperature?

design

measuring

Which part is this? I thought most FPGAs had built in thermal diodes to make it *easy* to measure die temperature. At this point all you need is to solder a couple of wires to the right pins and hook up a measurement device. Besides, you already said you had a die temperature from your IR photos...

off

One

PCB

ambient.

about

better,

and a

required

up.

The

made

Well, he spilled some beans in a post finally. His die temperature started at 65°C and his goal was to get it to 85°C. So he still needs to heat it up another 20°C. lol

It wasn't so much that they thought lots of it was better, but they thought it was a good thing rather than realizing it is a bad thing that you need to minimize. Once you get the bad thing close to zero, how much better than this other not quite as bad thing be?

Interesting. Can you really scrape it well enough to get the final surface flatter than without the dust?

When I looked up the difference in delay times for silicon transistors I realized that unless your CPU was running really hot, there wasn't much point in all the nonsense. My main goal became making the thing quiet, but I never completed the effort. I got sidetracked with a job.

I also looked at refrigeration and someone pointed out that condensation can be prevented by moving air over the cold surfaces fast enough they don't cool to the due point. It is static air that condenses. My thought was to just put the entire CPU in the fridge (without the hard drive). But the little dorm fridges likely won't move that much heat.

Uh, really? That is one of the parts in my basement that was going to be connected to a smallish, but standard air cooled heat sink. I was actually thinking of using convection to circulate the water, but never got that far.

Uh, can't nanoWatts be had without nuclear fuel? I

FPGA, as measured at its hot spot with the FLIR, drops from 60C to 45C with the Bergquist pad. I haven't been able to measure actual die temp. Whatever it was, the pad reduces it by 15 C.