Flexible heat pipes?

How thin? Thin plates will have high lateral (spreading) thermal resistance, so even if you solder a heat pipe to one, it will only be cold in the immediate vicinity of the pipe. How serious a problem this may be depends on your situation.

Why use thin copper? Can you describe the actual situation better?

Heat pipes are expensive and can have reliability problems.

The problem is interesting. In general, good heat conduction is incompatible with flexibility. One interesting geometry might be two conventional or pin-fin heat sinks with their fins/pins interleaving closely, with thermal grease glopped between.

You might get away with soldering a few hanks of stranded wire between the plates. Maybe a coil of fine-strand #10 wire, soldered broadside to both plates; it would start to resemble one of those wire shock mounts. The numbers are roughly in the ballpark.

John

John,

We have a blade with 8 FPGAs and 2 CPUs. We already have separate heatskinks on each of them with some of the sinks having embedded heat pipes to reduce the lateral temperature gradients. There's almost 300W total power for the

10 devices, and of course, the FPGAs have a max die temperature of 85C (no industrial temp versions available on the hot ones).

At a 35C ambient, some devices are in the 50C range and others are in the mid 90s. The ones in the downstream air are the hottest ones (obviously).

So, my idea is to have one common heat sink or vapor chamber such that the device's (almost) get their temperatures averaged. The other nice thing about a common sink is that we would get airflow on both sides of the sink.

The hard part is connecting the 10 devices to the common heat sink with minimal temperature drop while also accomodating the chip-to-chip and board-to-board mechanical differences between the device heights and planarity.

In the past, we've attempted to connect multiple devices to a common sink, and the differences in height and planarity would be filled with thermal grease or a gap pad. The problem is that both of these have thermal conductivities (K) around 2W/K-m. The temperature drop is just too big for any decent amount of gap and power.

I need a gap pad with a K of several thousand. Spongy diamond would be nice. Got any?

Bob

-- == All google group posts are automatically deleted due to spam ==

I wish. Even diamond-filled grease is a rotten heat conductor. The little gel-filled baggies and flexy gap-pads are terrible too.

How about my interleave idea? Envision layers, top down...

Big finned heat sink

On the bottom of that, 10 small finned heat sinks [pointing down

On top of each device, epoxy a finned heat sink

Mate the mess, interleaving the fins, with some grease. The effective greased surface is large. Of course, you need to keep the gaps down, and do the math.

With a little thought, it wouldn't be too big a horror to assemble and rework.

Better yet, epoxy a little aluminum or copper block on top of each chip and circulate water through them all, with small flexy silicone hoses. 0.05 GPM will rise 20 deg C at 300 watts; put the hottest chips first in the string, or do some parallel flows. Of course, you have to circulate the water, and dump the heat somewhere. That would eliminate your existing, presumably expensive array of heat-pipey stuff. Zig-zag waffle channels inside the blocks would have excellent across-the-surface cooling. Should be fun to design.

I once bought an AlienWare computer that used a pumped-water cooling system with a radiator and colorful hoses and stuff. Cute, but it was a piece of junk for other reasons, and I sent it back.

Or, how about a mushroom heat sink for each chip, with some sort of compliant support structure above to sort of hold them all together? Scale the areas as needed to balance the temperatures.

John

[snip] [snip]

We have pondered liquid cooling and even an external air conditioner. It's too much stuff and too expensive. Our customers are used to a small standalone chassis. We currently have the most dense solutions in the industry, so if we add anything more and they'll go to the competition.

Having said that, I think we're about at the limit of bandwidth/power/density without resorting to one of these approaches. It's going to get interesting.

Thanks for the ideas. I'll let you know if we come up with anything that's clever and that works (sometimes those two concepts are mutually exclusive).

Bob

--
== All google group posts are automatically deleted due to spam ==

Thermal design is fun, because it's so hard.

One other observation: air prefers to flow around heat sinks and not through them; the fins get in the way. You can dissipate a lot of heat if you get serious about ducting air through the fins, and just not nearby.

I have an unproven theory that a heat sink should reduce the free-flow air volume in half for best cooling. Sort of like the maximum-power-transfer theorem.

John

I don't know of any that are specifically sold as flexible, but the wick-lined copper tubes that connect heat source and heat sink could be made flexible enough to accommodate this sort of separation - 12.5mm isn't a lot of space to accommodate the bends that you'd need to make the tubes long enough to be that flexible, but I'd guess that it could be done.

Basically, you've stream of steam coming off the hot plate to condense on the cooled plate, and you need enough cross-sectional area in the connecting tubes such that the necessary flow rate doesn't create too much of a pressure drop from one end to the other.

You can get the mass of steam - in moles per second - from the heat you are dissipating at the hot end and the latent heat of vaporisation of water (or whatever you are using as your working fluid) and the volume from the vapour pressure above your cooled plate.

The viscosity of the vapour is pretty much independent of pressure and only weakly dependent on temperature.

Hope this helps.

--
Bill Sloman, Nijmegen

I once saw a design where they had six MOSFETS on a heatsink about 15" long, in a current sink circuit. The six MOSFETS had different source resistors, such that the ones that dissipated more power were closer to the air source. (It was a dummy load for a battery charger/analyzer.)

Cheers! Rich

Makes sense. Heatsinks are seldom isothermal. Stuff like this is best tuned by experiment.

John

Heat pipes depend on evaporating a working fluid - usually water - into an evacuated space, and condensing it at some slightly cooler surface elsewhere, then having the condensed working fluid wicked back to the hotter surface to be evaporated again.

The evaporating and condensing chambers are usually separate, and connected by wick-line copper pipes which are as flexible as any other metal pipe - and if you make the connection moderately serpentine it will be plenty flexible enough to accommodate your mounting tolerances.

If you could get wick-lined bellows, and mount the condensing and evaporating chambers firmly enough to prevent the bellows collapsing under atmospheric pressure, you could have quite a compact solution.

--
Bill Sloman, Nijmegen

Thermacore does have flexible heat pipes that are similar in concept to wick-lined bellows. The problem is that they don't currently have anything that is as flexible as what we need. They also claim that the price is very large when compared with conventional heat pipes.

At least I'm having some fun experimenting with different ideas. It's cool to see water boil at room temperature when there's no air pressure to keep its boiling point at 100C, and then watch as the boiling subsides when the water vapor pressure increases.

Bob

--
== All google group posts are automatically deleted due to spam ==

Ideally, the "square copper plates" should be square copper boxes, with the bottom of the box (where the evaporation is going on) covered with copper mesh to wick the water in to be evaporated. The heat transfer "across the plate" is then done by moving steam around.

Eventually, the steam generated at the hot plate has to be condensed at the cool plate and the condensed water wicked back to the hot plate, but the pipes doing that job can be pretty flexible if they are made long enough.

Heat pipes aren't cheap, unless you can latch onto something being made in volume for a mass market application (like processor cooling in PCs).

They don't have much in the way of moving parts, and the only subtle failure mechanism that I know of is the build up of non-condensible gas in the gas circulation space. Oxygen and nitrogen will diffuse through copper, but only slowly. Hydrogen and helium go through faster, but there isn't much of either in most working environments.

--
Bill Sloman, Nijmegen

They have a number of failure mechanisms: corrosion, leaks, non-condensable gas generation from chemical reactions, blocked pores in the wick. It takes a lot of care to make one that's likely to be reliable for 20 years.

John

Copper is widely used for plumbing because it is fairly hard to get it to corrode. A system that depends on keeping its internal space at well below atmospheric pressure is indeed sensitive to leaks, but they don't happen without provocation.

It's kind of hard to get water to react with copper to produce a non-condensible gas. Iron and aluminium will do it, but they aren't used in regular heat pipes - if you bought any made from that kind of material, you'd have to be more careful.

Blocked with what? It's a sealed and evacuated system.

Perhaps. But it would take serious carelessness to make one that was likely fail much earlier than that.

--
Bill Sloman, Nijmegen

on each of them with

gradients. There's

max die temperature of

90s. The ones in the

'Downstream air' implies forced cooling.

What about better forced cooling?

A client I worked with a couple of years ago used forced cooling in their chassis which was used in an Opthamologist's office. The Pabst fans they were using sounded like they were for some sort of vertical take off device, it definitely would have bee obtrusive in a quiet office.

I proposed using these...

They're PC fans which move much more air than the pabst fans _much_ more quietly.

The mech engineer on the project poo-poo'd these saying they weren't well enough rated but when we looked into it they are more highly rated than the pabst fans with a much longer MTBF.

Is more, quieter fans an option?

Nial.

"Nial Stewart" wrote in message news: snipped-for-privacy@mid.individual.net...

With the amount of airflow we have, we're already deep into the thetaSA vs airspeed curve for our heatsinks.

The only options are cooler intake air, liquid cooling, or better heatsinks. We're working on the latter.

Bob

--
== All google group posts are automatically deleted due to spam ==

chips

to

eliminate

Zig-zag

was

=20

competition.

=20

into an=20

=20

will=20

under=20

anything=20

very=20

cool=20

keep=20

the=20

Sometimes half the trick is stepping way way back. Heat pipes are just one phase change method. Too bad you cannot pull off a Cray redux with a freon "waterfall".

Vertical or near-vertical heat pipes don't need wicks (heat source at bottom, heatsinks / fans at top) - just simple tubes, and a method to seal in a quantity of liquid / vapor under a partial vacuum. Make your own, with Al or Cu ends and any tubing in between. For PC and data storage applications the whole electronics package can be insulated (sound and heat), so it can just shut down on over-temperature. Even if it's in a fire, the heat pipe acts as a diode and the electronics pack survives.

Cheers Tony

heatsinks /

/ vapor

"Partial vacuum" isn't quite the right description. You don't want anything in the sealed volume except the heat transfer fluid - any vapour that is less condensible than the stuff that is doing the heat transfer slows the flow of gas onto the condensing surfaces. As far as I know, the usual technique is to boil off enough of the heat transfer liquid to flush everything else out of the system, but a perfectionist would pump it down to an hardish vacuum - say 10^-5 torr - then inject the right volume of the heat transfer material, and pump some of that off while sealing the system.

insulated (sound and

the heat pipe

If the dose of working fluid is calculated so that the pressure inside the system doesn't get high enough to blow it open when the whole dose has been evaporated by your fire. A good fire can melt even hard- brazed joints, so calculating the survivable peak pressure needs some interesting assumptions about the nature of the fire.

-- Bill Sloman, Nijmegen