Re: Heat-tolerent computers.

Here we go:

a place to start learning about the exciting world of silicon carbide heat-tolerant electronics.

Silicon carbide can work at up to 650 degrees C... but even silicon can go to 350 degrees C.

But because silicon's performance degrades as temperatures go up, running silicon hotter doesn't really help things...

John Savard

Packaging is a huge problem at high temperatures. The differential thermal expansion of silicon and FR4 causes a great deal of stress, which is one of the three primary limitations on die size. (The others are yield and routability when there are too many I/Os per unit area.)

Cheers

Phil Hobbs

Yield I can understand.

Why the differential thermal expansion of silicon and FR4 - which, I see, is what printed circuit boards are made of - should matter, however, puzzles me.

The silicon is mounted to a substrate in a package, the package is plugged into a socket, the socket is on a PC board.

So the plastic of the socket, and then the plastic or ceramic chip package, need to match the FR4.

The chip, meanwhile, is bonded to metal in order to carry heat away. That metal plate can be put in the package in a way that allows for differential expansion. The pins on the package are connected to the pads on the chip by fine flexible wires.

So I would have thought that it's the immediate clash between silicon and aluminum that matters.

John Savard

"Plugged" is so 1980. A whole lot of them are big LGAs. The CTE mismatch can rip the corner pads off, or crack the solder balls.

Not nowadays, at least not in high performance designs.

Cheers

Phil Hobbs

" The main reason why this hasn't been tried is because the microprocessor itself is not very heat tolerant. "

I don't believe this.

I think I have seen pentium III chips run at close to 100 degrees celcius maybe even over up to 110 degrees celcius.

Perhaps you thought I ment farhenheit ?

Bye, Skybuck.

I'll admit I'm thinking of a consumer product desktop computer, as opposed to either a blade in a rack or a tablet. And you should know lots more about this kind of stuff than I do, of course.

I did some searching on this topic - besides finding out what FR4 _was_, I found out that silicon's expansion coefficient is quite different from that of just about all the other materials used with it.

But I would assume that if one is running a chip at a very high power level, so that it is getting hot, one is talking about something like a multi-chip module (MCM) or a chip in a socket with a big heatsink and fan on top of it. Not a compact package bonded right on the PCB.

Thus, when a desktop machine is overclocked, the chip's clock rate and power consumption both increase. The suggestion of the OP was, as I understood it, that if the chip could still function at higher temperatures, chips could be overclocked more, and perhaps with less elaborate cooling.

Less elaborate cooling could go all the way down to the iPad level, but one really doesn't want an iPad with a chip running at 500 degrees Celsius.

For one thing, the chip couldn't be isolated from the surface of the device well enough. You could burn yourself on something like that. And then there's battery life.

John Savard

I worked a few years in the packaging department at IBM Research. (It was a great place actually--that's where I made the infrared antenna coupled tunnel junction detectors.) Even though I'm not a packaging guy, I palled around with them and did some consulting for them, and in the process learned a lot about this stuff.

The man-years of wasted effort that ROHS caused was criminally irresponsible.

Cheers

Phil Hobbs

Was this the fab that IBM had trouble getting useful JJs out of after a shut-down?

Best regards, Spehro Pefhany

IBM never made Josephson devices in a fab, AFAIK. That project got shut down about 1982ish. My gizmos were normal metal - oxide - normal metal. Our best results came out in 2007--we got about 50-100x efficiency improvements over previous work by doing the electromagnetics right. (The result was a blistering 7% quantum efficiency at 1.6 microns, but the previous best was around 0.1% at 10.6 microns.)

It's certainly true that it often takes a few days to get a plant up and running again after a shutdown--stuff like corroded connections that worked before shutdown but fail due to inrush current on power-up, hoses that split due to water hammer, all sorts of marginal situations turn into hard failures.

Cheers

Phil Hobbs