Gallium Arsenide CPU's ie Cray 3 and solar panels

The GaAs logic was all n-channel, either CML or depletion load or something like that. Ultra power hogs.

GigaBit Logic used to make a GaAs logic family. It was pretty awful. They are long gone.

I'd say zero. GaAs doesn't integrate well. Really fast logic these days is SiGe ECL/CML, but that's low gate count stuff, many milliwatts per gate function. The only thing that makes a usable CPU these days is silicon CMOS, with a little germanium here and there maybe.

Some fast opamps and comparators are SiGe.

GaAs has no p-channel parts. And its defect density is high.

GaN is great for power fets, like for radar and fast switching power supplies. There may be a few GaN ICs, RF amps I think.

The buzz these days is the demise of Moore's Law. People can fab 12 nm features in silicon, but a chip has to sell a billion dollars to make it worth developing. The good news is that, if chip densities level off at 10 or 14 nm, and stay affordable at 28, people may design uPs and FPGAs and leave them in production for decades.

Cray had large investment in GigaBit Logic, who made the chips. I used some GigaBit their chips in one of my projects, where they worked fine, and the GigaBit Logic data book is still on my bookshelf - for the high-speed logi c layout tips than rather than the devices, which GigaBit Logic could never make with high enough yield to make the business profitable.

In 1991 Gigabit merged with Triquint and Gazelle

and stopped selling the logic parts soon after - the firm seems to have con centrated on parts for the mobile phone market. which is rather larger than the market for supercomputers.

Slight.

Hi,

Cray 3 was made with GigaBit IC's afaik.

There is this company who apparently can make CMOS in GaAs efficiently:

Also VCSEL apparently are made in GaAs.

They have p-channel just require 9 times the gate area of the n-channel apparently, except for the POET link above, maybe that is closer to silicon 2 to 1 ratio for p to n channel gate ratio?

A lot of the RF amps, ie cell phones, high performance wifi etc have GaAs RF amps in them.

cheers, Jamie

Boy, if Moore's law is dead (sure has been true for Intel processors for some time) the long production life of pedestrian semis would be a great side benefit!

Jon

Yeah, but sometimes you hit a point of inflection.

WHAT belongs to "CPU"?

On Mon, 5 Oct 2015 17:30:09 -0700, Jamie M Gave us:

Bullshit. You are an illiterate, uneducated punk.

Half of that is right. You don't seem to be able to buy GigaBit logic chips any more.

They were a pain to use. Vcc was 0V (like ECL) and they needed a -2V rail f or the resistors terminating the transmission line you needed for the fast logic signals (also like ECL). You could get by without it, but the alterna tives were a bit power hungry and messy (as they are with ECL).

The negative power supply rail - Vss - had to be between -3.3V and -3.5V, a nd you needed a bias rail - Vee - between -5.2V and -5.3V which didn't nee d to sink much current as the power rails, which sank quite a bit.

On the other hand they were quite a lot quicker than 100k ECL, which was al l that was around at the time, and appreciably quicker than ECLinPS which w as a couple of years down the pipe-line - we got preliminary data sheets in 1990, but not a smell of real part.

They weren't awful, merely a pest to use, and the speed was awe-inspiring. We used synchronous logic synchronised to an 800MHz clock, and did a bit of interpolation between edges.

transmission line you needed for the fast logic signals (also like ECL). You could get by without it, but the alternatives were a bit power hungry and messy (as they are with ECL).

between -5.2V and -5.3V which didn't need to sink much current as the power rails, which sank quite a bit.

and appreciably quicker than ECLinPS which was a couple of years down the pipe-line - we got preliminary data sheets in 1990, but not a smell of real part.

synchronised to an 800MHz clock, and did a bit of interpolation between edges.

Hi,

What do you think about this new "revival" of GaAs for multipurpose system on a chip etc applications?

It looks like the company has been working on it for a long time without much to show so far, but they state big claims of disrupting the silicon semiconductor industry.

cheers, Jamie

It looks very much like hype for investors. The web site and paper is all about the number of patents they have, and how huge potential their ideas have. If they can persuade enough people to invest, they might manage to make some niche products, such as very high speed optical transceivers using GaA, but that's about it.

Remember how Seymour Cray described himself? "I'm a mediocre computer designer, but a world-class plumber" (or words to that effect). Cray machines were all about speed, regardless of how much power they took, and were designed as inside-out refrigerators with circuit boards around the outside.

None.

GaA may well work as solar panels, and for all I know they could be a touch more efficient than Si panels. But they don't come close in terms of economics or production scaling, and GaA is much more of a hazardous material (the gallium half is not poisonous, but has severe transport restrictions, while the arsenide part is particularly unpleasant). GaA is okay in small quantities - it is used in high-end radio and radar devices, and other niche areas, but you wouldn't want it all over the place for solar panels.

Other types of semiconductor have uses today or potential uses in the future, but don't expect to see anything replacing silicon. Silicon is excellent for general purpose electronics, and will continue to be an excellent choice long into the future. We will see things like GaN for high temperature, and perhaps graphene (or other flat materials) in a few high speed devices, but not for mainstream electronics.

Looks like a very long shot. 1000:1 maybe.

The hole mobility in GaAs is way worse than in silicon, whereas the electron mobility is (iirc) about 3x. So GaAs devices are pretty much N-channel.

"Gallium arsenide is the material of the future, and always will be."

Because by and large it stinks, except for NFETs, which can be amazing.

Gallium tends to occur in lots of lead/zinc mines, along with arsenic, antimony, and lots of other useful nonferrous metals.

Cheers

Phil Hobbs

There used to be a lot of discrete GaAs mesfets, which were pretty good, small stuff up to 10s of amps, fairly low voltages, under 10 usually. Many have been discontinued.

Phemts behave better (somewhat higher voltages, lower gate swing) but they tend to get discontinued too.

GaN is the new-new thing. High voltages, high Gm.

All nfets.

Hi,

Fair enough, but some industry bigwigs seem to be on board, so it is interesting that people involved in Global Foundries would be on board with this GaAs company I think.

Ajit Manocha ? Executive Co-Chairman, formerly CEO of GlobalFoundries

Dr. Suresh Venkatesan ? Chief Executive Officer, formerly Senior Vice President, Technology Development at GLOBALFOUNDRIES

Todd A. DeBonis ? Director, formerly Vice President of Global Sales and Strategic Development at TriQuint Semiconductor (played an integral part in the recent merger with RFMD and subsequent creation of Qorvo, Inc)

Mohandas Warrior ? Director, president and chief executive officer (CEO) of Alfalight Inc. since February 2004. Alfalight, Inc is a GaAs based high power diode laser manufacturing company

cheers, Jamie

Any good tales about this company? A search doesn't show much at all. Who bought these chips and what for?

I don't have any stories. They did make a really nice "pin driver" chip that I got a sample of, but it cost over $100 so I didn't use it.

I hear that Analog Devices makes cool pin drivers, but they aren't public.

Pin driver?

Logic in, logic out, and the output Vhi and Vlo are programmable. Tri-state is a nice option. Maybe even slew rate. They are used by the zillion in IC tester machines.