why are they not using gallium arsenide on PC chips?

[Cartoon Corpse]

or are they?

i used to hear all about how gallium arsenide chips are the next big thing years ago. seems cell phones and other apps are using them. why not PC CPU's?

what about GPU, PPU?

(im assuming they're still predominantly silicon (silly con...hmm) maybe im wrong though.

 

[arjan de lumens]

Mainly power consumption. IIRC, for digital logic, gallium arsenide draws about an order of magnitude more power for a given performance level. Also, the material is brittle, toxic and expensive.

Its use in mobile phones is mainly tied to analog circuitry around the phone's antenna.

 

[KimB]

Really? Well, I suppose it would depend upon the manufacturing method, but GaAs semiconductors have much greater electron mobility (lower resistance). But I seem to remember that it's only for n-type regions, so that you need to significantly increase the number of transistors to balance out current flow (silicon has the nice benefit that its mobility is nearly the same for n-type or p-type, but still benefits from this balancing trick). So I would have thought that the main drawback would be increased die area, not power consumption, which should be lower in the end.

I do hear that cost is a huge concern, however.

 

[Guden Oden]

The Cray 1, which had GaAs chips running at a couple hundred MHz AFAIR, drew something like 40.000 watts of electricity and had +5V copper rails with a cross-cut area about the size of train tracks coming into it from the power supply... Scary stuff.

Some laughed at the Xbox 360's external PSU and the amount of heat it puts out. Consider that it offers greater computational throughput and system bandwidth in a package that must be a hundredth the size at the very most, whilst drawing roughly 0.375% of the power... Pretty nice development during the last 30 years.

 

[arjan de lumens]

The low mobility of GaAs in p-type regions means that GaAs logic families that resemble silicon CMOS generally will only have a very minimal performance advantage over their silicon CMOS counterparts; instead, to obtain really good performance from GaAs, you need to use logic styles that more resemble pure NMOS - these are perhaps 5x faster but have the disadvantage that the logic gates continue to draw power even while not switching.

As for total power consumption in CMOS-like logic families, it is not so much the gate resistance itself that controls the amount of power drawn, but instead the capacitive load that the gate feeds; as such, higher electron mobility itself will not have much of an effect on power consumption.

Did Cray1 use GaAs? That would be new to me. I know that they tried - and failed miserably - to make a Cray3 that used GaAs.

 

[_xxx_]

Theory aside, it's too expensive.

 

[Mintmaster]

The biggest reason we use silicon is that silicon oxide grows so well on it with very few interface states (which cause scattering and ruin the bandgap property of the semiconductor). It's really tough to find a good insulator for any other material. We're already down to about 15 angstroms for the insulator thickness, and even though high-K dielectrics will let us increase that, the insulator in the MOS transistor will continue to get thinner for faster switching speed.

One of my profs said silicon dioxide is the "chosen one". Another said it's a "gift from nature". I heard Freescale (spun off from Motorola) recently claimed to have solved the problems and created a MOSFET on GaAs that scales well, but who knows. I'm told people have fraudulently made this claim in the past.

 

[Xentropy]

Silicon does a few things very well, but it's hardly the "chosen one". The main reason GaAs hasn't caught on is due to reasons stated earlier, it has crappy mobility for p-type or enhancement Fets. I think this is true for the other 3/5 type semi's also (i.e. Indium Phosphide, Indium Antimonide). If you can't do a CMOS equivalent process easily, then you can't compete with Silicon for large scale circuits, beacuse the power draw dissipation would be huge.

It's not really due to cost, since the cost would come down if they were doing GaAs in high volume. The largest GaAs wafers I have seen are 6 inch diameter so far, so thats the cost limiter right now.

Silicon has one huge flaw however: the baseline substrate is not semi-insulating. Bulk Silicon substrates are on the order of 10 ohm*cm resistivity. Compare this to GaAs or Indium Phosphide or AL2O3 (Sapphire) which are on the order of 1e8 ohm*cm.

This means as you go up in frequency you are relying on a built in depletion capacitance surrounding each interconnect line and active device, and this isolation will linearly degrade with frequency. Silicon on Insulator (SOI) helps reduce this effect to an extent.

 

[nelg]

Silicon does a few things very well, but it's hardly the "chosen one". The main reason GaAs hasn't caught on is due to reasons stated earlier, it has crappy mobility for p-type or enhancement Fets. I think this is true for the other 3/5 type semi's also (i.e. Indium Phosphide, Indium Antimonide). If you can't do a CMOS equivalent process easily, then you can't compete with Silicon for large scale circuits, beacuse the power draw dissipation would be huge.

It's not really due to cost, since the cost would come down if they were doing GaAs in high volume. The largest GaAs wafers I have seen are 6 inch diameter so far, so thats the cost limiter right now.

Silicon has one huge flaw however: the baseline substrate is not semi-insulating. Bulk Silicon substrates are on the order of 10 ohm*cm resistivity. Compare this to GaAs or Indium Phosphide or AL2O3 (Sapphire) which are on the order of 1e8 ohm*cm.

This means as you go up in frequency you are relying on a built in depletion capacitance surrounding each interconnect line and active device, and this isolation will linearly degrade with frequency. Silicon on Insulator (SOI) helps reduce this effect to an extent.

Enough with the technical distractions. You and your NSA buddies have cornered the market for GaAs. Now get back to work on bugging my phone.

 

[_xxx_]

It's not really due to cost, since the cost would come down if they were doing GaAs in high volume. The largest GaAs wafers I have seen are 6 inch diameter so far, so thats the cost limiter right now.

You're not a beancounter, I guess?

That's not how economics work, it's exactly the other way around. Only if the costs go down will it ever be mass produced. Or if the advantages for the given product clearly overweigh the pricing pains.

 

[Xentropy]

You're not a beancounter, I guess?

That's not how economics work, it's exactly the other way around. Only if the costs go down will it ever be mass produced. Or if the advantages for the given product clearly overweigh the pricing pains.

Nah, I'm an engineer. But, yeah, it's probably more a chicken and egg situation. Someone has to make the investment first in infrastructure and assume there will be a market.

The starting material for GaAs is quite a bit more expensive, and likely to stay higher. GaAs and InP are significantly more brittle than silicon. I've seen silicon wafers dropped to the floor and not break. If you drop a GaAs wafer to the floor, the only question is whether it's 100 or a 1000 pieces.

The manufacturing process for GaAs circuits is not inherently higher cost, which I guess was my main point.