USS Clueless              Voyages of a restless mind

Main: normal long no graphics Contact Log archives Best log entries Other articles Site Search

Stardate 20010904.0726 (On Screen): It always blows me away to read accounts of technological advances which were clearly written by people who don't understand them. This report describes an announcement by Motorola that they've figured out how to fabricate GaAs components on a Silicon substrate. If true, this is a major advance in IC fab technology. But it's not quite what this report says. For instance, the report says that GaAs can "can transmit signals at much higher speeds". Well, no. The speed with which signals move around on the IC is about the same. What changes is how fast transistors are capable of switching.

It also describes this as being a boon for cell phones, something I know a great deal about. It will permit much faster CPUs to be put into cell phones without raising the price, it says. But there are three constraints on the CPUs used in cell phones: price, speed and power consumption; and of the three the last one is actually the gating factor. In the cell phones I used to work on, we were running the CPU at less than 20 MHz even though it was capable of going 200 MHz. That's because higher clock rates burn more power and our battery was tiny. We did everything we possibly could to reduce power consumption. It may be that GaAs will provide 35 times the CPU speed (though I'm skeptical about a number that large: 7 GHz??? in a cell phone???), but cell phones don't need blazing CPU speeds. (Our CPU spent 99% of the time halted.) GaAs isn't going to reduce power consumption per unit computing by 35-fold; I am not sure it will reduce it at all, in fact. GaAs has a lot of virtues but I never heard that low power consumption was one of them. This is the case for nearly everything which runs off batteries, by the way; the CPUs in such devices rarely run at even what current technology permits. Raising the clock ceiling is unimpressive if no-one is pushing the ceiling anyway. Even PDAs don't push the clock ceiling; the CPU in an iPaq runs about 235 MHz (IIRC) which is fast by PDA standards but isn't remotely as fast as it could run. (I'm virtually certain the CPU they're using is capable of twice that speed.)

They also claim that it's going to reduce price. I don't see how; what does GaAs bring to the table which will reduce die sizes or increase yields, which is what would be needed to reduce price? The only claim they make that I believe is increased computing speed, but the kinds of embedded devices they want to target with this aren't pushing the state of the art in that regard yet anyway. Motorola has found a very fine solution -- but it's not a solution to a problem that embedded devices have. Their marketing department needs to look again at their technology and figure out who really needs it. (discuss)

Update: Ah. Apparently what's going on is that this new hybrid process can produce parts with GaAs performance for a tenth of what a pure GaAs process would cost, and that's their justification for "cheaper". It's not clear that the hybrid process would be cheaper than the pure silicon process used now, however, so in practice it's not clear that it would really reduce costs given that no-one in that business is using GaAs anyway.

Update: OK; I probably should have gone to EETimes first. A cell phone chipset is actually three chips. One is huge and has the CPU and all the digital circuitry on it and represents the majority of the price. The other two are RF parts and it wouldn't surprise me to learn that they're currently fabricated with a pure GaAs process, since analog microwave is hard to do in silicon. This new technology will make those two parts cheaper, but they're a minor part of the total expense of the chipset, which is a minor part of the cost of a phone. Reducing the cost of those parts will result in a fraction of a percent reduction in the sales price a cellphone, but it will probably get used nonetheless.