IBM's 500GHz processor? Not so fast...
6/22/2006 3:29:45 PM, by Jon Hannibal Stokes
I'm going to resist the urge to go on a rant about J-school grads who horribly mangle their coverage of technical topics, because, to be fair, I'm not exactly an analog RF guru myself. However, I will say that yesterday's round of press reports about the IBM/Georgia Tech 500GHz transistor was just a train wreck: "OMG a 500GHz supercooled computer chip!! Terahertz CPUs here we come! Also, cell phone chips only run at 2GHz!! (I'm, uh, not sure why this cell phone detail is relevantshouldn't we be comparing these to Pentiums?but that's what the PR guy said so I'll just throw that tidbit in my article.)"
If you read the coverage yesterday, let me see if I can untangle it for you. Like I said, I'm not really an analog circuits guy (I hated those classes, and I don't follow that industry), but I'll give it a shot. And if you didn't read the coverage, then hopefully the following will save you a headache.
Switching vs. clocking
First off, what the Georgia Tech/IBM team demonstrated was a transistor that can be switched at 350GHz at room temperature, and 500GHz when cooled to near absolute zero. Now, just because a transistor can switch that fast, it doesn't mean that you could build a processor that is clocked that fast. The transistors that make up, say, a Pentium 4 processor could theoretically be switched much faster the clockspeed of even the fastest Pentium 4 chip; but how fast an individual transistor can be switched and how fast you can clock a complex digital circuit that consists of millions of those transistors communicating in lock-step are two very, very different numbers.
But really, there's no need to even talk about desktop processors here, because the kind of transistor that IBM and GA Tech goosed up to 500GHz is intended for use not in digital circuits but in analog RF devices.
CMOS = digital, like for CPUs; BiCMOS SiGe = analog RF for wireless devices
The other thing that was confusing about the coverage of the IBM/GA Tech announcement was this universally repeated comparison of the 500GHz transistor to a 2GHz "cell phone chip." To understand where comparison this came from, you first have to know that wireless devices like cell phones use multiple kinds of silicon-based circuits to do their thing.
The kind of silicon circuit that most people are familiar with is the low-power, embedded microprocessor that the phone uses to run its software (Java games, the interface, media playback, etc.). This processor is produced on a typical complementary metal oxide (CMOS) process, and it certainly does not run at 2GHz, or anything close. What runs at 2GHz is a much smaller, simpler chip containing an analog circuit that processes the wireless signal in real-time. Unlike the cell phone's CPU, such chips are made on a silicon-germanium (SiGe) BiCMOS process.
Because these SiGe circuits are performing analog processing of a high-frequency radio signal, the transistors that make up the circuits have to be able to switch at the same frequency as the radio signal that's passing through them. The faster these transistors can switch, the higher the frequencies they can be used to process. So a phone that operates in the 2GHz spectrum needs an analog RF chip that can operate at 2GHz; likewise, a radar device that operates in the 24GHz spectrum needs an analog processing circuit that can operate at 24GHz, and so on.
By making SiGe BiCMOS transistors that can operate at ever higher frequencies, IBM and others hope to enable the wireless industry to produce electronics that effectively utilize new portions of the radio spectrum.
At any rate, there's a good thread in the CPU & Motherboard forum on this, so check it out if you get the chance.
