IBM Breaks the Silicon Speed Barrier
At the GaAs IC Symposium in Baltimore last week, scientists from IBM Research presented experimental results showing that SiGe electronic circuits could sustain line rates greater than 56 Gbit/s. "It's the first time anyone has seen silicon moving at such high speeds," says Modest Oprysko, senior manager for IBM's communications link architecture project.
IBM created what it considers to be the key building blocks of 40-Gbit/s circuits: 4-to-1 multiplexers, 1-to-4 demultiplexers, transimpedance amplifiers, voltage controlled oscillators, and modulator drivers. All the chips were fabricated in the company's commercial 0.18-micron 7HP process -- which fact highlights the big advantage of SiGe technology: The manufacturing process already exists.
Previously, it was thought that front-end circuits for OC768 would require the use of so-called III-V materials technologies such as indium phosphide and gallium arsenide. Electrons move more quickly inside these materials, enabling them to respond at higher speeds.
Several vendors have gone this route, including Gtran Inc., Inphi Corp., and Vitesse Semiconductor Corp. (Nasdaq: VTSS). But they have needed to develop new processing techniques in order to do so (see Startup Claims 40-Gig First).
III-Vs have other drawbacks, too. The wafers are fragile and only readily available in 4-inch sizes, which limits the production scale advantages. And since the processing is less mature, yields are lower. In contrast, SiGe processing rides on the back of the success of silicon technology, which is now being commercially implemented in 12-inch wafer sizes.
Oprysko points out that SiGe technology keeps improving, just like the standard silicon processes on which it is based. In IBM's previous process generation, for example, the "unity gain frequency" -- the maximum operating frequency at which a transistor can still produce useful gain -- of SiGe was 47 GHz. In the process generation used to manufacture the circuits detailed above, this figure of merit has been increased to 120 GHz, which is approaching that of gallium arsenide (150 GHz) or indium phosphide (160 GHz).
Speed is not enough, however. Critics of silicon germanium often cite a couple of other reasons why SiGe isn't going to make it, as far as OC768 is concerned. One, they say it consumes too much power; and two, the breakdown voltage is too low to be of practical use.
As far as power consumption goes, the numbers also improve with each process generation. IBM's multiplexer and demultiplexer circuits consume about 1.2 W, which compares not too unfavorably with the 900 mW claimed by Inphi for its mux and demux chips in indium phosphide (see Inphi Pitches OC768).
Figures for breakdown voltages can mislead, says IBM in its paper. It's common practice to quote the breakdown voltage across the collector to the emitter of the transistor when its base is open circuited. For IBM's latest process, the breakdown voltage under these conditions is 1.8 V, which would imply that the circuit would self destruct when connected to any normal power supply.
However, proper circuit design can guarantee that open circuit situations never occur. Under less demanding operating conditions, the maximum breakdown voltage is typically 50 to 100 percent bigger. This has allowed IBM to make a modulator driver with a 3V single-ended voltage swing.
Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC) has a development relationship with IBM, giving it early access to new processes, and has already announced products based on the 0.18-micron 7HP process, including an OC768 modulator driver.
Does this mean the end of the line for indium phosphide? "Certainly, it will put additional pressure of other technologies," says Oprysko. But it's early days yet, he modestly warns. "So far there's only been speculation about performance metrics. As people start to demonstrate real devices, then we will really know where we stand."
— Pauline Rigby, Senior Editor, Light Reading