A Superconducting Chip Operates at 20 Gigahertz
Hypres Inc. has demonstrated a fast operating clock speed for an arithmetic-logic unit (ALU). Their 8-bit chip has a frequency of 20 gigahertz. It can perform mathematical operations. This is a superconducting circuit and it could lead to energy saving high-end computing applications. Researchers from Stony Brook University collaborated with them to make a novel architecture. This technique is scalable to 32 or 64-bit processing. HYPRES did further work on the designs of the processors. The company also fabricated and tested the ALU. Single flux quantum electronics are superconducting in nature. They have been around for decades and are a possible way of overcoming various bottlenecks to progress. Hypres uses a niobium-based compound to construct the ALU. The whole thing is cooled to a very low temperature.
A SFQ has Josephson junctions in order to carry out calculations. The SFQ voltage pulses do not last very long and they happen on a picosecond scale. Other devices have integrated 3,000 to 20,000 junctions into a single piece. The minimum feature sizes are larger than other transistor arrangements. They cannot put as many components per wafer as leading manufacturers can. This means it may be hard to compete with ordinary silicon for specific HPC jobs. One product with the SFQ is a digital sampling oscilloscope. About a decade ago, there was some hope that they could help to run petaflop/s machines. They were never able to overcome the challenges necessary to do this. Instead of supercomputers, these higher frequencies are for specialized tasks where swifter cycling is crucial. The academics are closer to obtaining 32-/64-bit RSFQ processors. In Aug. 2011, they made a 16-bit wave-pipelined parallel-prefix adder that has a 30-GHz target clock frequency.
Conventional silicon has reached a plateau of 3 to 5 GHz around 5 years ago. The highest overclocked AMD CPU operates at 8.429 GHz. That also needs cooling, but it is not superconducting. Adding more cores enables a greater performance to a certain extent. However, most programs do not take advantage of this parallelism. Despite the past claims of the megahertz myth, clock speed is important. This is not the sole factor in evaluating a machine. There are fewer architectures now, though, that can make them more efficient. The academics appear to believe a superconducting exaflop/s system might still be doable. One hundred gigahertz electronics are considered achievable. These numbers may not directly comparable to conventional hardware. Perhaps these sorts of gains can make this a viable platform for various fields. It could be difficult to beat out other technology.
HYPRES collaboration achieves a 20 GHz 8 Bit digital circuit for future microprocessors. The main researcher’s page is on this site.
