Jason Perlow claims that Moore’s Law is going to have to be completely re-written — instead of microprocessor technology doubling its performance every two years, we’ll be looking forward to ten to twenty fold increases in computational power, at a bare minimum, every five years.
This would mean over ten years that computer speed would improve 100 to 400 times instead of 32 times.
Onchip photonics at very low power are what is required to enable zettaflop computing with systems that are relatively similar to what exists today. After that another re-architecting will be required to achieve more speed. Memristors for faster memory and memory that is beside the logic at a low level will also provide faster computing over this period. Memristor computing could also enable more gains from three dimensional computing structures. Memristors and photonic communication could enable yottaflop computing.
The first application of IBM’s nanophotonics technology will be used in the Blue Waters supercomputer, which will be installed at the National Center for Supercomputing Applications (NCSA) in Urbana, Illinois in the Summer of 2011. This system, which is based on large blade clusters of IBM’s 8-core POWER7 chip, will have a maximum peak throughput of 10 Petaflop/s, roughly four times faster than the Tianhe-1. It will also be much more compact and power-efficient as well, making it the most “green” supercomputer ever created.
However, even the Blue Waters supercomputer will become a clunky dinosaur once more and more nanophotonics elements are integrated onto microprocessors and replace many of the functions that traditional semiconductors perform today. And as with all supercomputing technologies, the equivalent of N minus 2 generations of processor performance will quickly find their way into enterprise systems and even consumer electronics.