Extremetech
IBM has become the first company to integrate electrical and optical components on the same chip, using a standard 90nm semiconductor process. These integrated, monolithic chips will allow for cheap chip-to-chip and computer-to-computer interconnects that are thousands of times faster than current state-of-the-art copper and optical networks. Where current interconnects are generally measured in gigabits per second, IBM’s new chip is already capable of shuttling data around at terabits per second, and should scale to peta- and exabit speeds.
After more than a decade of research, and a proof of concept in 2010, IBM Research has finally cracked silicon nanophotonics (or CMOS-integrated nanophotonics, CINP, to give its full name). IBM has proven that it can produce these chips on a commercial process, and they could be on the market within a couple of years. This is primarily big news for supercomputing and the cloud, where the limited bandwidth between servers is a major bottleneck.
There are two key breakthroughs here. First, IBM has managed to build a monolithic silicon chip that integrates both electrical (transistors, capacitors, resistors) and optical (modulators, photodetectors, waveguides) components. Monolithic means that the entire chip is fabricated from a single crystal of silicon, on a single production line; i.e. the optical components are produced at the same time as the electrical components, using the same process. There aren’t two separate regions on the chip that each deal with different signals; the optical and electrical components are all mixed up together to form an integrated nanophotonic circuit.
Second, and perhaps more importantly, IBM has manufactured these chips on its 90nm SOI process — the same process that was used to produce the original Xbox 360, PS3, and Wii CPUs. According to Solomon Assefa, a nanophotonics scientist at IBM Research who worked on this breakthrough, this was a very difficult step. It’s one thing to produce a nanophotonic device in a standalone laboratory environment — but another thing entirely to finagle an existing, commercial 90nm process into creating something it was never designed to do. It sounds like IBM spent most of the last two years trying to get it to work.
The payoff makes all the hard work worthwhile, though. IBM now has a cheap chip that can provide a truly mammoth speed boost to computers. It’s not too hyperbolic to say that this advancement will single-handedly allow for the continuation of Moore’s law for the foreseeable future.
Holy hell, this will be huge.
