Yorktown Heights (NY) - IBM announced today reaching a significant milestone in practical on-chip optical data transmissions. Using a device called a Mach-Zehnder electro-optic modulator (MZEOM), one that’s 100 to 1,000 times smaller than similar previous devices, IBM has managed to send 100x more data between processor cores using 10x less power than over copper wires. IBM plans to integrate this technology into their future many-core architectures, eventually bringing our desire for 1000s of processing cores and true supercomputing power, to even our notebooks.

The silicon-based MZEOM is a device which has both a laser source input and electrical signal input. The electrical input signal is used to instruct the MZEOM when to switch the laser on and off. The laser itself is carried through something called a silicon nano-photonic waveguide, like an on-die fiber optic wires which connects two points both terminated by MZEOMs. The signal is modulated as required by the on-die circuitry, sending data to its destination. By using light, instead of electricity, not only can much more data be communicated, but with a fraction of the energy cost and generated heat.

According to IBM, the new technology "significantly reduces [manufacturing] cost [and therefore end-user cost], energy consumption and heat while increasing communications bandwidth between the cores more than 100x over wired chips." IBM has big plans for the technology. According to their press release, "The new technology aims to connect 100s, even 1000s of processor cores together on a single chip that would have the power of today’s large supercomputers."

IBM’s optical modulator uses silicon nanophotonic waveguides to control the flow of light on a silicon chip. The waveguides are made of tiny silicon strips (marked by purple color) with dimensions 200 times smaller than the diameter of a human hair, in a silicon-on-insulator (SOI) wafer. Digital electrical signals are applied to the p+/ i /n+ doped silicon nanophotonic waveguide through the electrodes (marked by gold color). Electrical charges (holes - green particles; electrons - red particles) are injected into the waveguide and change the optical properties of silicon, which is used to perform the modulation function. Extremely high modulation switching speeds are possible using this technology.

Background While visiting an Intel campus in October, I was taken through a silicon photonics laboratory. I saw several projects underway, and the basics of what IBM has introduced here was explained to me. The emitter/receivers I saw were quite large, about the size of a bread crumb from a piece of toast. A large blow-up of the device was shown to me on a poster explaining how it worked. Basically, the laser emitter sat flat on the surface. It beamed upward to a gold plated 45 degree mirror which literally sat right on top of it. That laser light was then directed via waveguides wherever it was needed. In most instances Intel showed me, it was simply to a nearby fiber optics wire which had to be aligned a particular way for optimum signal strength.

Conclusion plus opinion It appears IBM has made that emitter much, much smaller. Small enough, in fact, to have 1000s of them on a single processor die. If true, then we could be witnessing the first step between pure silicon and pure optical computers. These early hybrids will likely be able to keep Moore’s Law moving along, but will ultimately be replaced by a full-on optical solution, or something else more radical like a quantum computer hybrid.