Silicon photonics

Researchers quickly discovered that increasing the power of the laser could only go so far before something called ‘Two Photon Absorption’ got in the way – usually photons travelling through the silicon aren’t strong enough to excite an electron in the silicon atoms, which is how the light passes through in the first place. However sometimes if two photons arrive at an atom at the same time, there’s enough power for the electron to break free. Although rare, the stronger the laser the more frequently the impacts occur, and while normally the free electrons would recombine with the silicon atoms in the waveguide, with a consistent beam the electrons build up and start to intercept the photons, reducing the effectiveness of the beam.

The solution to this was a pretty clever one indeed – line the waveguide with diodes such that, when its charged, electrons are drawn to one side and allow a focused amplified beam to travel relatively unhindered.

Finally, by coating the ends of a waveguide with mirrors the laser reflects back and forth gaining strength before exiting the cavity in a focused beam – and thus the world’s smallest laser is born.

Modulating a beam by first splitting it and then shifting phases before recombining. Source: Intel

Data at light speed
This, of course, is just one step in the silicon photonics dream – now that we have the laser, there’s still the issue of encoding data in it, transporting it and receiving it.

Data is encoded with a modulator to turn the light off and on and produce the 0s and 1s that are the building blocks of computing. This happens at extremely high data rates – and in fact the most recent breakthrough in silicon photonics was the development of a modulator capable of operating at speeds in excess of 1GHz. It sounds small by today’s standards, but it’s expected that hitting ten or twenty times this won’t take long.

To gain these speeds the modulator in silicon photonics works by breaking the beam down into two smaller beams, making one of them out of sync (or out of phase) before merging them together again in a technique known as phase shifting. If you’ve studied your basic science on waveforms, you’ll know that an out of phase signal will cancel out its original source, much like noise-cancelling headphones. When the beams are recombined, out of phase beams negate the source beam and turn it on and off, and thus data can be encoded into bits.

It’s at the modulator where the limits of speed and future advances will take place. The faster data can be encoded and read, the more information that can be transmitted per second. However, with the lasers and modulators being mere microns long it’s possible to not just improve modulating techniques to increase bandwidth, but to combine many lasers onto a single chip and multiplex the signals for a combined data bandwidth that’s also infinitely scalable.

Naturally, waveguides need to exist in silicon to transport the signals and, presumably, optical fibre will be used to transfer it between components in a system. At the receiving end silicon combined with germanium is used in the photo detectors – remembering that silicon is transparent to the light, the germanium increases the silicon’s sensitivity to the infrared spectrum and in turn absorbs the light and, ultimately, decodes the signals and passes them on.

Performance
Intel has already demonstrated streaming high definition video using its silicon photonics chip and 5km of optical fibre in real-time with instantaneous playback. So far, while the technology is still in its infancy, it’s expected that with multiplexed chips data bandwidths as high as 40Gb/s won’t be uncommon within a few years.

The full capability of the technology is expected to be in the terabits per second category. You think your spanky 1GHz DDR2 RAM is all the rage? Wait until you’re running it at twenty times that, and that’ll be for budget stuff.

Well, that might be quite a few years away – we need optical memory, interconnects, controllers and of course CPUs, but it’s just a matter of time. The physical properties of metal conductors have real limitations that we’re already hitting.

For the first time in the history of computing the future won’t be about trying to pump more electrons, but to make the leap from electrons to photons.

Combining multiple lasers and modulators and multiplexing them onto a single optical fibre.