Photonic Computing

Paxton Leung

The majority of computing performance improvement in the past decade can be attributed to the shrinking of the width of silicon transistors on chips. From 22 nanometres down to the latest 3 nanometres, the future of silicon transistor shrinking seems uncertain, with the investment required to further shrink transistors increasing exponentially. Computing power that continues to flourish enables humanity to build ever more powerful software: from allowing us to simulate rovers on Mars with greater accuracy to understanding human speech for Siri to set an alarm. However, without another breakthrough in computer science, we may finally reach the limits of exclusively-silicon based computing, and face a significant drag on global economic growth.

A potential breakthrough technology is photonic computing, where chips replace electronic-based logic, and light is used instead. Instead of copper wires, transparent vias, known as waveguides, are used, which traps and guides light akin to copper wires in a circuit. Akin to fibre optics, a bit would be represented by a pulse of light, but unlike electronics, a light pulse can travel for much further without significant signal loss or noise, and waveguides do not lose energy in the form of heat, compared to resistive heating of copper conduits on traditional CPUs.

The primary advantage of photonic computing is the fundamental principle that light of vastly different wavelengths do not interact with each other. A singular photonic circuit can have several different data sets being computed simultaneously, as long as these data sets are represented by light of different wavelengths. For example, matrix multiplication could be massively accelerated by photonics through the ability of multiplying several matrices at the same time on the same chip, drastically reducing circuit size, and time required to compute. Traditional silicon processors such as the graphics processors (which perform the calculations for all graphical tasks, and produce a display output to your monitor) would need to calculate the matrices one element at a time, or at best, one matrix multiplication at a time.

Certain optical properties can also be utilised in computing to deliver a quantum leap in certain scientific workloads, such as Fourier transforms using lenses, and optical properties to complete the calculation, instead of plugging numbers into complicated scientific equations. An image would be inputted into an array of lenses using a high pixel density display, and a light detector (such as the camera) would take the output. The two components take a lot less energy than a higher precision calculation performed.

Unfortunately, optical computing is still many years aways from delivering a general purpose processor, like your CPU in the heart of your laptop. Optical logic gates are large, and still require electricity to switch the gate on or off, or to boost the signal. In the near future, such as within the next 5 years, we may see photonic chips running in parallel to classical silicon chips, in the form of co-processing ASICs (application specific integrated circuits, circuits designed to do one, and only one thing really well). Photonic circuits could come pre-set with all the logic for a specific process, such as video decoding for a specific standard, and accessed if and only if a video needs decoding. The parallel nature of photonic computing permits significant speed ups, and power improvements compared to regular silicon based ASICs.

Page updated

Report abuse