Grant: Discovery Grants Program - Individual

The recent explosion in AI applications promises profound societal benefit. However, the power dissipation of these systems cannot be ignored. An emerging and promising approach to reduce energy and latency of deep neural networks uses analog photonic computation. Successful deployment of these systems requires that they can be scaled to larger network size and the potential energy bottleneck of input and output is solved. This proposal aims to improve the high-speed electrical-to-optical and optical-to-electrical conversion required in these systems. Although such conversion has parallels with high-speed optical transceivers for datacom, several challenges, specific to or more stringent in the case of photonic accelerators must be addressed, such as linearity, non-balanced data, and noise. The applicant will develop solutions to these challenges as well as the challenge of tuning and calibrating the optical devices. This program will support and train two PhD and three MASc students over the next five years, who will finish their degrees as IC-design experts, having a sought-after understanding of electronic-photonic integration and co-design. These highly qualified personnel will fill an urgent need in Canada's microelectronics sector. Research outcomes will have application to photonic accelerators as well as datacom.