Matěj Hejda is currently working as an Architecture & System Design Research Scientist within the Belgium-based team of the Large-Scale Integrated Photonics (LSIP) group of Hewlett Packard Enterprise (HPE). Previously, he has obtained his Ph.D. from the Institute of Photonics, University of Strathclyde (UK) working on spiking lasers and optoelectronic systems as functional primitives towards light-enabled neuromorphics. His main focus is on optical computing and neuromorphic photonics. He investigates integrated photonic circuits for AI acceleration, functional and system-level evaluations of such photonic accelerators, use-case driven hardware-software co-design for optical computing, and spike-based photonic/optoelectronic computing. His other interests also include neuroAI and photonic-electronic co-integration
Recent explosive growth in AI brought a resurgence of interest in application-specific computing hardware. Given the significant energy and resource demands of modern AI datacenters, research into emerging computing substrates and approaches has gained significant traction over the recent years. Among those, photonics (and integrated photonics in particular) stands as a promising technology not only for communications, but also for high throughput, physics-based computation. In this talk, I will provide an overview of our recent research work on photonics-enabled AI accelerators and neuromorphic systems from the architecture and system level.