Abstract: Integrated photonics offers unique advantages for accelerating artificial intelligence (AI) and neuromorphic computing by enabling ultrafast, energy-efficient, and massively parallel information processing. This talk will explore both the emerging opportunities and critical challenges in developing scalable photonic computing architectures, with a focus on issues such as device variability, optical loss and crosstalk noise, and integration complexity. I will discuss robust-by-design methodologies for enhancing the reliability and computational accuracy of photonic neural processors, as well as application-level techniques like model and architectural pruning to optimize photonic neural networks. I will present new architectures that leverage phase-change materials (PCMs) and photonic memory to further improve performance and energy efficiency for AI training and inference. Finally, I will discuss scenarios where photonic computing can outperform electronic computing for edge applications and share results from our ongoing work in this area.

Bio: Mahdi Nikdast is an Associate Professor in the Department of Electrical and Computer Engineering at Colorado State University (CSU), Fort Collins, where he is directing the Electronic-PhotoniC System Design (ECSyD) Laboratory. He received his Ph.D. in Electronic and Computer Engineering from The Hong Kong University of Science and Technology (HKUST), Hong Kong, in 2014. From 2014 to 2017, he was a postdoctoral fellow at McGill University and Polytechnique Montreal, Quebec, Canada. His research interests are at the intersection of integrated photonics, interconnection networks, and high-performance computing. Prof. Nikdast and his students have published numerous papers in referred journals and international conference publications and across different areas of VLSI, EDA, Photonics, Embedded Systems, Systems-on-Chip (SoCs), Artificial Intelligence (AI), and Computer Architecture. He edited a book on Silicon Photonics for High-Performance Computing and Beyond, published by Taylor and Francis Group in 2022, and another book on Photonic Interconnects for Computing Systems: Understanding and Pushing Design Challenges, published by River Publishers in 2017. Prof. Nikdast currently serves as an Associate Editor for IEEE Transactions on Very Large Scale Integration Systems (IEEE TVLSI), and has served on the TPC of various international conferences, including DAC, OFC, IPC, DATE, ICCAD, NOCS, etc. Prof. Nikdast and his team were the recipient of various awards, including the Second Best Project Award at the AMD Technical Forum and Exhibition (AMD-TFE 2010, Taiwan), the Best Paper Award at the Asia Communications and Photonics Conference (ACP 2015, Hong Kong), the Best Paper Award at the Design, Automation, and Test in Europe (DATE) Conference (DATE 2016 – Test Track, Dresden), the Best Paper Award Candidate at ACM Great Lake Symposium on VLSI (GLSVLSI 2018, USA), and the Best Paper Honorable Mention Award at ACM Great Lake Symposium on VLSI (GLSVLSI 2020, China). Prof. Nikdast received the prestigious NSF CAREER Award (2021), the George T. Abell Award for Outstanding Early-Career Faculty (2022), the Rockwell-Anderson Professorship (2022), the George T. Abell Award for Teaching and Mentoring (2023), and the IEEE CEDA Ernest S. Kuh Early Career Award (2024). He is a Senior Member of the IEEE.