Reconfigurable Intelligence: FPGAs for AI & AI for FPGAs

Abstract:

As AI systems transition from training-centric development to pervasive, continuous inference, the constraints governing computation are changing. Inference increasingly emphasizes predictability, latency, energy efficiency, and integration with physical systems—challenging assumptions embedded in training-era infrastructure and architectures.

This talk explores how these shifts reshape the role of reconfigurable hardware. We discuss how emerging trends, including quantization and logic-centric computation, align with the strengths of FPGAs, and how reconfigurable fabrics can serve not only as inference engines but also as system-level companions in physical AI systems. At the same time, scaling specialization exposes fundamental challenges in backend tooling and design productivity. We discuss how AI-assisted interfaces and optimization agents can help navigate complex backend search spaces and make reconfigurable hardware more accessible at scale. The talk concludes with perspectives on heterogeneous AI systems, future forms of reconfigurability, and open opportunities for industry–academia collaboration.

Bio:

Alireza Kaviani has nearly three decades of experience in FPGA and ASIC technologies, spanning architecture, EDA tools, IC design, and low-power systems. He has authored more than 65 patents and publications and has been active in technical communication through university teaching and the delivery of multiple FPGA tutorials at major conferences. The speaker holds a PhD in Electrical and Computer Engineering from the University of Toronto.