Towards General-Purpose Unary Computing

Speaker

Josh San Miguel
University of Wisconsin - Madison

Host

Professor Daniel Sanchez
CSAIL MIT
Abstract:
There is a growing need for ultra-low-power processor systems that are not only capable of fitting in tiny form factors (e.g., wearables, implants) but also general-purpose enough to run a diverse set of applications under stringent power budgets. To build such systems, we make the case for unary computing. This paradigm uses unary logic and datatypes to shrink circuits by orders of magnitude; for example, a traditional binary multiplier can be reduced to a single AND gate in unary. Though promising, the barrier-to-entry for unary computing is currently too high, and the architectural design space has been sparsely explored. This talk will offer a primer on unary computing and present our recent work towards making it more efficient and programmable for emerging applications. Namely, we develop uGEMM, a unary building block for learning architectures, and uBrain, a unary brain computer interface. This talk will also offer a glimpse of our ongoing work towards a general-purpose unary CPU.

Bio:
Josh San Miguel is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Wisconsin-Madison. His research spans broadly across topics in computer architecture and systems, with recent focus on approximate computing, interconnection networks and intermittent computing. He is a recipient of the NSF CAREER Award in 2021. His work has garnered several honors, including the ISCA Best Paper Award in 2022 and three IEEE Micro Top Picks citations in 2016, 2017 (honorable mention) and 2021.