Paul Whatmough

Paul Whatmough is a Research Associate at Harvard University, where he focuses on energy-efficient machine learning hardware and edge AI systems. He earned a B.Eng. with first-class honors from Lancaster University, an M.Sc. with distinction from Bristol University, and a Ph.D. from University College London. His research spans hardware acceleration for deep neural networks, sparse transformer processors, and TinyML model compression, with recent work on soft-error resiliency in embedded ML accelerators and noise-robust speech/NLP edge inference SoCs. He previously held research positions at ARM Ltd. and Philips/NXP Research Labs in the U.K. His contributions include a 12nm 18.1TFLOPs/W sparse transformer processor presented at ISSCC 2022, and he has secured NSF funding for a cloudless universal translator project. He collaborates extensively with institutions across the U.S., Europe, and Asia, including IBM, ETH Zurich, and Peking University.