Link: https://eu02web.zoom-x.de/j/66719911736?pwd=N5n96kFJbauLi2u79eJI0ZD15hgNsi.1
4:30pm Central European time is (usually) 7:30am Pacific time and 11:30pm Beijing time
Machine learning at the edge calls for computing architectures that are both energy-efficient and high performance. Traditional Von Neumann systems struggle meet these requirements, as they have to cope with the cost of data movement between memory and logic. Addressing this challenge, we present Faraday, a scalable near-SRAM computing architecture for edge machine learning. By exploiting SRAM parallelism for core dot-product operations and supporting full network execution near memory, Faraday minimizes memory-to-CPU data movements, resulting in speedups of up to 200x, with minimal area overheads. Application mapping on Faraday is performed by an dedicated SDK enabling the automatic deployment of ONNX-defined models.
Grégoire Eggermann is currently working towards his PhD degree with the Embedded System Laboratory (ESL) of EPFL, in Switzerland, under the supervision of Prof. David Atienza and Dr. Giovanni Ansaloni. He received his MSc degree in Micro and Nanotechnologies for integrated systems from Grenoble-INP Phelma in 2022. His main research interests include the development of near-memory computing architecture for low-power AI applications.
