RISC-V Vector Processor for High-throughput Multidimensional Sensor Data Processing & Machine Learning Acceleration at the Edge

The computing solutions of tomorrow must be more energy efficient than those of today, which requires combined efforts to be conducted on multiple research areas: from new transistor technologies to innovative software algorithms by way of original processor architectures. This project enters into this last research area by revisiting the vector processing model, which provides a highly efficient way of exploiting data parallelism in scientific computations, sensor processing, and machine learning algorithms. Indeed, the efficiency of vector processors comes from their ability to perform parallel-data computations on very large vectors, thereby amortizing the overhead of fetching and decoding instructions. In this project, we aim at developing a state-of-the-art energy-efficient open-source vector processor that follows the RISC-V “V” specifications. To this end, we will explore original architectures – low-precision instructions and autonomous memory subsystems – that will lead to better performances in a reduced power envelope.

Intern: 
Mohammad Hossein Askari Hemmat;Mickaël Fiorentino;Matheus Cavalcante;Matteo Perotti;Alireza Ghaffari
Superviseur universitaire: 
Yvon Savaria;Jean-Pierre David;Luca Benini
Province: 
Quebec
Partner University: 
Programme: