CGRA Virtualization
Main content start
Speaker: Kalhan Koul, Stanford University
Date: April 1, 2021
There is an increasing demand in hardware to accelerate several applications simultaneously. For example, a public service may have several users requesting a single accelerator at the same time or a mobile phone may need to accelerate image processing and audio processing concurrently. This talk will present an overview of how we spatially and temporally schedule applications to maximize accelerator utilization and minimize end-to-end latency and describe a CGRA-optimized SoC architecture that supports this virtualization.
Presentation (pdf)
(1.13 MB)