Presentation
Leveraging the Residue Number System for Designing High-Precision Analog Deep Neural Network Accelerators
DescriptionAchieving high accuracy, while maintaining good power efficiency, in analog DNN accelerators is challenging due to the costly high-precision data converters. We overcome this challenge by using the residue number system (RNS) to compose high-precision operations from multiple low-precision operations. Our study shows that RNS can achieve 99% FP32 accuracy for state-of-the-art DNNs using data converters with only 6-bit precision. We use redundant RNS to achieve fault tolerance against errors in the analog accelerator. We show that an RNS-based photonic core achieves better power efficiency compared to its electronic counterpart with the same precision.
Event Type
Work-in-Progress Poster
TimeWednesday, July 12th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
RISC-V
Security