Close

Presentation

Optimizing ADC Utilization through Value-Aware Bypass in ReRAM-based DNN Accelerator
Time
Location
Event Type
Research Manuscript
Virtual Programs
Hosted in Virtual Platform
Keywords
AI/ML System Design
Topics
Design
DescriptionReRAM-based Processing-In-Memory (PIM) has been widely studied as a promising approach for Deep Neural Network (DNN) accelerator with its energy-efficient analog operations. However, the domain conversion process utilizing power-hungry Analog-to-Digital Converter (ADC) hinders the overall energy efficiency. In this paper, we propose a value-aware bypass method to optimize the ADC utilization of the ReRAM-based PIM. By exploiting the property of bit-line (BL) level data distribution, the proposed work bypasses the redundant ADC operations depending on the magnitude of value. Evaluation results show that our method successfully reduces ADC access and improves the overall energy efficiency by 2.48x-3.07x compared to baseline.