Session
Domain-Specific PIM Accelerators from Client to Cloud
Session Chair
Event TypeResearch Manuscript
In-memory and Near-memory Computing
Cloud
Design
TimeWednesday, July 13th10:30am - 12:00pm PDT
Location3002, Level 3
DescriptionAs Moore’s Law enters its twilight years, domain-specific designs have emerged as a primary mechanism to deliver ever-increasing levels of performance under strict energy-efficiency requirements. The papers in this session cover a wide range of domain-specific PIM solutions, from network architecture search techniques targeting in-memory compute platforms, to accelerators for graph neural networks, stereo matching, recommendation systems, and beyond.
Presentations
10:30am - 10:53am PDT | NAX: Neural Architecture and Memristive Xbar based Accelerator Co-design | |
10:53am - 11:15am PDT | MC-CIM: A Reconfigurable Computation-In-Memory For Efficient Stereo Matching Cost Computation | |
11:15am - 11:37am PDT | An In-Memory-Computing Architecture for Recommendation Systems | |
11:37am - 12:00pm PDT | ReGNN: A ReRAM-based Heterogeneous Architecture for General Graph Neural Networks |