Presentation
TiPU: A Spatial-Locality-Aware Near-Memory Tile Processing Unit for 3D Point Cloud Neural Network
DescriptionEnergy-efficient 3D point cloud neural network accelerators are desired for autonomous driving and AR/VR applications. This paper proposes TiPU, a spatial-locality-aware tile processing unit where the point clouds are partitioned into tiles to process spatial features locally. Intra-tile farthest point sampling and cross-tile neighbor search are employed to avoid unnecessary distance computing. To efficiently facilitate the tile operations, TiPU architecture consists of a unified distance computing unit, a near-CAM feature extractor, and a near-SRAM-computing MLP engine. The experimental results show that, compared to GPU implementation, TiPU achieves 2.8x processing speed and reduces 460x energy consumption.
Event Type
Research Manuscript
TimeThursday, July 13th1:40pm - 1:55pm PDT
Location3003, 3rd Floor
Autonomous Systems
Cloud
Autonomous Systems (Automotive, Robotics, Drones)