Presentation
FLNA: An Energy-Efficient Point Cloud Feature Learning Accelerator with Dataflow Decoupling
DescriptionGrid-based feature learning network plays a key role in recent point-cloud based 3D perception. However, high point sparsity and special operators pose great challenges to deployment. We propose FLNA, a novel accelerator with algorithm-architecture co-design. In algorithm, the dataflow-decoupled graph is adopted to reduce 86% computation by exploiting inherent sparsity. In hardware, we customize a pipelined architecture with block-wise processing, and introduce transposed SRAM strategy to save 82.1% access power. Implemented on a 40nm technology, FLNA achieves 19.1 − 70.3× speedup over RTX 2080Ti GPU. It rivals the state-of-the-art accelerator by 1.21× energy-efficiency improvement with 50.7% latency reduction.
Event Type
Research Manuscript
TimeTuesday, July 11th4:10pm - 4:25pm PDT
Location3003, 3rd Floor
AI
Cloud
RISC-V
AI/ML Architecture Design