A 65.7ms Quick-Instant-On and 50.7mJ/Frame Tiny YOLOv3-Based Object-Detection 16nm-AI-SoC with 32MB-Embedded-MRAM and Hybrid-Multiplication-Architecture
Student Contest:
No
Affiliation Type:
Academia
Keywords:
AI SoC, Near computing-in-memory, STT-MRAM, Application processor, Quick-instant-on
Abstract:
An embedded STT-MRAM-based near-memory computing AI SoC with hybrid multiplication architecture is developed, and its object detection function and performance are demonstrated. STT-MRAM serves as tiny YOLOv3-based AI accelerator weight memory and OS boot memory. The near-memory computing architecture reduces data transfer latency and energy consumption. The AI SoC demonstrates fast compact-Linux OS boot (65.7 ms) and low-energy tiny YOLOv3 inference (50.7 mJ/frame) on a system-level board.
