Presentation
Neuromorphic System-on-Chip Towards Efficient Edge Healthcare
DescriptionAccurate classification methods for edge healthcare have been studied. However, reported works are computationally expensive and power-hungry. Furthermore, there has been a lack of co-design exploitation to integrate sensor and processor in system-level. To address these issues and empower ultra-efficient edge healthcare, this paper proposes NeuPI, an all-spike near-sensor neuromorphic system with a novel algorithm-sensor-processor co-design strategy. A unique hierarchical pipeline structure is also designed to further speed up the system response. The whole system is implemented in TSMC 40nm process and verified. The proposed design achieves a 97.96% accuracy on arrhythmia detection tasks with an energy consumption of <1uJ/classification.
Event Type
Work-in-Progress Poster
TimeTuesday, July 11th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
RISC-V
Security