ULP FPGA-Based CNN for Cardiac Monitoring in Space
A research team has developed an ultra-low-power FPGA-based CNN for real-time seismocardiography feature extraction on wearable health sensors, targeting both spaceflight and terrestrial applications. The system uses quantization-aware training and a systolic-array accelerator on a Lattice iCE40UP5K FPGA, achieving 98% validation accuracy while consuming only 8.55 mW and completing inference in 95.5 ms. The FPGA's power efficiency and radiation resilience make it suitable for battery-powered deployments in space environments. The implementation uses minimal hardware resources: 2,861 LUTs and 7 DSP blocks.
Key facts
- Ultra-low-power FPGA-based CNN for SCG feature classification
- Uses Lattice iCE40UP5K FPGA
- Achieves 98% validation accuracy
- Power consumption: 8.55 mW
- Inference time: 95.5 ms
- Hardware resources: 2,861 LUTs, 7 DSP blocks
- Quantization-aware training and systolic-array accelerator
- Targets spaceflight and terrestrial health monitoring
Entities
Institutions
- arXiv