Ultra-Low-Power FPGA-Based CNN Achieves Real-Time Cardiac Feature Extraction for Astronaut Health Monitoring

A new ultra-low-power Field-Programmable Gate Array (FPGA) solution has been developed for real-time Seismocardiography (SCG) feature classification, addressing the needs of wearable health sensors for astronauts. This approach utilizes Convolutional Neural Networks (CNNs) with quantization-aware training and a systolic-array accelerator, allowing efficient integer-only inference on the Lattice iCE40UP5K FPGA. The implementation demonstrates a validation accuracy of 98% while consuming just 8.55 mW and completing inference in 95.5 ms with minimal hardware resources. These findings indicate the feasibility of on-device cardiac feature extraction, paving the way for energy-efficient, autonomous health monitoring in space missions.