A High Accuracy & Ultra-Low Power PPG-Derived HR Estimation AI Processor for Wearable Devices

PPG-derived heart rate (HR) estimation has been widely used in wearable intelligent devices and provides a convenient way to monitor cardiac conditions in daily life. However, HR estimation of PPG signals disturbed by motion artifact (MA) remains hampered. Recently, some end-to-end neural network methodologies have been reported to successfully realize accurate HR estimation from MA-induced PPG, but with significantly increased computational complexity, posing challenges to real-time performance and low power consumption. In this work, an HR estimation AI processor is proposed. Several techniques have been proposed to achieve both ultra-low power consumption and high HR estimation accuracy. Verified using FPGA and implemented with a 55nm CMOS process technology, the proposed design consumes low energy (90.0 µJ), while achieving the lowest MAE of 3.45±2.31 bpm compared with other existing designs. The processor is suitable to be integrated into energy-constrained wearable health monitoring devices for ultralow power and high-accuracy HR estimation.