Low-Power Consumption Inference of ECG Signals Using Differential Logic Networks

The electrocardiogram (ECG) serves as a pivotal diagnostic tool for identifying arrhythmia. The swift and energyefficient detection and analysis of ECG signals hold paramount significance for medical institutions and patients. In this work, we propose a low-power consumption and highly efficient classification technique for ECG signals based on Differential Logic Gate Networks (DLGNet). By employing dynamic threshold encoding techniques on the ECG signals, we have drastically reduced power consumption for ECG signal processing compared to other neural networks, while achieving a notable detection accuracy rate of $\mathbf{9 8. 0 5 \%}$. Circuit implementation using the CMOS 55nm Ultra-Low Power (ULP) process demonstrated that the power consumption to process a single heartbeat is 5.67 nJ for realtime diagnosis and $1.48 \mu \mathrm{~J}$ for batch mode offline processing, respectively.