Efficient Training and Energy Saving Using Ternary Hardware Acceleration for PCG Classification

Cardiovascular diseases (CVDs) necessitate continuous heart sound monitoring to distinguish normal from abnormal signals for early diagnosis. While wearable devices using phonocardiogram (PCG) technology show promise, energy constraints limit their effectiveness. This paper presents a binary PCG classification method utilizing a ternary neural network with nonvolatile ternary logic gates. Based on NC-CNTFET technology, the design incorporates ternary full adders and multipliers for efficient MAC operations, reducing transistor count and enhancing energy efficiency. Results demonstrate significant improvements in energy dissipation, extending device lifespan and enhancing long-term disease detection capabilities.