Energy-Efficient Hardware Design for CNN-Based ECG Signal Classification in Wearable Bio-Medical Devices

As traditional medical resources become increasingly scarce, current portable monitoring healthcare devices are proving inadequate. Consequently, there is a growing preference for wearable healthcare devices with diagnostic capabilities. Nonetheless, designing these wearable devices poses challenges due to limited hardware resources and the need for high diagnostic accuracy. This paper presents a scalable, energy-efficient hardware architecture optimized for ECG beat classification in the context of edge computing for wearable biomedical devices. We implemented selective clock gating on the most computationally intensive components to achieve low power consumption. Our findings reveal a substantial reduction in power usage, with our design consuming 8.2x and 4.4x less power compared to the highest and lowest power-consuming benchmarks, respectively. Our design demonstrates exceptional speed performance, with a latency of just 0.00906 seconds, surpassing several other implementations and ensuring the quick processing necessary for real-time applications in wearable technology. Furthermore, our design is highly resource-efficient, utilizing 23.7x and 2.99x fewer FPGA programmable logic resources than competing designs. These results highlight the advantages of our design for edge computing, making it particularly suitable for wearable health monitoring devices where minimizing power consumption, enhancing speed, and optimizing resource usage are critical.