A Reconfigurable Non-Volatile Memory Architecture for Prolonged Wearable Health Monitoring Devices

Wearable devices have broadened the horizons of healthcare systems within consumer digital products. However, these devices have limited power resources, emphasizing the importance of power management at all design levels. This paper proposes a novel energy-efficient, non-volatile SRAM designed to address the power constraints of wearable healthcare devices. The proposed design offers a non-volatile alternative to commonly used flash memories, exhibiting a remarkable 1000-fold reduction in power for both read and write operations. An in-memory computing architecture based on Ternary Content Addressable Memory (TCAM) is proposed to optimize power and mitigate data transfer challenges between memory and processor. The versatility of the proposed design extends to security considerations, as it functions as a physical unclonable function (PUF) to enhance system security. Comprehensive simulation results demonstrate a substantial reduction in energy consumption during data backup and restore operations in the non-volatile SRAM, accompanied by a 47% improvement in search energy. In addition to these advancements, a low-power TCAM-based dual-stream convolutional neural network is implemented to enhance the detection of heart sounds. The findings affirm that the proposed approach stands as an efficient solution for wearable healthcare devices with limited power resources, extending the continuous heart sound-capturing capability to approximately one year.