Real-Time Stress Monitoring with Flexible Electronics: A Co-Design Approach from ADCs to Feature Extractors and Classifiers

Prolonged stress has significant consequences for mental and physical health, making its continuous monitoring a necessity. Wearable devices can offer real-time physiological tracking but primarily rely on rigid silicon-based hardware, which does not adapt well to the body surface, leading to discomfort during extended use, and is frequently too expensive for broad accessibility. Flexible Electronics (FE) provide an attractive alternative, offering lightweight and bendable systems at ultra-low costs, making them ideal for affordable and comfortable healthcare devices. Current FE approaches primarily focus on identifying stress levels via Machine Learning (ML) classifiers, or propose standalone hardware (e.g., bio-sensors) but overlook the impact and potential co-design benefits of endto-end system design, such as sensor interfacing and feature extraction costs, which may dominate the system's requirements. In our work, we propose an automated hardware-aware framework to co-design all major system components: Analog-to-Digital Converters (ADCs), feature extractors, and ML classifier. Our co-optimization framework tailors the flexible hardware implementation to the application requirements, thus complying with demanding area and power constraints associated with FE. Our experiments demonstrate that our framework significantly reduces interfacing and feature extraction costs by 69.24%, and enables real-time stress monitoring with silicon-level accuracy along with the unique advantages of flexible, conformal hardware.