Robust human-machine interfaces enabled by a skin-like, electromyogram sensing system

摘要

Recent advances in materials engineering, chip miniaturization, wireless communication, power management, and manufacturing technologies have shaped our perspectives on wearable electronics in ways that wearing a Fitbit is as casual as driving to work. However, establishing a robust sensor-to-skin interface remains as a significant challenge due to the drastic contrast in soft, dynamic human skin and rigid electronics, limiting the adoption of technology to leisurebased applications. Here, we present an engineering solution by combining the respective merits of thin-film nanostructures, soft materials, and miniature electronic components and developing a soft, hybrid, wireless, wearable platform. We exploit conventional CMOS processes to fabricate metal/polymer nanostructures, implemented as dry contact electrodes (thickness ~3 μm) as well as a flexible interconnection system (thickness ~10 μm). The electrodes are further optimized by incorporating an open-mesh network geometric features allowing for prolonged, intimate contact throughout repeated and dynamic deformation of human skin. The skin-electrode impedance and the signal-to-noise ratio are ~18 kΩ and 29.52 dB, respectively, from electromyogram (EMG) recordings, matching the qualities of Ag/AgCl hydrogel electrodes. The stretchable circuit layer contains pad metal structures compatible for integration of surface mount chip components using a conventional reflow soldering process, allowing for easy integration of commercially available integrated circuit solutions. Silicone-based elastomer is used as both the carrier substrate for the thin-film structures and the backing layer providing the necessary adhesiveness to the skin. We verify that the completed system can be stretched up to 10% based on computational and experimental analysis. Finally, we demonstrate the robustness of the system functionality by showcasing human-machine interfaces (HMI) based on a single-channel forearm EMG with a real-time classification distinguishing four different hand gestures (accuracy: 95.9%) as well as the control of a robotic hand using three devices simultaneously.