We present a hermetically sealed capacitive microelectromechanical system (MEMS) strain sensor designed for arterial pulse waveform extraction using the strain plethysmography (SPG) modality. The MEMS strain sensor features a small form factor of 3.3 mm × 3.3 mm × 1 mm, leverages a nano-gap fabrication process to improve the sensitivity, and uses a differential sensing mechanism to improve the linearity and remove the common mode drift. The MEMS strain sensor is interfaced with an application-specific integrated circuit (ASIC) to form a compact strain sensing system. This system exhibits a high strain sensitivity of 316 aF/µε, a gauge factor (GF) of 35, and a strain sensing resolution of 1.26 µε, while maintaining a linear range exceeding 700 µε. SPG signals have been reliably captured at both the fingertip and wrist using the MEMS strain sensor with high signal quality, preserving various photoplethysmography (PPG) features. Experimental results demonstrate that heart rate (HR) and heart rate variability (HRV) can be estimated from the SPG signal collected at the fingertip and wrist using the sensor with an accuracy of over 99%. Pulse arrival time (PAT) and pulse transit time (PTT) have been successfully extracted using the sensor together with a MEMS seismometer, showcasing its potential for ambulatory BP monitoring (ABPM) application.

Carotid artery disease, including atherosclerotic stenosis and non-atherosclerotic abnormalities, substantially increases ischemic stroke risk and motivates accessible tools for early screening. Current diagnostic pathways rely on clinic-based imaging and skilled operators, creating barriers to frequent monitoring and scalable deployment. We present a non-invasive diagnostic approach using a wearable MEMS accelerometer patch to capture mechano-acoustic vibrations generated by carotid blood flow at the neck. The miniature device integrates a hermetically sealed wideband accelerometer with out-of-plane sensitivity and micro-g resolution to detect subtle flow-induced vibrations. We validated the approach in a carotid flow phantom and a clinical study of 74 patients. Time–frequency representations were computed using the continuous wavelet transform (CWT), from which interpretable spectral and scalogram-derived candidate biomarkers were extracted. Six non-redundant features were then selected for multivariate classification, distinguishing pathology, defined as 50% or greater stenosis or a non-atherosclerotic abnormality, from non-pathology, defined as less than 50% stenosis. Finally, model interpretability was assessed using SHapley Additive exPlanations (SHAP) to quantify the contribution of each biomarker to predicted disease probability. These findings resulted in an AUROC of 0.97 and AUPR of 0.947, with 81.7% sensitivity and 93.6% specificity at the prespecified threshold (precision 85.4%, F1 83.5%, accuracy 89.8%), highlighting the potential of wearable seismic sensing combined with interpretable machine learning for fast screening and longitudinal monitoring of the right and left carotid arteries.