Mixed-Signal Classifier Chip on Flexible Substrate for Cardiovascular Health Monitoring

This work presents a fully integrated on-chip classifier on a flexible substrate for on-device detection of atrial fibrillation (AFib) from a electrocardiogram (ECG) signal. The ECG signals are digitized using a 14-bit analog-to-digital converter followed by time-domain feature extraction. The features are provided as inputs to an in-memory computing analog, 3-layer artificial neural network (ANN) for classification. A prototype test-chip is fabricated in 65nm and consumes $58.3 \mu \mathrm{~J} /$ classification while detecting AFib with 99.6% accuracy on Physionet dataset and 86.7% accuracy on participants in a small-scale human subjects study at Mayo Clinic Arizona.