A 65nm compressive-sensing time-based ADC with embedded classification and INL-aware training for arrhythmia detection

Abstract

In-sensor analytics are in high demand to avoid high computation at server back end. Traditional analog sensors require high supply voltage in the range of 1–1.2V even when digital supplies are scaled down to 0.4V-0.6V. We propose a time-based compressed-domain analog-to-digital (ADC) based encoder with parallel processing units for arrhythmia classification. The computationally enhanced ADC performs in-situ compression along with analog to digital conversion. An accuracy of 84% accuracy is achieved with 10.5nJ energy per classification for an 8X compression ratio.