A Silicon Neuron-based Bio-Front-End for Ultra Low Power Bio-Monitoring at the Edge

Abstract

This paper presents the circuits for an edge-based bio-front-end implemented using an integrate-and-fire silicon neuron model in 22nm SOI CMOS Technology. The proposed implementation encodes both positive and negative input signals separately and, like its biological counterpart, provides asynchronous output. This asynchronous output allows for maximum sensitivity to high-information content input signals and low sensitivity for low-information content. In the proposed design, the firing rate can be controlled by an adaptation circuit to achieve maximum power savings. We demonstrate this design with a sinusoidal test signal and pre-recorded ECG signals. The proposed design achieves ultra-low-power consumption; by applying a sinusoidal input and ECG input the power consumption without adaptation (with adaptation) is 4. 069SnW(3.999nW) and 5. 1529nW (3.311SnW), respectively. In addition, the reconstruction of the ECG signal is demonstrated and the signal to error for the reconstructed ECG signal is 30.2 dB.