A Mixture-Gas Multisensor Interface with On-Chip Classification and On-Edge Regression

Abstract

This paper presents a mixture-gas detectable edgecomputing device with two step on-sensor classification and regression capabilities. Mixture-gas classification is achieved through a proposed analog on-chip AI circuit, and an analog auto-normalization circuit for better AI accuracy is integrated together in the readout integrated circuit (ROIC). For on-edge mixture-gas concentration analysis, a proposed cross-iterative tuning (CIT) regression algorithm is embedded in the edge device. For system-level feasibility, an edge-computing IoT device prototype with metal-oxide-semiconductor (MOS) sensor devices is manufactured based on the ROIC and experimentally verified to achieve 25.9% better classification and 10.9% better regression.