A Wide Dynamic Range Optical Particulate Matter Sensor With On-Chip Machine Learning Calibration

초록

This article presents an optical particulatematter (PM) sensor featuring on-chip machine learning (ML)calibration to enhance accuracy across diverse particlesizes. A programmable gain amplifier (PGA) with dynamicrange extension is integrated to improve the precision of lowPM concentration measurements (<30 mu g/m(3)), fully utilizingthe analog-to-digital converter's (ADC) dynamic range. Thesensor employs an ML calibration approach combining leastmean square (LMS) error and singular value decomposition(SVD)-based pseudoinverse matrix to mitigate variations inmodules and components, significantly enhancing systemreliability. These innovations enable accurate classificationof four PM categories: PM1.0, PM2.5, PM4.0, and PM10.0.For particle concentrations above 30 mu g/m(3), the maximummeasurement error is+8.85%/-8.84%; for concentrationsbelow 30 mu g/m(3), the error is within+2.9 mu g/m(3)/-2.6 mu g/m(3).Fabricated using a standard 0.13 mu m CMOS process, the5.80 mm2read-out integrated circuit compensates formodule-level and component-level variations, ensuringrobust and reliable performance.

키워드