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

초록

This article presents an optical particulate matter (PM) sensor featuring on-chip machine learning (ML) calibration to enhance accuracy across diverse particle sizes. A programmable gain amplifier (PGA) with dynamic range extension is integrated to improve the precision of low PM concentration measurements (<30 μg/m3), fully utilizing the analog-to-digital converter's (ADC) dynamic range. The sensor employs an ML calibration approach combining least mean square (LMS) error and singular value decomposition (SVD)-based pseudoinverse matrix to mitigate variations in modules and components, significantly enhancing system reliability. These innovations enable accurate classification of four PM categories: PM1.0, PM2.5, PM4.0, and PM10.0. For particle concentrations above 30 μg/m3, the maximum measurement error is +8.85%/-8.84%; for concentrations below 30 μg/m3, the error is within +2.9 μg/m3/-2.6 μg/m3. Fabricated using a standard 0.13 μm CMOS process, the 5.80 mm2 read-out integrated circuit compensates for module-level and component-level variations, ensuring robust and reliable performance. © 2025 IEEE.

키워드