Sensorless Power Measurement for Personal Computers Using CMOS MEMS Magnetometers and Machine Learning

We present a novel, sensorless power meter for accurate and real-time power estimation of personal computers (PCs). Our innovative system integrates CMOS MEMS magnetic field sensors with machine learning-based IoT analysis, allowing power consumption to be measured without traditional sensors. At the core of our device is a CMOS MEMS search coil magnetometer, fabricated using the TSMC 1P6M 180nm process and calibrated against a commercial power meter. Coupled with Python software that monitors GPU utilization, CPU usage, and two temperature metrics, we developed a multivariate nonlinear regression (MNR) model with 59 terms to predict power consumption. After an initial training phase, our system can measure the power consumption of any PC sensorlessly, achieving less than a 2.95% discrepancy in average power measurements over 20 minutes. This scalable approach facilitates deployment across multiple PCs or servers in data centers, enabling real-time power reporting and carbon footprint calculations. By efficiently monitoring power usage in diverse computing environments, our method significantly enhances energy management and sustainability efforts.