Efficient Energy Harvesting for Self-Powered IoT Nodes Using Machine Learning–Driven MPPT and Integrated Power Management

Abstract

The increasing deployment of self-powered Internet of Things (IoT) devices has created a demand for efficient photovoltaic (PV) energy harvesting and power management systems. This paper presents a Physics-Informed Regression-Based Maximum Power Point Tracking (PIR-MPPT) approach integrated with a low-power power management unit for photovoltaic energy harvesting applications. The proposed PIR-MPPT model directly predicts the maximum power point voltage from irradiance and temperature conditions, reducing the computational complexity associated with conventional iterative MPPT techniques. The complete energy harvesting system consists of a voltage-controlled oscillator (VCO), non-overlapping clock generator, ramp charge pump, Banba bandgap reference (BGR), and programmable low-dropout regulator (LDO). The system is modeled using Verilog-A and Cadence Virtuoso and implemented using 45-nm GPDK technology. Simulation results demonstrate accurate maximum power point prediction with an $R^{2}$ value of 0.9082, RMSE of 20.56 mV, and MAE of 9.71 mV. The proposed ramp charge pump boosts the PV input voltage from 1.0–1.35 V to an output voltage range of 3.3–3.5 V with high conversion efficiency. The programmable LDO subsequently generates regulated output voltages of 1.2 V, 1.8 V, and 2.5 V to support diverse low-power IoT loads. Furthermore, the Banba BGR provides a stable 0.8 V reference voltage with excellent temperature stability, while the integrated system achieves reliable end-to-end operation, low output ripple, and efficient power conversion. These results validate the proposed architecture as a promising solution for self-powered IoT sensor nodes and low-power embedded systems.