A noninvasive glucose monitoring system-on-chip (SoC) based on near-infrared (NIR) Photoplethysmography (PPG) is proposed in this paper. The proposed SoC includes a PPG-readout circuit, and a glucose estimation processor (GPP). Six different temporal and spectral features are extracted from the PPG signal after motion artifact and baseline removal from the PPG signal. The GPP utilizes Ensembled Boosted Trees to predict blood glucose levels. The SoC is implemented in 180um CMOS and consumes 208μW with an area of 4.5mm2. It achieves a mean absolute relative difference (mARD) of 5.83% (mARD) verified on 200 subjects. This improves (accuracy/power) figure of merit (FoM) by 5.5% compared to the state-of-the-art work.