A $0.18-\mu \mathrm{g}$ Bias-Instability $0.8-\mu \mathrm{g}/\sqrt{\text{Hz}}$ Resolution Silicon Oscillating Accelerometers with Low-Hysteresis Temperature Extraction and On-Chip Learning-Based Compensation

Authors:

Abstract
Temperature drift remains a major limitation to the long-term stability of silicon oscillating accelerometers. Conventional compensation techniques encounter two critical challenges: (1) hysteresis-induced errors in in-situ temperature measurements, and (2) time- and environment-dependent variations in compensation model coefficients, which hinder complete drift cancellation. This work introduces two key innovations to address these issues. First, a digitally programmable gain amplifier enables real-time measurement of the MEMS resonator's quality factor, allowing low-hysteresis temperature estimation by directly linking the resonator amplitude to its structural temperature. Second, a recursive least squares on-chip learning algorithm adaptively updates the compensation coefficients over time, effectively modeling temperature dependencies and suppressing longterm drift. The proposed readout integrated circuit, fabricated in a $\mathbf{0. 1 8}-\mu \mathrm{m}$ CMOS process, integrates MEMS sensing and compensation functionalities. Experimental results demonstrate a bias instability of $0.18 \mu \mathrm{g}$ and a noise density of $0.8 \mu \mathrm{g}/\sqrt{\mathbf{H z}}$. These results validate the framework's capability for highprecision inertial sensing in UAVs, UUVs, and compact navigation systems operating under dynamic environmental conditions.