Research on Aging State Prediction Method of Computing Chip System

With the development of integrated circuits, circuit aging has become an increasingly serious problem, particularly for advanced chiplet. Accurately predicting the aging state of a circuit and replacing it before failure can greatly improve the reliability of electronic systems. However, existing circuit prediction methods suffer from accuracy issues. In this paper, we propose a novel method for predicting circuit aging based on real-time data collected from chiplet in actual system workloads. Specifically, we design an online monitoring system to track the aging state of chiplet, and use an ARMA method to predict circuit aging. Our validation results on FPGA demonstrate the accuracy and reliability of this approach.