Real-Time IC Aging Prediction via On-Chip Sensors

Real-time aging prediction for nanoscale integrated circuits (ICs) is a crucial step for developing prevention and mitigation actions to avoid unexpected circuit failures in the field of operation. Current practices for predicting aging-related performance degradation in ICs consist of recording the operating conditions (e.g. workload, temperature, etc.) throughout ICs' usage time and building a learning model that maps historical operating conditions to actual performance degradation. While some operating conditions such as IC workload can be readily recorded using existing on-chip structures (e.g. registers), other operating conditions such as historical temperature values may not be available for real-time aging degradation prediction. In this paper, we develop a novel real-time IC aging prediction scheme using a set of on-chip sensors that can accurately record historical operating condition parameter values, which will in turn be used for aging-related performance degradation prediction. Experimental results show that by using a machine learning based prediction model and the notion of equivalent aging time, we can achieve accurate aging degradation prediction with the proposed on-chip sensor structure.