Aging Prediction of Integrated Circuits Using Ring Oscillators and Machine Learning

The reliability of integrated circuits (ICs) is a growing concern with modern electronics. Between the complex design and manufacturing processes of ICs, designs are susceptible to being tampered with, counterfeited, or manufactured with a compromised process technology. This can lead to accelerated aging and shorter lifespans of ICs. In this paper, we propose a light-weight machine learning-based method for approximating future ring oscillator values for aging anomaly detection. Our method relies on an LSTM neural network to predict future ring oscillator frequencies on an FPGA. This prediction is then compared with the measured frequencies to ensure proper aging and reliability of the FPGA. We demonstrate that an increase in the prediction error of ring oscillator frequencies over time shows that there is additional hardware active on the FPGA, i.e., a hardware trojan is present.