Estimation of Remaining Useful Life from the Aging Degradation in IC Performance Parameters

This paper addresses the estimation of the remaining useful life (RUL) for the used integrated circuits (ICs). The RUL is a crucial parameter in productizing the second-hand ICs, that may fail earlier than fresh products due to reduced performance margins caused by aging. In estimating the RUL, we estimate the use-period of an IC from its present value of performance parameters and their long-term degradation due to aging. We present a machine-learning based probabilistic estimator to increase the estimation accuracy, by reflecting the statistical significance of the aging against the process variation, and by utilizing the correlation in IC performances. The experimental results are presented on DDR5 DRAM products, with package level performance parameters including valid time window and the max operation frequency. Compared to the conventional method, the proposed method shows 2.4x and 1.8x decrease respectively for the standard deviation and the RMS error in the estimation result.