Research on UFS Endurance Test and Life Prediction Method based on Back-Propagation Neural Network

Universal Flash Storage (UFS) is growing increasingly popular due to its high-speed performance and low power consumption, making it become a preferred choice for mobile devices, especially for high-bandwidth applications such as 5G networks and automotive applications. However, the repetitive program/erase (P/E) cycles and extreme temperature conditions could lead to degradation of devices. Error correction codes (ECC) could correct some memory errors, but uncorrectable errors would result in memory failure or incorrect data storage. Thus, for mitigating these problems, UFS manufacturers set conservative endurance specifications and adopt strict life management methods to ensure the reliability of their chips. Herein, this study aimed to investigate the impact of temperature and P/E cycles on the endurance of UFS devices by conducting endurance tests under different temperature conditions. Furthermore, we proposed a UFS lifetime prediction model based on Backpropagation neural network (BpNN) to predict the lifetime stages of devices from the lifetime characteristic data. The proposed model by our experiments achieved 94.5% accuracy on the test set, which can help users understand the performance of their devices and assist manufacturers to optimize the lifetime management methods for UFS chips.