ML-Based Aging Monitoring and Lifetime Prediction of IoT Devices With Cost-Effective Embedded Tags for Edge and Cloud Operability

The growing number of smart connected devices raises challenges in system reliability. Prediction of failures in Internet of Things (IoT) ecosystems is a significant problem, especially in smart healthcare or financial transfers. In this work, a novel framework and a cost-efficient (power consumption and area) aging tag for the remaining useful life estimation of IoT devices are introduced. Utilizing the proposed comprehensive system prevents failures in large IoT ecosystems. We chose the threshold voltage, a fundamental parameter of transistors as the aging criterion. Also, we considered process, supply voltage, and temperature (PVT) effects on the threshold voltage in the creation or selection of the proper comparison profile. The proposed framework is implemented in three distinct types of cloud machine learning (ML), edge ML, and a combination of proportion estimation and Zscore. Choosing between these three types is a tradeoff between accuracy and minimum resources requirement. A sample tag is fabricated and simulation and fabrication results verified the performance of the proposed system. The cloud ML framework estimated the remaining useful life of various test scenarios with less than 1% error. This means the system estimates the remaining lifetime for the next ten years with two weeks error. Furthermore, the implementation of optimized anomaly detection methods eliminates probable errors of measurement, storage, and data transmission.