Use of Silicon-based Sensors for System Reliability Prediction

More sensors mean better visibility and subsequently increased systems reliability… up to a point. Additional relevant silicon-based sensors require additional software, test cases, and product costs. In this paper we argue that the cost/benefit balance has shifted towards having more sensors with the advent of telemetry mechanisms to carry the sensor readings and machine learning facilities to low-overhead anomaly detection models and resources. The choice of the sensors to use and their association to high-level function-related sensors has demonstrated that silicon-based sensors in system components add important value on top of high-level function-related sensors, such as root cause analysis; early failure prediction, and system reliability prediction.