Fault Syndrome Approach for Measuring System-Level Radiation Effects: A Case Study on Smartphones

Abstract

Measuring radiation effects in computer-based systems is essential to ensure system reliability. System-level radiation effects are usually evaluated based on the architectural vulnerability factor (AVF) characterization of the target system. AVF is the weight of the failure rate of a hardware or software element, indicating the degree to which it contributes to the failure rate of the target system. This value can be obtained by software- or hardware-based fault injection techniques (e.g., simulation of the system architecture model or radiation testing of the device). As an alternative to radiation testing, a field data collection method can be used that analyzes fault records from thousands of devices operating in the field. Regardless of which of these methods are used, measuring system-level radiation effects is a significant effort due to the complexity of the error propagation path in the system. In this dissertation, a fault syndrome approach is proposed as an alternative method of measuring system-level radiation effects. This approach indirectly measures the radiation effects based on an analysis of the target system’s symptoms observed during the radiation test. As a means to determine symptoms, measurement of device current and temperature or observation of abnormal behavior revealed through the device’s screen displaying can be used. Since system errors are not directly detected or diagnosed, this approach has the advantage of minimizing modifications to the target system. As a case study, the measurement and analysis of system-level radiation effects through this approach were performed on smartphone systems. Smartphones were chosen for this study because they have been used as commercial off-the-shelf (COTS) devices in scientific projects requiring high reliability. First, the test results reported in the previous three studies were reviewed according to this approach. In this review, it was noted that system crash failures such as reboots and hangs were common to all target systems. Moreover, these fault symptoms occur with a certain probability regardless of the level of software abstraction, so they can be used as a criterion for evaluating system vulnerabilities. Subsequently, the neutron beam experiment of the Galaxy S7 Edge product specially designed to measure radiation effects with this approach is discussed. During the beam test, measurement equipment and two cameras were used to observe the device current, temperature, and screen image. Through these observations, seven symptoms of malfunction were identified, and it is shown that these symptoms are causally related to a neutron-induced single-event effect (SEE) fault. In this study, the total failure rate of cache memory is used as a key parameter to assess the vulnerability of a target system to neutron SEE. Cache failure rates for the six target systems covered in the three studies reviewed and the Galaxy phone experiment ranged from 100 to 2000 failure-in-time (FIT). Meanwhile, system failure rates were between 2.6 and 71 FIT. The ratio between these two failure rates, the system-to-component FIT ratio, is presented as an indicator of the system’s vulnerability. This evaluation confirms that the mean time between failures (MTBF) is only about 1.4 days for a million smartphones operating on the ground. This case study, conducted on smartphone systems, demonstrates that system-level radiation effects can be measured and evaluated through symptom-based analysis, an approach proposed in this dissertation. In particular, this approach makes the measurement and evaluation work easier as the difficulty of the device modification required is lower than that of conventional approaches. Therefore, this approach can be useful when access to the inside of the target system is severely limited or when minimization of analysis time and cost is required.