Using the ANOVA F-Statistic to Isolate Information-Revealing Near-Field Measurement Configurations for Embedded Systems

The analysis of variance (ANOVA) F-statistic is proposed as a tool to isolate near-field measurement configurations that are sensitive to targeted chip processes in embedded systems. It is hypothesized that the desired measurement configurations have high F-values, i.e., the variation in a target process is a major contributor whereas obfuscating background processes and measurement uncertainty are minor contributors to the variance of measured signals. The concept is demonstrated by isolating data-dependent measurement configurations for a commercially available variant of the 8051 micro-controller: First, a multi-stage measurement protocol using F-values is developed to rapidly isolate optimal measurement configurations within the 4-D search space of 2-D probe location over chip area, probe orientation, and time. Then, signals captured using configurations with high F-values are analyzed to identify the Hamming weights of the output data computed by a randomized test code running on the 8051. It is shown that configurations with higher F-values generally result in more accurate classification of the output data; the configuration with the highest F-value results in 100% accuracy.