Local environmental variability monitoring using hypothesis tests

Modern electronic embedded devices have to run more and more quickly while consuming less and less energy. In distributed architectures (MultiProcessor System On Chip for example), the supply voltage and the operating frequency of each processor is usually tuned dynamically to obtain an efficient performance/power consumption trade-offs. Then, to locally adapt efficiently the parameters of the circuit, it is necessary to estimate its physical state in particular the current supply voltage and temperature values. In this paper a method based on statistical tests is proposed to estimate the supply voltage and temperature of a local area of an integrated circuit. The measurements come from standard ring oscillators (ROs) integrated on-chip and the results show the effectiveness of the method with mean estimation errors of 5mV and 7.5°C. The unsensitivity of the method face to IR drops events is also exhibited.