Efficient Programmable Random Variate Generation Accelerator From Sensor Noise

Abstract

We introduce a method for nonuniform random number generation based on sampling a physical process in a controlled environment. We demonstrate one proof-of-concept implementation of the method, that doubles the speed of Monte Carlo integration of a univariate Gaussian. We show that we must measure and compensate for the supply voltage and temperature of the physical process to prevent the mean and standard deviation from drifting. The method we present and our detailed empirical hardware measurements demonstrate the feasibility of programmable nonuniform random variate generation from low-power sensors and the effect of ADC quantization on the statistical qualities of the approach.