Probabilistic computing considers a range of applications, all which share a requirement for a high volume of samples from different probability distributions. For example, modeling some nuclear physics using Monte Carlo codes results in half their run time spent generating uniform pseudo random numbers, and significant computational overhead transforming those numbers to sample relevant distributions. Thus, there is a benefit in connecting the stochasticity of a device to statistical sampling in a way that makes sampling ubiquitous and cheap, in time and energy, which may further motivate the development of algorithms that continue to shift the burden from calculation to sampling. This talk focuses on statistical analysis of device bitstreams based on their ability to generate quality statistical samples. We focus on understanding the implications of these requirements on two promising devices – magnetic tunnel junctions and tunnel diodes. We conclude with resource estimates for circuits capable of efficiently producing samples for large probabilistic calculations.
–
Presenters
Shashank Misra
Sandia National Laboratories, Sandia National Laboratories, Albuquerque, New Mexico 87185, USA
Authors
Shashank Misra
Sandia National Laboratories, Sandia National Laboratories, Albuquerque, New Mexico 87185, USA
Christopher R Allemang
Sandia National Laboratories
Laura Rehm
New York University, New York University (NYU), Center for Quantum Phenomena, Department of Physics, New York University, New York, NY 10003, USA
Andrew D Kent
New York University, New York University, Department of Physics, Center for Quantum Phenomena, Department of Physics, New York University, New York, NY 10003, USA
Jean Anne C Incorvia
The University of Texas at Austin, University of Texas - Austin
Leslie C Bland
Temple University
Catherine Schuman
University of Tennessee - Knoxville
Suma G Cardwell
Sandia National Laboratories
J. Darby Smith
Sandia National Laboratories, Sandia National Laboratories, Albuquerque, New Mexico 87185, USA
