High Performance Computing (HPC) systems have led to advances in practically every area of research: aerospace, automobile, environmental, energy, healthcare, etc. Their enhanced capabilities enable scientific simulations, critical to experimental methods, as never before. As Argonne Laboratory’s HPC prepares to reach Exa-scale output, with many other HPCs already boasting Tera- and Peta-scales, it is imperative to consider what new technological advances will best serve the needs of the HPC community. As one goes to higher outputs the methods for generating the random noise, critical to the simulations that in turn drive scientific innovation, have a significant flaw. IC Chips, used in everyday computers, involve pseudo-random number generation that eventually trace back over the same set of numbers. This makes it impossible to ensure that a simulation has included a fair representation of the dynamical phase-space it attempts to probe. Alternatively, scientist can use NIST tables of vetted random numbers (RN). While the latter provides a work around, as many of the scientist using HPC centers to perform simulations are not computer scientist, these tables can be a source of confusion and lost time. For HPCs to best serve the interest of the scientific community, improvements are needed in the techniques for access and integration of truly-random numbers.

This proposal is to develop a Truly-Random Number Generation (TRNG) system for use in an HPC environment. TRNG, in the proposed design, is accomplished by continuous, micro-splitting of a laser beam through quantum interaction with a birefringent material (Fig. 1a) in an optical cavity. The splitting process eventually leads to random walking of the photons [1-2]. This random walking produces a fluctuation of the energy intensity as the beam exits the cavity and impinges on a detector. In the process, the beam profile gets stretched, making it ideal for parallel application. The ability to produce random values for Monte Carlo styled simulations of dynamics and for mimicking sources of noise will improve the quality of simulations particularly for high outputs (Tera-, Peta- and Exa-scales) [3-5]. The primary technical objective, enhance the quality of simulations performed on HPC systems through a high-speed, parallel, truly-random numbers generator (TRNG), can only be achieved through the use of quantum mechanical processes as described. To tailor a system to the HPC network, the final system will mirror the HPC architecture with only minimal inserted micro-circuitry to provided one stream per processing unit.