SNG-TRNG: A Spin-Based Neuromorphic Gaussian TRNG

Algorithms such as Monte Carlo simulations, genetic algorithms, stochastic optimization, Bayesian networks, and generative AI often encounter limitations due to the deterministic nature of von Neumann architectures, underscoring the need for innovative stochastic computing systems. In this context, hardware true random number generators with customizable probability density functions emerge as a vital enabling technology. In this paper, we introduce SNG-TRNG, a novel spin-based neuromorphic true random number generators that leverages spin-transfer torque magnetic tunnel junction as a random entropy source. Using a neural network, we transform uniform random sequences into Gaussian distributions with desired statistical properties. The Kolmogorov-Smirnov test results, with a KS statistic of 0.0088 and a p-value of 0.4251, confirm that the generated random data closely follows a standard normal distribution, demonstrating the high quality of the SNG-TRNG output. This work highlights the potential of spin-based devices in random number generation and paves the way for next-generation stochastic computing systems.