Zhang Ran, Li Xiaohan, Zhao Mingkun, Wan Caihua, Luo Xuming, Liu Shiqiang, Zhang Yu, Wang Yizhan, Yu Guoqiang, Han Xiufeng
Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Beijing, 100190, China.
Songshan Lake Materials Laboratory, Dongguan, Guangdong, 523808, China.
Adv Sci (Weinh). 2024 Jun;11(23):e2402182. doi: 10.1002/advs.202402182. Epub 2024 Apr 15.
The incorporation of randomness into stochastic computing can provide ample opportunities for applications such as simulated annealing, non-polynomial hard problem solving, and Bayesian neuron networks. In these cases, a considerable number of random numbers with an accurate and configurable probability distribution function (PDF) are indispensable. Preferably, these random numbers are provided at the hardware level to improve speed, efficiency, and parallelism. In this paper, how spin-orbit torque magnetic tunnel junctions (SOT-MTJs) with high barriers are suitable candidates for the desired true random number generators is demonstrated. Not only do these SOT-MTJs perform excellently in speed and endurance, but their randomness can also be conveniently and precisely controlled by a writing voltage, which makes them a well-performed Bernoulli bit. By utilizing these SOT-MTJ-based Bernoulli bits, any PDF, including Gaussian, uniform, exponential, Chi-square, and even arbitrarily defined distributions can be realized. These PDF-configurable true random number generators can then promise to advance the development of stochastic computing and broaden the applications of the SOT-MTJs.
将随机性融入随机计算可为诸如模拟退火、非多项式难题求解和贝叶斯神经网络等应用提供大量机会。在这些情况下，大量具有精确且可配置概率分布函数（PDF）的随机数是必不可少的。优选地，这些随机数在硬件层面提供，以提高速度、效率和并行性。本文展示了具有高势垒的自旋轨道扭矩磁性隧道结（SOT-MTJ）如何成为所需真随机数发生器的合适候选者。这些SOT-MTJ不仅在速度和耐久性方面表现出色，而且其随机性还可以通过写入电压方便且精确地控制，这使其成为性能良好的伯努利比特。通过利用这些基于SOT-MTJ的伯努利比特，可以实现任何PDF，包括高斯分布、均匀分布、指数分布、卡方分布，甚至任意定义的分布。这些可配置PDF的真随机数发生器有望推动随机计算的发展，并拓宽SOT-MTJ的应用范围。
