Generation, Modulation and Application of Spintronic Markov Chain Signal

Stochastic processes are widely used in many real-world fields. As a basic type of stochastic process, Markov chain (MC) represents a stochastic transition model in which the probability of each event depends only on the state attained in the previous state. The hardware generation and arbitrary modulation of MC signals are urgently needed because of the large energy and circuit consumption of existing software methods. However, the hardware MC remains challenging due to the difficulty of modulating randomness in devices. In this work, a novel MC hardware generation method based on a single magnetic tunnel junction (MTJ) device is proposed and further verified via electrical experiments and mathematical derivation. The crucial MC model parameters such as transition matrix and average dwell time (ADT) can be flexibly modulated by changing the probabilistic switching voltage in the proposed waveform. Furthermore, based on the MC generation method and inspired by the divide-and-conquer strategy, a new hardware architecture to estimate eigenvector of an n×n stochastic matrix where n is the power of 2 is proposed using only log₂n MTJ devices. This method provides a new hardware solution for the generation, modulation and application of stochastic signals in semiconductor IC chips.