Arbitrary Modulation of Average Dwell Time in Discrete-Time Markov Chains Based on Tunneling Magnetoresistance Effect

Stochastic processes (SPs) are widely used in many real-world fields, especially AI algorithms and models. A discrete-time Markov chain (DTMC) is a fundamental SP where the probability of each event depends only on the state attained in the previous event. DTMC is extensively used in signal processing and information theory, but the hardware generation of DTMC remains hindered by the difficulty in arbitrarily modulating the averaged dwell times (ADTs), i.e. the average time that the DTMC stays at one state. In this letter, we propose a two-step procedure to modulate the ADTs of a DTMC generated from single magnetic tunnel junction (MTJ), without being limited by additional restrictions such as a fixed ratio between the ADTs, widening the SP applications scope of the MTJ-based DTMC. This method has been verified via mathematical derivation and electrical characterization. The generation throughput and power consumption can also be conveniently modulated. This procedure provides a new hardware solution for the generation of stochastic signal in semiconductor IC chips.