Self-stabilized true random number generator based on spin–orbit torque magnetic tunnel junctions without calibration

Magnetic tunnel junction (MTJ)-based true random number generators (TRNG), which are promisingly utilized as hardware accelerators for probabilistic computing, may suffer intrinsic performance drifting and fluctuation in the probability space and performance variations in device arrays. In order to preventively avoid these concerns, it is necessary to develop strategies that can help MTJ-based TRNG to be self-stabilized and calibration-free. Here, we derived a simple and easily handleable strategy that only depends on the current sampled state of an MTJ device without any historical or statistic information or a pre-calibrated probabilistic switching property to dynamically tune its sampling probability for the next sampling operation. In this way, the long-term sampling probability of the MTJ can be proved to automatically converge into a desired probability with little fluctuation and drifting. We further experimentally demonstrate such a self-stabilized MTJ-based TRNG. Our results show that the TRNG can generate binary random numbers with any destined probability regardless of its initial state. Furthermore, the TRNG also displays the desired capability of excellently hedging against performance drifting and fluctuations. All these combined allow the self-stabilized TRNG without calibration to be more viable for high-quality entropy source applications.