Engineering-Oriented Design of Drift-Resilient MTJ Random Number Generator via Hybrid Control Strategies
Authors:
Ran Zhang,
Caihua Wan,
Yingqian Xu,
Xiaohan Li,
Raik Hoffmann,
Meike Hindenberg,
Shiqiang Liu,
Dehao Kong,
Shilong Xiong,
Shikun He,
Alptekin Vardar,
Qiang Dai,
Junlu Gong,
Yihui Sun,
Zejie Zheng,
Thomas Kämpfe,
Guoqiang Yu,
Xiufeng Han
Abstract:
Magnetic Tunnel Junctions (MTJs) have shown great promise as hardware sources for true random number generation (TRNG) due to their intrinsic stochastic switching behavior. However, practical deployment remains challenged by drift in switching probability caused by thermal fluctuations, device aging, and environmental instability. This work presents an engineering-oriented, drift-resilient MTJ-bas…
▽ More
Magnetic Tunnel Junctions (MTJs) have shown great promise as hardware sources for true random number generation (TRNG) due to their intrinsic stochastic switching behavior. However, practical deployment remains challenged by drift in switching probability caused by thermal fluctuations, device aging, and environmental instability. This work presents an engineering-oriented, drift-resilient MTJ-based TRNG architecture, enabled by a hybrid control strategy that combines self-stabilizing feedback with pulse width modulation. A key component is the Downcalibration-2 scheme, which updates the control parameter every two steps using only integer-resolution timing, ensuring excellent statistical quality without requiring bit discarding, pre-characterization, or external calibration. Extensive experimental measurements and numerical simulations demonstrate that this approach maintains stable randomness under dynamic temperature drift, using only simple digital logic. The proposed architecture offers high throughput, robustness, and scalability, making it well-suited for secure hardware applications, embedded systems, and edge computing environments.
△ Less
Submitted 19 April, 2025; v1 submitted 25 January, 2025;
originally announced January 2025.
Probabilistic Greedy Algorithm Solver Using Magnetic Tunneling Junctions for Traveling Salesman Problem
Authors:
Ran Zhang,
Xiaohan Li,
Caihua Wan,
Raik Hoffmann,
Meike Hindenberg,
Yingqian Xu,
Shiqiang Liu,
Dehao Kong,
Shilong Xiong,
Shikun He,
Alptekin Vardar,
Qiang Dai,
Junlu Gong,
Yihui Sun,
Zejie Zheng,
Thomas Kämpfe,
Guoqiang Yu,
Xiufeng Han
Abstract:
Combinatorial optimization problems are foundational challenges in fields such as artificial intelligence, logistics, and network design. Traditional algorithms, including greedy methods and dynamic programming, often struggle to balance computational efficiency and solution quality, particularly as problem complexity scales. To overcome these limitations, we propose a novel and efficient probabil…
▽ More
Combinatorial optimization problems are foundational challenges in fields such as artificial intelligence, logistics, and network design. Traditional algorithms, including greedy methods and dynamic programming, often struggle to balance computational efficiency and solution quality, particularly as problem complexity scales. To overcome these limitations, we propose a novel and efficient probabilistic optimization framework that integrates true random number generators (TRNGs) based on spin-transfer torque magnetic tunneling junctions (STT-MTJs). The inherent stochastic switching behavior of STT-MTJs enables dynamic configurability of random number distributions, which we leverage to introduce controlled randomness into a probabilistic greedy algorithm. By tuning a temperature parameter, our algorithm seamlessly transitions between deterministic and stochastic strategies, effectively balancing exploration and exploitation. Furthermore, we apply this framework to the traveling salesman problem (TSP), showcasing its ability to consistently produce high-quality solutions across diverse problem scales. Our algorithm demonstrates superior performance in both solution quality and convergence speed compared to classical approaches, such as simulated annealing and genetic algorithms. Specifically, in larger TSP instances involving up to 70 cities, it retains its performance advantage, achieving near-optimal solutions with fewer iterations and reduced computational costs. This work highlights the potential of integrating MTJ-based TRNGs into optimization algorithms, paving the way for future applications in probabilistic computing and hardware-accelerated optimization.
△ Less
Submitted 8 January, 2025;
originally announced January 2025.
