Superparamagnetic Tunnel Junctions for Reliable True Randomness

Stochastic devices have the potential to disrupt computing, revolutionizing low-power machine learning acceleration, probabilistic computing, and hardware security. As implemented, however, superparamagnetic tunnel junctions (sMTJs) face significant challenges including the need for external magnetic fields, and poor reliability and scalability. Here, we present experimental demonstration of three-terminal sMTJs as scalable and reliable sources of true randomness under a field-free regime. By leveraging dual-current controllability and incorporating feedback systems, we substantially enhance the stability and reliability of sMTJ-based systems under varying conditions, even in the field-free regime. Our findings demonstrate the generation of cryptographic-quality random bitstreams and the practical use of sMTJs as efficient and reliable random number generators, successfully integrated into advanced computing algorithms like generative artificial intelligence. Field-free, truly random sMTJs promise to address critical challenges in cryptography, edge computing, and beyond, significantly advancing the field of random number generation.