Generating 5 million truly random bits each second

Random number generation is crucial for many fields like cryptography, Monte Carlo simulations, machine learning, and probabilistic computation. However, generating truly random bits generally requires dedicated hardware and significant post-processing, both of which raise costs and limit throughput.

Criss et al. developed a random bit generator based on a magnetic tunnel junction (MTJ) controlled by a field-programmable gate array (FPGA), enabling the direct generation of application-ready random bits without post-processing.

MTJs have been used for decades for data storage, relying on the property whereby applying a high voltage across the junction causes the magnetization direction of one of the two layers to switch, changing the resistance.

“The switching of these devices is inherently stochastic,” said author Andrew Kent. “For a given bias and pulse duration, the junction switches only with a certain probability. By tuning the pulse conditions so that this probability is 50%, the device naturally produces a random binary signal.”

In tests, however, the team found a correlated drift in their device outputs. To correct this, they applied a bias feedback to modulate the write pulse and fed the MTJ output into a circuit that eliminates matching results. With these FPGA-based improvements, the authors generated truly random bits that passed the National Institute of Standards and Technology (NIST) statistical test suite for randomness.

With this approach, the researchers achieved random bit generation at five million bits per second from nanoscale MTJs, demonstrating a scalable entropy source compatible with modern integrated-circuit technology.

The authors are exploring pathways to integrate these MTJ-based entropy sources directly into semiconductor platforms and scale them to larger arrays. Such systems could open new routes to solving complex problems in physics and engineering, from particle physics simulations to combinatorial optimization.

Source: “Magnetic tunnel junction as a real-time entropy source: Field-Programmable Gate Array based random bit generation without post-processing,” by Troy Criss, Ahmed Sidi El Valli, Naomi Li, Andrew Haas, and Andrew D. Kent, Journal of Applied Physics (2026). The article can be accessed at https://doi.org/10.1063/5.0308636 .