&Bullet; physics 14, s51
A device that is normally associated with stable information storage can be modified to be extremely sensitive to thermal noise and provide a random source for probabilistic computing.
Kerem Camsari of the University of California at Santa Barbara and colleagues created what may be the worst computer storage device. Their arrangement of electronic structures – so-called magnetic tunnel junctions (MTJs) – have magnetic states that are so volatile that the states are constantly changing due to thermal noise, so that the MTJs can no longer store information  . But the device is far from useless. The researchers hope that its fluctuating electrical signals could be a valuable resource for probabilistic computing.
An MTJ consists of two nanomagnets separated by an insulating film. Traditionally, one of these nanomagnets is fixed while the other rotates freely. As the magnetizations align, electrons tunnel across the isolation barrier, creating a low resistivity state. When the magnetizations are opposite, tunneling is suppressed and the MTJ assumes a high resistivity state.
MTJs have attracted interest because the stability of their states makes the devices promising platforms for information storage applications. However, the free nanomagnet in each device can be constructed in such a way that the device is sensitive to ambient noise, which causes the states to switch spontaneously under a certain voltage. It has been shown that such “stochastic” MTJs are efficient random sources.
Camsari and colleagues propose a variation on the conventional stochastic MTJ in which both nanomagnets rotate freely. When modeling their MTJ with double free layer, they find that its states fluctuate faster than those of MTJs with fixed layer. Your device also maintains its behavior over a wider voltage range. The team says the symmetrical design should make it easier to mass-produce their MJTs using existing techniques. Recently, they used a series of eight conventional stochastic MTJs to solve a simple factoring problem  . With their new double free layer design, they could use thousands of MTJs to do more ambitious calculations.
Marric Stephens is Corresponding Editor for physics based in Bristol, UK.
- KY Camsari et al., “Magnetic tunnel junctions with double free layer for probabilistic bits” Phys. Rev. Appliedfifteen044049 (2021).
- WA limits et al., “Integer factorization using stochastic magnetic tunnel junctions” nature573390 (2019).