Generation of Probabilistic Bits by Exploiting Orthogonal Spin Currents in Magnetic Trilayers.

Probabilistic bits (p-bits), non-deterministic classical bits fluctuating between two digital states, constitute a core element in probabilistic computing. This adeptly addresses computationally complex problems, offering some of the envisioned capabilities of quantum computers while mitigating their major challenges such as shielding, cooling, and scalability. A recent study demonstrated the successful integer factorization using nanoscale magnetic tunnel junction (MTJ)-based p-bits. However, their stochasticity originates from the superparamagnetic properties of nanoscale MTJs, rendering p-bits sensitive to the temperature and the dimensions of the device. Here, we demonstrate reliable p-bits based on stochastic spin-orbit torque (SOT) switching in micrometer-sized magnetic trilayers. In a Fe/Ti/CoFeB structure, two spin currents with orthogonal spin polarizations are generated from the Fe/Ti bilayer, and their magnitude and sign determine SOT switching polarity of the top perpendicular CoFeB layer. This enables systematic control of the probability of having a 'UP' magnetization state of the CoFeB by either an external magnetic field or applied current for SOT switching. Furthermore, by harnessing the generated p-bit streams, we effectively demonstrate invertible AND gate operations and stochastic neural networks with improved energy and area efficiency, thus highlighting the potential utility of our p-bits in device applications.