Variations in Stochastic Magnetic Tunnel Junctions

Probabilistic computing, as a natural computing paradigm, leverages intrinsic device‐level physical randomness to implement tunable stochastic bits, for which stochastic magnetic tunnel junctions offer a promising hardware realization. In this paper, we provide a detailed examination of the fabrication process of stochastic magnetic tunnel junctions, including the stack deposition, the formation of nano‐dots using the electron‐beam lithography, as well as the development of an undercut resist profile designed to improve pattern definition and facilitate a reliable lift‐off process. Furthermore, we identify and discuss several critical sources of variations, specifically retention‐time variation, state‐distribution variation, and variations related to the fabrication, observed during electrical and magnetic characterization, all of which significantly influence both the fabrication outcome and the resulting device performance. Our findings offer a refined experimental methodology for making the stochastic magnetic tunnel junction and future probabilistic computing applications.