Low Energy Barrier Nanomagnet Design for Binary Stochastic Neurons: Design Challenges for Real Nanomagnets With Fabrication Defects

Much attention has been focused on the design of low energy barrier nanomagnets (LBMs), whose magnetizations vary randomly in time owing to thermal noise, for use in binary stochastic neurons (BSNs) that serve as hardware accelerators for machine learning. The performance of BSNs depends on two important parameters: the correlation time τcassociated with the random magnetization dynamics in an LBM, and the spin-polarized pinning current Ip, which stabilizes the magnetization of an LBM in a chosen direction within a chosen time. We show that common fabrication defects in LBMs make these two parameters unpredictable because they are strongly sensitive to the defects. That makes the design of BSNs with real LBMs very challenging. Unless the LBMs are fabricated with extremely tight control, the BSNs that use them could be unreliable or suffer from poor yield.