Process Variation Sensitivity of Spin-Orbit Torque Perpendicular Nanomagnets in DBNs

Neuromorphic architectures with low energy barrier nanomagnetic devices have been attracting increasing interest over the past few years. More recently, a low barrier nanomagnet (LBNM)-based probabilistic device (p-bit) has been shown to be the basis of neuronal nodes in deep belief networks (DBNs). The LBNMs with perpendicular magnetic anisotropy (PMA) are analyzed and optimized in the interest of achieving stochasticity present in the learning system. In p-bit-based DBNs, several defects, such as variation of the nanomagnet diameter ( $\sigma d$ ), thickness ( $\sigma t_{f}$ ), and anisotropy field ( $\sigma H_{K}$ ), which results in alteration of the fluctuation speed of the p-bit’s nanomagnet can impair functionality. In this article, the accuracy of p-bit-based DBNs is examined under variation of nanomagnet diameter, thickness, and anisotropy field for various tilt angles and temperatures. As evaluated for the MNIST data set for temperature and tilt angle of 300 K and 25°, accordingly, it is shown that the process variation (PV) of $\sigma d$ , $\sigma t_{f}$ , and $\sigma H_{K}$ can be tolerated up to 8%, 23%, and 25%, respectively. A new method is developed to control the fluctuation frequency of the output of a p-bit device by employing a feedback mechanism. The feedback can alleviate this PV sensitivity of p-bit-based DBNs. The compact and low complexity method, which is presented by introducing the self-compensating circuit, can alleviate the influences of PV in fabrication and practical implementation.