Stochastic Spintronics Device Based Bayesian Networks for Efficient Uncertainty Modeling

Spintronics devices are well-suited for edge computing due to their unique properties that enable efficient and low-power data processing. Edge computing scenarios involve handling large volumes of data from sensors and IoT devices, requiring real-time processing capabilities [1]. By configuring the magnetic tunnel junction (MTJ) characteristics of spintronics devices can be fine-tuned to meet the specific needs of various applications, making them versatile and adaptable solutions for a wide range of computing tasks [2]. The MTJ consists of two ferromagnetic layers: a free layer and a pinned layer (Refer Fig 1). These layers are separated by an non-magnetic layer. The magnetization of the free layer can switch between parallel and anti-parallel orientations relative to the pinned layer, resulting in distinct resistance states(RP and RAP). By utilizing the discrete resistance state of MTJ, a novel hardware computing framework called Probabilistic Spin Logic (PSL) has been introduced (fundamental unit known as Probabilistic Bits (p-bits)) [3]. P-bits can be interconnected to solve a wide array of problems, ranging from optimization tasks to inference processes [4].