Spintronic Stochastic Neuron-Based Deep Belief Networks for Image Classification
Aijaz H. Lone, Meng Tang, Daniel N. Rahimi, Divyanshu Divyanshu, Camelia Florina Florica, Selma Amara, Hossein Fariborzi, Gianluca Setti · 2025
Spintronic devices hold promise for AI, particularly in Boltzmann computing architectures, due to their tunable stochastic switching. We demonstrate a SOT-tuned spintronic neuron integrated into a deep belief network for pattern classification. Its switching probability, driven by current, mirrors biological neuron activation through a sigmoid function. Integrated with RBM and logistic regression, the network achieves 97% classification accuracy, stable across temperature variations, underscoring spintronics' potential in robust AI applications.