Ferrimagnet-Based\nNeuromorphic Device Mimicking the\nVentral Visual Pathway for High-Accuracy Target Recognition

The\nventral visual pathway (VVP) of the human brain efficiently\nimplements target recognition by employing a deep hierarchical structure\nto build complex visual concepts from simple features. Artificial\nneural networks (ANNs) based on spintronic devices are capable of\ntarget recognition, but their poor interpretability and limited network\ndepth hinder ANNs from mimicking the VVP. Hardware implementation\nof the VVP requires a biorealistic spintronic device as well as the\ncorresponding interpretable and deep network structure, which have\nnot been reported so far. Here, we report a ferrimagnetic neuron with\na continuously differentiable exponential linear unit (CeLu) activation\nfunction, which is closer to biological neurons and could mitigate\nthe issue of limited network depth. Meanwhile, we also demonstrate\nthat a ferrimagnet can construct artificial synapses with high linearity\nand symmetry to meet the requirements of weight update algorithms.\nBased on these neurons and synapses, we propose an all-spin convolutional\nneural network (CNN) with a high interpretability and deep neural\nnetwork, to mimic the VVP. Compared to the state-of-the-art spintronic-based\nneuromorphic computing model, the CNN with bionic function, using\nexperimentally derived device parameters, achieves high recognition\naccuracies of over 91% and 98% on the CIFAR-10 datasets and the MNIST\ndatasets, respectively, showing improvements of 1.13% and 1.76%. Our\nwork provides a promising method to improve the bionic performance\nof spintronic device-based neural networks.