Artificial optoelectronic synaptic characteristics of Bi 2 FeMnO 6 ferroelectric memristor for neuromorphic computing

Wen-Min Zhong,Xin-Gui Tang,Qiu-Xiang Liu,Yan-Ping Jiang

DOI: https://doi.org/10.1016/j.matdes.2022.111046

2022-08-21

Abstract:Photoelectric synapse is a new type of artificial synapse that combines light pulse excitation and electrical pulse excitation, and is an important component for the development of neuromorphic hardware. In this work, the Bi 2 FeMnO 6 ferroelectric artificial synapse with optoelectronic response is reported. It can generate an excitatory post-synaptic current through ultraviolet light irradiation and then modulate the excitatory post-synaptic current through ferroelectric polarization, thereby realizing a photo-ferroelectric dual-control synapse. A training method based on ultraviolet excitation and ferroelectric enhancement is proposed, which greatly reduced the number of training pulses and increased the plasticity of the synapse. The optimized spike scheme based on ferroelectricity is proposed, which made the synapse have very linear weight adjustment function. This synapse is used to build a convolutional neural network, which can achieve 98.80% recognition accuracy on the MNIST dataset. In addition, the ferroelectricity and multi-level resistive switching effect of synapse have also been reported.

materials science, multidisciplinary

What problem does this paper attempt to address?

The problem this paper attempts to address is the development of a new type of artificial synapse to improve the performance of neuromorphic computing hardware. Specifically, the paper reports a ferroelectric optoelectronic artificial synapse based on bismuth iron manganese oxide (Bi₂FeMnO₆), which combines the excitation of optical pulses and electrical pulses to achieve optoelectronic dual-control synapses. The main objectives include:

1. **Improving the linearity and plasticity of the synapse**: By utilizing ferroelectric and optoelectronic properties, the number of pulses required for training is reduced, enhancing the plasticity of the synapse.
2. **Reducing energy consumption**: A training method based on ultraviolet excitation and ferroelectric enhancement is proposed, making the synapse more energy-efficient during the training process.
3. **Achieving high-precision handwritten digit recognition**: A convolutional neural network based on this synapse was constructed, achieving a recognition accuracy of 98.80% on the MNIST dataset.

Through experiments and theoretical analysis, the paper demonstrates the potential of this ferroelectric optoelectronic synapse in neuromorphic computing, particularly in applications such as image processing and data storage.