Optoelectronic memristor sets new bar for accuracy in path toward neuromorphic vision
Optoelectronic memristor sets new bar for accuracy in path toward neuromorphic vision lead image
Human vision involves a complex interplay of sensing, storage and processing that remains difficult to emulate in an artificial device. Optoelectronic memristors with in-sensor computing architecture are promising candidates to achieve this, with what is called neuromorphic vision.
Shan et al. have demonstrated an optoelectronic synaptic device that dramatically improved recognition accuracy. Based on a zinc oxide-hafnium oxide ZnO/HfOx heterojunction, their memristor achieves higher recognition accuracy than zinc oxide alone.
“Our work focuses on the basic optoelectrical memristive behaviors and the subsequent complex neuromorphic computing functions,” said author Haiyang Xu. “The ZnO/HfOx heterojunction synaptic device exhibits conductance evolution with high linearity and nonvolatile high current state.”
Optoelectronic synapses offer power efficiency, high performance, and scalability, which make them promising for use as memristors. These devices remember the electrical current that previously traveled through them and can retain memory without using power, called its nonvolatile state.
Rapid recombination of electron pairs in materials used in current approaches, however, has limited the ability of such materials to achieve a nonvolatile memristor state.
The group found their heterojunction enabled the effective separation of photo-excited electron pairs, which increases carrier lifetimes, leading to a nonvolatile high-current state.
“More importantly, the artificial vision system based on optoelectronic synaptic devices can achieved a high recognition accuracy of 96.1 percent,” Xu said. “Our work provides a feasible pathway toward the development of optoelectronic synaptic devices for use in high-performance neuromorphic vision systems.”
The group next looks to develop memristors with all-optical modulation as well as construct memristive arrays for more complicated visual functions.
Source: “Optoelectronic synaptic device based on ZnO/HfOx heterojunction for high-performance neuromorphic vision system,” by Xuanyu Shan, Chenyi Zhao, Ya Lin, Jilin Liu, Xiaohan Zhang, Ye Tao, Chunliang Wang, Xiaoning Zhao, Zhongqiang Wang, Haiyang Xu, and Yichun Liu, Applied Physics Letters (2023). The article can be accessed at https://doi.org/10.1063/5.0129642 .
