Low-power reconfigurable MoS2/MoTe2 optoelectronic synapse for visual recognition
image: Optical and electrical signals act as presynapses and modulatory synapses, respectively. The modulatory synapse modulates the built-in electric field by electrostatically modulating the surface adsorption of the material. Artificial synapses enable image self-noise reduction.
Credit: Nano research, Tsinghua University Press
The development of artificial synapses aimed at creating neuromorphological computing systems that are anticipated to fundamentally address the performance bottleneck issues in von Neumann architecture systems. Two-dimensional (2D) materials, with their atomic-scale thickness and van der Waals contact surfaces, offer exceptional optoelectronic properties, making them potential candidates for artificial synapse fabrication.
A team of 2D semiconductor scientists led by Professor Wen Luo at Wuhan University of Technology has recently developed a 2D heterojunction device: a MoS2/MoTe2 photoelectrical synapse that combines low power consumption with advanced visual processing capabilities. With low single-pulse power consumption (0.73 pJ per spike) and excellent long-term cyclic stability (repeatable reconstruction up to 700,000 times), the device far exceeds the performance of similar products.
The team published their research article in Nano research on August 17, 2025.
The key to the device’s performance lies in a smart material design. The team used a planar heterojunction of two-dimensional materials MoS2 and MoTe2, paired with surface charge transfer doping (SCTD)—a technique that modulates electron distribution by leveraging air molecules (O2/H2O) without damaging the material’s structure. “Traditional methods like impurity doping can harm the lattice, while sustained electric fields waste energy,” explains Wen Luo. “Our SCTD approach, controlled by brief gate voltage pulses, avoids both issues, allowing reconfigurable performance with minimal power.”
This design enables the device to mimic biological synapses: it can process light signals (like the retina), store short/long-term “memories” (via adjustable photocurrents), and hardware-level visual self-denoising. When tested on the MNIST handwritten digit dataset, images corrupted by 30% noise were restored to 93.94% recognition accuracy after 60 s of the device’s hardware-level noise reduction—39% higher than without processing.
The device also stands out for its efficiency and durability. Each signal spike consumes only 0.73 pJ—far less than similar devices—and it remains stable through 80,000 seconds of continuous operation and over 700,000 reconfiguration cycles. “This stability is critical for real-world use,” notes Xin Yan, a member of the team. “Imagine a robot navigating a dimly lit room: it needs to process visual data continuously without overheating or failing.”
Looking ahead, the team aims to integrate the synapses into larger neuromorphic systems. Potential applications include low-power visual sensors for autonomous robots, smart medical imaging devices that reduce noise in real time, and energy-efficient AI chips for edge computing.
“This work bridges materials science and neuromorphic engineering,” Luo adds. “By learning from how the human brain processes vision, we’re building devices that are not just powerful, but also sustainable.”
Other contributors include Wen Deng, Niannian Yu, and Xiuying Zhang from the School of Physics and Mechanics, Wuhan University of Technology, China; and Jinsong Wu from the State Key Laboratory of Advanced Processes for Materials Synthesis and Processing, Wuhan University of Technology.
This work was supported by the National Natural Science Foundation of China (52273231).
About Nano Research
Nano Research is a peer-reviewed, open access, international and interdisciplinary research journal, sponsored by Tsinghua University and the Chinese Chemical Society, published by Tsinghua University Press on the platform SciOpen. It publishes original high-quality research and significant review articles on all aspects of nanoscience and nanotechnology, ranging from basic aspects of the science of nanoscale materials to practical applications of such materials. After 18 years of development, it has become one of the most influential academic journals in the nano field. Nano Research has published more than 1,000 papers every year from 2022, with its cumulative count surpassing 7,000 articles. In 2024 InCites Journal Citation Reports, its 2024 IF is 9.0 (8.7, 5 years), and it continues to be the Q1 area among the four subject classifications. Nano Research Award, established by Nano Research together with TUP and Springer Nature in 2013, and Nano Research Young Innovators (NR45) Awards, established by Nano Research in 2018, have become international academic awards with global influence.