Recently, the research team led by Professor Huo Nengjie from the School of Electronic Science and Engineering (School of Microelectronics) at the Faculty of Engineering, South China Normal University, has achieved significant progress in the field of neuromorphic in-memory computing devices for marine visual–auditory perception. The related work has been published in Advanced Materials under the title "Neuromorphic In-Memory Computing for Marine Visual–Auditory Perception." Mr. Deng Qunrui, a Ph.D. student (Class of 2022), serves as the first author, with Professor Huo as the corresponding author, and South China Normal University as the primary affiliation.
The research team has proposed a neuromorphic floating-gate transistor (NFT) based on a MoS₂/h-BN/SnSe₂ van der Waals heterostructure. This device uniquely integrates both electrical and optical nonvolatile storage functionalities within a single unit, enabling the emulation of both visual and auditory synaptic behaviors. Consequently, it allows for dual-modal in-memory processing of visuo-auditory signals.
In the electrical domain, the device exhibits outstanding nonvolatile memory characteristics, including a switching speed of 14 µs, an on/off current ratio of 10⁶, an endurance exceeding 10⁴ cycles, and a retention time surpassing 10⁶ seconds. Benefiting from the ultrafast programming speed and high on/off ratio, a single NFT cell can distinguish 16 discrete current states, thereby achieving 4-bit multilevel storage. Furthermore, by integrating an artificial neural network for sonar echo signal processing, the system achieves high-accuracy classification (88%) of submarine minerals and rocks.
In the optical domain, the device enables continuous modulation from short-term to long-term plasticity under 405–808 nm laser pulses, exhibiting canonical synaptic functions such as paired-pulse facilitation (PPF). Leveraging the low-attenuation green-light window in seawater, the system, combined with RGB denoising and green-channel enhancement preprocessing, attains an accuracy of 80% in 23-class marine biological image recognition, which is comparable to that of an ideal software-based model.
This work presents a cross-modal optoelectronic in-memory computing architecture, offering an energy-efficient, low-power, and compact hardware solution for intelligent perception in complex underwater environments.
This research has been continuously supported by the National Natural Science Foundation of China, the Guangdong Provincial Natural Science Foundation, and other funding sources.