Electrospinning Equipment for Research| Near-Sensor Neuromorphic Computing System Based on a Thermopile infraredDetector and a Memristor for Encrypted Visual Information Transmission

Shandong University's Wang Tianyu & North University of China's Lei Cheng in Nano Letters: Memristor-Based Infrared-Sensing Neuromorphic Computing System Enables Data Encryption Through Near-Sensor Computing

Research Background:

The rapid advancement of artificial intelligence (AI) and Internet of Things (IoT) technologies has revealed limitations in traditional von Neumann computing architectures. The physical separation between memory and processing units leads to inefficiencies in handling massive data volumes and high energy consumption. Light, as a primary medium for modern information transmission, is favored for its high-speed transmission capability and security. However, conventional optical detection systems can only detect optical information without processing capabilities, making them inadequate for future demands of increasing data volume and processing requirements. Therefore, developing new optical detection systems that integrate both detection and information processing capabilities is crucial for next-generation information technology.

Research Findings:

Recently, a research team led by Researcher Wang Tianyu and Researcher Meng Jialin from Shandong University, in collaboration with Associate Professor Lei Cheng from North University of China, proposed a near-sensor neuromorphic computing system integrating perception, storage, and computation. The study, titled "Near-Sensor Neuromorphic Computing System based on Thermopile Infrared Detector and Memristor for Encrypted Visual Information Transmission," was published in the renowned international journal *Nano Letters*. Wang Zheng, a graduate student from the School of Integrated Circuits at Shandong University, is the first author, with Researcher Wang Tianyu, Researcher Meng Jialin, and Associate Professor Lei Cheng serving as corresponding authors.

The system consists of a thermopile infrared detector array capable of detecting near-infrared optical information and a memristor array considered a promising candidate for brain-inspired computing. The thermopile infrared detector array captures infrared signals, generating response voltages that are input into the memristor array to produce conductance values, during which weight updates occur. Using this system, the authors first demonstrated high-precision visual neuromorphic computing by improving the recognition rate of MNIST digits from 84.26% to 98.63% after noise preprocessing. Subsequently, they designed a novel information encryption transmission method based on the system and a hardware-software integrated convolutional neural network (CNN), successfully demonstrating its application scenarios.

Graphical Abstract:

Figure 1: Structure and working principle of the near-sensor neuromorphic computing system.

Figure 2: Encrypted information transmission and recognition based on the near-sensor neuromorphic computing system.

Figure 3: Application demonstration of secure encrypted information transmission and computation using the thermopile infrared detector-based near-sensor system.

Summary:

In summary, the researchers proposed a near-sensor neuromorphic computing system composed of a thermopile infrared detector array and a memristor array. The thermopile infrared detector exhibited a broad detection range and fast response capability, while the memristor demonstrated excellent retention, endurance, and neuromorphic computing performance. The team developed an artificial neural network for handwritten digit recognition, improving accuracy from 84.27% to 98.63% through noise reduction by the thermopile infrared detector. The near-sensor neuromorphic computing system was also used for encrypted password transmission and voice-activated secure access, showcasing significant potential for high-precision information transmission and encryption.

Reference Link:https://doi.org/10.1021/acs.nanolett.5c01843