Enhanced Image and Text Classification Tasks in Neuromorphic Computing Using TMD-Based Memristors

Memristors¹, known for their ability to mimic synaptic behavior, have emerged as promising components for neuromorphic computing systems. These devices offer high-density integration, low power consumption, and non-volatile storage, making them ideal for brain-inspired computational architectures^2–4. This study presents the transition metal dichalcogenide (TMD)-based memristors⁵ for neuromorphic computing applications and demonstrates the application of this device in complex image classification tasks and sentiment analysis using the Behance artistic media (BAM)⁶ dataset. The BAM dataset comprises around 74,000 images featuring diverse and creative visual content from various artists, making it a suitable and challenging example for evaluating the performance of the device. We utilized convolutional neural networks (CNNs) and the pre-trained Visual Geometry Group 16⁷ (VGG16) model for the classification tasks. We employed long short-term memory (LSTM) networks for sentiment analysis and information extraction. To evaluate and optimize the effectiveness of our device, we utilized CVD-grown MoS₂, WS₂, and WSe₂ to carry out a comprehensive series of image classification and annotations specifically targeting the BAM dataset. Detailed results will be presented in the meeting. This research will enable effective, high-performance TMD-based memristors for efficient neuromorphic computing applications.<br/><br/><b>References</b><br/>1. Li, H. <i>et al.</i> Memristive Crossbar Arrays for Storage and Computing Applications. <i>Adv. Intell. Syst.</i> <b>3</b>, 2100017 (2021).<br/>2. Kunwar, S. <i>et al.</i> An Interface-Type Memristive Device for Artificial Synapse and Neuromorphic Computing. <i>Adv. Intell. Syst.</i> <b>5</b>, 2300035 (2023).<br/>3. Liu, X., Zeng, Z. & Wunsch II, D. C. Memristor-based LSTM network with in situ training and its applications. <i>Neural Netw.</i> <b>131</b>, 300– 311 (2020).<br/>4. Smagulova, K. & James, A. A survey on LSTM memristive neural network architectures and applications. <i>Eur. Phys. J. Spec. Top.</i> <b>228</b>, (2019).<br/>5. Teja Nibhanupudi, S. S. <i>et al.</i> Ultra-fast switching memristors based on two-dimensional materials. <i>Nat. Commun.</i> <b>15</b>, 2334 (2024).<br/>6. Michael J. Wilber, Chen Fang, Hailin Jin, Aaron Hertzmann, John P. Collomosse, and Serge J. Belongie. Bam! the behance artistic media dataset for recognition beyond photography. In ICCV, 1211–1220, (2017).<br/>7. K. Simonyan, A. Zisserman, “Very deep convolutional networks for large-scale image recognition”. International Conference on Learning Representations, (2015).