Exploration of Analog Synaptic Plasticity and Convolutional Neural Network Simulation in Bilayer TiO<sub><i>x</i></sub>N<sub><i>y</i></sub>/SnO<sub><i>x</i></sub> Memristor for Neuromorphic Systems
Muhammad Ismail, Doohyung Kim, Eunjin Lim, Maria Rasheed, Chandreswar Mahata, Yeongkyo Seo, Sungjun Kim
Abstract
In this study, a TiN/SnO 2 /Pt sandwich structure is explored for its dual functionalities in electronic synapses and multistate memory. The SnO 2 layer is fabricated via reactive sputtering, leading to the formation of a TiN/TiO x N y /SnO x /Pt memristor. This configuration, confirmed by HRTEM and XPS analyses, exhibits several advantageous features: consistent bipolar nonvolatile switching at low operating voltages, endurance up to 500 cycles, an on/off ratio of ∼10 2, and robust data retention. Set and reset times are approximately 300 and 400 ns, with energy consumption of 3.24 nJ and 3.26 nJ, respectively. The memristor achieves multilevel resistance states, simulating synaptic behaviors such as LTP/LTD, SADP, PPF, and PPD. Utilizing LTP and LTD data, CNN simulation achieved 91.3% recognition accuracy, surpassing the 70.5% accuracy of ANN simulation. These findings suggest the TiN/TiO x N y /SnO x /Pt memristor’s potential for artificial neural network applications.