Exploring the Potential of a 1T-1M HfO<sub>X</sub>-Based Resistive Switching Device for Articulating Artificial Neural Network Hardware
In this work, we investigate the pattern recognition accuracy and energy consumption of crossbar architectures based on HfO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X</inf>-based one-transistor one-memristor (1T-1M) cells, utilizing a fully calibrated analytical model. Our results highlight the uniform and well separated low resistance states (LRS) with HfO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X</inf>-based 1T-1M cells provide excellent noise-immune operations with a recognition accuracy of approximately 47% in the presence of a 26% noise level. Further, the HfO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">X</inf>-based crossbar array maintains an average energy consumption of around 18 nJ during data storage within each cell. Finally, the artificial neural network designed with 1T-1M demonstrates the 90% recognition capabilities in handwritten MNIST digits.
Authors
- Brajesh Rawat (ORCID: https://orcid.org/0000-0001-9097-8050)
- Venkata Kalyan Tavva (ORCID: https://orcid.org/0000-0001-6596-2292)
- Mani Shankar Yadav
- Sushma Gupta (ORCID: https://orcid.org/0000-0002-5382-4383)
Institutions
- Indian Institute of Technology Ropar (IN)
Publication Details
- Published
- 2024-10-09
- DOI
- https://doi.org/10.1109/cas62834.2024.10736761
- Primary Topic
- Neural Networks and Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00