Microwave artificial neurons based on magnetic tunnel junction nano-oscillators for image recognition and denoising
Abstract We report a microwave artificial neuron based on magnetic tunnel junction spin torque nano-oscillator (STNO). Based on STNO neurons, the three-layer full connected artificial neural network (ANN) are constructed to recognize handwritten digit with a produced accuracy of up to 90.95%. To address noise inherent in real circuits leads to poor recognition accuracy, we harness the frequency mutation characteristics of the STNO output near the critical threshold current for denoising handwritten digits corrupted by noise. Before and after denoising using the aforementioned ANN, the recognition accuracy is improved from 55.98% to 88.76%.
Authors
- Bin Fang (ORCID: https://orcid.org/0000-0002-6916-587X)
- Yanxiang Luo (ORCID: https://orcid.org/0000-0001-5288-0677)
- Junwei Zeng (ORCID: https://orcid.org/0000-0002-9456-1583)
- Kexin Zeng (ORCID: https://orcid.org/0000-0002-1862-6728)
- Zhongming Zeng (ORCID: https://orcid.org/0000-0001-7240-2058)
- Zhenhao Liu (ORCID: https://orcid.org/0009-0004-9909-5528)
- Like Zhang (ORCID: https://orcid.org/0000-0002-0869-3417)
Institutions
- National University of Defense Technology (CN)
- Nanjing University of Information Science and Technology (CN)
- Suzhou Institute of Nano-tech and Nano-bionics (CN)
- National Supercomputing Center in Wuxi (CN)
- Wuxi Institute of Technology (CN)
Publication Details
- Journal
- Japanese Journal of Applied Physics
- Published
- 2024-10-01
- DOI
- https://doi.org/10.35848/1347-4065/ad8418
- Primary Topic
- Advanced Research in Systems and Signal Processing
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China