Electrospinning Machine | Ferroelectric/Antiferroelectric HfZrOx Artificial Synapses/Neurons for Convolutional Neural Network−Spiking Neural Network Neuromorphic Computing

Recently, the team from the School of Integrated Circuits at Shandong University and the School of Microelectronics at Fudan University, through joint research, successfully constructed a novel HfZrOx memristor device based on a ferroelectric-antiferroelectric synergistic mechanism, achieving important progress in the field of low-power neuromorphic computing hardware. The related findings were published in the authoritative international journal Nano Letters under the title "Ferroelectric/Antiferroelectric HfZrOx Artificial Synapses/Neurons for Convolutional Neural Network−Spiking Neural Network Neuromorphic Computing". Jinhao Zhang, a graduate student from the School of Integrated Circuits at Shandong University, and Kangli Xu, a Ph.D. candidate from the School of Microelectronics at Fudan University, are the co-first authors. Researcher Jialin Meng and Researcher Tianyu Wang from the School of Integrated Circuits at Shandong University, and Professor Lin Chen from Fudan University are the co-corresponding authors.

With the rapid development of artificial intelligence, edge computing, and brain-like computing, the "memory wall" problem caused by the "separation of memory and computing" under the traditional von Neumann architecture has become a core bottleneck restricting computing efficiency and energy consumption optimization. Neuromorphic computing, by simulating the event-driven information processing mechanism of biological neurons and synapses, provides a new path to break this architectural barrier. Memristors are widely regarded as important building blocks for brain-like hardware systems due to their multi-state regulation and synaptic plasticity characteristics. However, existing devices often face challenges such as high energy consumption and unstable states. This study proposes a hybrid memristor architecture integrating ferroelectric and antiferroelectric characteristics. The device achieves low-energy consumption programming at low voltage and is adaptable to various neural network models. The research team further constructed and verified its application potential in Convolutional Neural Networks (CNN) and Spiking Neural Networks (SNN) based on this device, demonstrating good recognition accuracy and neural plasticity performance.

Figure 1 CMR application process based on HfZrOx.

Figure 2 Characterization and testing of HfZrOx ferroelectric properties.

Figure 3 Neuromorphic integrate-and-fire behavior and classification performance based on HfZrOx.

Figure 4 Application of ferroelectric-antiferroelectric HfZrOx devices in CMR classification.

The memristor proposed in this study is based on Hf_xZr_1-xO₂ material composition ratio control, achieving a synergistic regulation mechanism of ferroelectric and antiferroelectric effects through precise design, and exhibits excellent stability and endurance (>10⁹ cycles), suitable for large-scale brain-like hardware deployment. The research team further applied it to cardiac magnetic resonance (CMR) image recognition tasks. By constructing CNN and SNN architectures incorporating this memristor model, tests were conducted on large-scale medical image datasets. The results show that the device can support high-accuracy recognition tasks for CMR images, with a recognition accuracy of 92.7%, fully verifying its practical application potential in intelligent medical image processing.

Ferroelectric/Antiferroelectric HfZrOx Artificial Synapses/Neurons for Convolutional Neural Network−Spiking Neural Network Neuromorphic Computing
Jinhao Zhang, Kangli Xu, Lin Lu, Chen Lu, Xinchen Tao, Yongkai Liu, Jiajie Yu, Jialin Meng, David Wei Zhang, Tianyu Wang, Lin Chen*
Original link: https://pubs.acs.org/doi/10.1021/acs.nanolett.5c02889