Answers are grounded in this paper and cite their sources.
Recent progress in deep learning extends the capabilities of artificial intelligence for various practical tasks, making neural networks an extremely popular and useful tool. However, even the most powerful and modern neural networks operate on chips built on the classical von Neumann architecture. At present, extensive research is being conducted to create devices, based on non-von Neumann architecture to radically improve the efficiency of training and operation of neural networks. This area of research includes the development of neuromorphic computing systems and synaptic devices. The paper demonstrates the synaptic behavior of a composite multiferroic heterostructure based on lead zirconate titanate piezoceramics and nickel at resonant excitation of the magnetoelectric (ME) effect and by applying magnetic field pulses of different amplitudes and polarity. The ME coefficient αE was considered as an analogue of the synaptic weight, and the output electrical voltage generated by the heterostructure was interpreted as a postsynaptic potential. The controlling the ME coefficient by changing the remanent magnetization of nickel when applying magnetic field pulses is shown. The possibility of simulating long-term potentiation and depression in the ME heterostructure, as well as spike-timing-dependent plasticity, is demonstrated.
No takes yet. Share an insight, caveat, or question.
Федулов et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: