研究目的
Investigating the potential of a fully CMOS compatible, forming-free and non-filamentary memristive device as an artificial synapse for neuromorphic computing applications.
研究成果
The Pd/Al2O3/TaOx/Ta memristive device demonstrates excellent potential as an artificial synapse for neuromorphic computing, with bidirectional analog switching behavior, multilevel conductance states, and nearly linear conductance change. Achieving over 94% recognition accuracy on the MNIST dataset highlights its applicability in neuromorphic systems.
研究不足
The nonlinearity of the weight update in the device can harm neural network performance. The study also notes the challenge of realizing excellent analog SET and RESET processes with satisfying retention time.
1：Experimental Design and Method Selection:
The study involves the fabrication and characterization of a bilayer memristive device (Pd/Al2O3/TaOx/Ta) for synaptic applications. The device's electrical characteristics under DC sweeping mode were evaluated to demonstrate its analog switching behavior and synaptic plasticity.
2：Sample Selection and Data Sources:
The device was fabricated on a Si substrate with Pd and Ta as bottom electrodes, TaOx formed via oxidation, Al2O3 via ALD, and Pd as the top electrode.
3：List of Experimental Equipment and Materials:
Agilent B1500A semiconductor parameter analyzer for electrical measurements, magnetron sputtering for electrode deposition, rapid thermal annealing (RTA) for TaOx formation, and atom layer deposition (ALD) for Al2O
4：Experimental Procedures and Operational Workflow:
The device was subjected to DC sweeping to evaluate its resistive switching behavior. Synaptic characteristics were assessed using non-identical training pulses to optimize conductance change linearity.
5：Data Analysis Methods:
The synaptic weight update was simulated in a two-layer perceptron neural network to estimate recognition accuracy using the MNIST handwritten digit dataset.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
