WO_x channel engineering of Cu-ion-driven synaptic transistor array for low-power neuromorphic computing

Area of Science:

• Materials Science and Engineering
• Neuroscience and Computing

Background:

• Artificial neural networks (ANNs) require energy-efficient synaptic devices for cognitive computing.
• Large-scale synaptic arrays necessitate low-current regimes for reduced power consumption.

Purpose of the Study:

• To develop a three-terminal synaptic transistor array with analogously modulated channel current states in the nanoampere range.
• To enable low-power, high-performance ANNs through engineered synaptic devices.

Main Methods:

• Fabrication of a CuO_x/HfO_x/WO₃ synaptic transistor array using reactive sputtering and annealing.
• WO₃ channel engineering with controlled oxygen and nitrogen incorporation.
• Temperature-dependent studies to identify transport mechanisms (Poole-Frenkel conduction).
• Demonstration of synaptic characteristics in a 7x7 array with parallel programming.

Main Results:

• Engineered WO₃ channels achieved analog current modulation in the tens of nanoamperes.
• Optimized annealing and gas incorporation enabled stable, low-current synaptic operations.
• Channel current modulation correlated with Cu ion movement and gate current response.
• Achieved 90% recognition accuracy on the Fashion-MNIST dataset using the synaptic transistors in a multilayer neural network.

Conclusions:

• The proposed CuO_x/HfO_x/WO₃ synaptic transistor array effectively mimics synaptic behavior for energy-efficient ANNs.
• WO₃ channel engineering is crucial for achieving precise low-current modulation.
• The device shows promise for next-generation neuromorphic computing applications.