Two-Terminal Ferroelectric Artificial Synaptic Devices with Asymmetric Structure

As an excellent nonvolatile memory device, two-terminal ferroelectric devices have been widely used in neuromorphic computing, promoting research on the integration of in-memory sensing and computing. In this study, we proposed a two-dimensional ferroelectric device based on a band engineering strategy and asymmetric interface design, demonstrating 16 stable conductance states that could maintain under 16,000 pulses excitation. The device achieves a switching ratio of 10⁴ by incorporating an hBN dielectric layer between the α-In₂Se₃ channel and source electrode, emulating synaptic typical plasticity with nonlinearity values ≈0.5 for long-term potentiation/depression. The synaptic functionality was simulated through pattern recognition using an artificial neural network, achieving high accuracies of 96.3% for clean 28 × 28-pixel handwritten digit. Besides the handwritten images with one-third occlusion were recognized and reconstructed by using the U-Net structure. The work lays the foundation for the improvement of synapse devices working on high-precision anti-interference recognition systems.