A Compact Memristor Model Based on Physics-Informed Neural Networks.

Area of Science:

• Solid-state physics
• Computational materials science
• Device modeling

Background:

• Memristor devices exhibit diverse physical models based on their unique structures.
• Describing memristor physical properties often involves complex differential equations, necessitating a unified approach.

Purpose of the Study:

• To propose a unified, physics-based compact memristor model.
• To leverage physics-informed neural networks (PINNs) for memristor physical analysis and model integration.

Main Methods:

• Developed a compact memristor model utilizing physics-informed neural networks (PINNs).
• Employed PINNs to intuitively solve complex differential equations governing memristor behavior.
• Extracted weights and biases from the PINN for implementation in a Verilog-A circuit simulator.

Main Results:

• The PINN-based model accurately predicts memristor device characteristics.
• Verification using two distinct memristor devices confirmed the model's efficacy.
• Demonstrated the capability of PINNs to extensively integrate various memristor device models.

Conclusions:

• Physics-informed neural networks offer a powerful framework for unifying diverse memristor models.
• The proposed PINN-based approach simplifies memristor physical analysis and enables accurate device prediction.