Experimental Study of Artificial Neural Networks Using a Digital Memristor Simulator

Area of Science:

• Electronic Engineering
• Computer Science
• Materials Science

Background:

• Memristors are crucial for next-generation computing, particularly in artificial neural networks (ANNs).
• Accurate hardware simulation of memristor behavior is essential for designing and testing memristive circuits and ANNs.
• Existing analog solutions often lack compactness, reconfigurability, and precise model matching.

Purpose of the Study:

• To present a fully digital implementation of a memristor hardware (HW) simulator.
• To develop a compact, reconfigurable, and accurate memristor emulator based on a behavioral model.
• To demonstrate the simulator's utility in complex memristive circuits and ANN applications.

Main Methods:

• Behavioral modeling of voltage-controlled threshold-type bipolar memristors.
• Digital circuit design and implementation using field-programmable gate array (FPGA) families.
• Validation using Altera Quartus II and ModelSim tools.
• Testing in associative memory and unsupervised learning ANNs with simplified spike-timing-dependent plasticity (STDP).

Main Results:

• The digital memristor HW simulator is compact, reconfigurable, and shows excellent agreement with the mathematical model.
• Functionality was validated on low-cost FPGAs, demonstrating suitability for complex circuits.
• Successful implementation of synapse functions in ANNs for associative memory and spatiotemporal correlation learning was achieved.
• Full circuit schematics and resource scaling trends are provided.

Conclusions:

• The proposed digital memristor HW simulator offers a viable and efficient alternative to analog solutions.
• The design framework facilitates the development of diverse applications leveraging memristive hardware simulation.
• This work enables advanced research and development in neuromorphic computing and beyond.