From Memristive Materials to Neural Networks.

Area of Science:

• Materials Science
• Computer Engineering
• Artificial Intelligence

Background:

• Moore's Law is reaching physical and architectural limits, necessitating new computational paradigms.
• Memristors, with their unique properties, are emerging as a key technology for future computing.
• The dual functionality of data storage and computation in memristors addresses current efficiency challenges.

Purpose of the Study:

• To review memristor-based computing technologies from materials to applications.
• To explore the potential of memristors in artificial intelligence and advanced computing.
• To discuss current challenges and future prospects of memristor technology.

Main Methods:

• Review of memristor materials and mechanisms.
• Analysis of memristor applications in synapse simulation.
• Investigation of memristor-based in-memory logic computing, deep neural networks (DNNs), and spiking neural networks (SNNs).

Main Results:

• Memristors demonstrate potential for high-performance computing and AI due to their simple structure, low power consumption, and CMOS compatibility.
• Successful simulation of synaptic behavior and implementation in DNNs and SNNs.
• In-memory computing architectures leveraging memristors show promise for enhanced data processing efficiency.

Conclusions:

• Memristors are crucial for overcoming post-Moore's Law limitations.
• Continued research into memristor materials and device physics is vital.
• Memristor technology holds significant promise for future AI and high-performance computing applications.