Mortise-tenon-shaped memristors for scientific computing.

Area of Science:

• Materials Science
• Computer Engineering
• Computational Science

Background:

• In-memory computing using memristors offers parallel processing for scientific computing.
• Memristor nonuniformity necessitates complex peripheral circuits, increasing power consumption and limiting practical applications.

Purpose of the Study:

• To address memristor nonuniformity and enhance in-memory computing efficiency for scientific applications.
• To develop a novel memristor structure with improved uniformity and performance.

Main Methods:

• Fabrication of a mortise-tenon-shaped (MTS) memristor by integrating a mortise-shaped hexagonal boron nitride (h-BN) flake onto a HfO2 switching layer.
• Characterization of MTS memristor uniformity, including cycle-to-cycle and device-to-device variations.
• Implementation of MTS memristors in a partial differential equation solver to evaluate performance.

Main Results:

• The MTS memristor demonstrated ultrahigh uniformity with significantly reduced cycle-to-cycle (~2.5%) and device-to-device (~6.9%) variations compared to standard HfO2 memristors.
• A partial differential equation solver built with MTS memristors achieved a five-times faster convergence speed for solving the Poisson equation than traditional memristor-based solvers.

Conclusions:

• The MTS memristor structure effectively mitigates nonuniformity issues in memristor-based in-memory computing.
• This approach offers a promising solution for reducing hardware resources and enhancing the speed and accuracy of scientific computing.