Related Experiment Video
Updated: Jun 23, 2026
09:05
Flow-pattern Guided Fabrication of High-density Barcode Antibody Microarray
Published on: January 6, 2016
4K Self-Rectifying Resistive Memory Crossbar Array for Reliable Pattern Recognition
Ik Joon Seo1, Dong Chan Lee1, Kanghyeok Jeon2
1Department of System Semiconductor Engineering, Yonsei University, Seoul 03722, Republic of Korea.
ACS Nano
|June 20, 2026
Summary
Researchers developed a novel resistive switching memory with self-rectifying capabilities, eliminating electroforming and improving reliability for large-scale integration. This breakthrough enhances memory performance and enables efficient analogue vector-matrix multiplication.
Area of Science:
- Materials Science
- Electrical Engineering
- Computer Engineering
Background:
- Resistive switching memories (RSMs) offer nonvolatile data storage but often suffer from stochastic variability and require electroforming.
- Integrating RSMs into large crossbar arrays (CAs) presents challenges in yield and operational stability.
- Efficient analogue computation, particularly vector-matrix multiplication (VMM), is crucial for next-generation AI hardware.
Purpose of the Study:
- To demonstrate an interface-controlled, self-rectifying resistive switching memory.
- To integrate this memory into a 4K (64x64) crossbar array (CA).
- To verify its suitability for analogue vector-matrix multiplication (VMM) and elucidate the switching mechanism.
Main Methods:
- Fabrication of a Ru/HfAlOx/TiN stack for nonvolatile resistive switching and rectification.
- Integration of the memory cells into a 4K (64x64) crossbar array (CA).
- Experimental verification of analogue vector-matrix multiplication (VMM) and conductance modulation, supported by current-conduction fitting, drive-level capacitance profiling (DLCP), and atomistic numerical simulations.
Main Results:
- Achieved 100% functional yield in the 4K CA without operational failure.
- Demonstrated interface-controlled operation that suppresses stochastic variability and eliminates the need for electroforming.
- Verified analogue vector-matrix multiplication (VMM) with linear and symmetric conductance updates, suitable for inference simulations.
Conclusions:
- The developed interface-controlled, self-rectifying memory offers a pathway to highly reliable, mass-producible nonvolatile memory.
- The technology enables efficient analogue computation, particularly VMM, addressing key requirements for AI hardware.
- The switching mechanism is elucidated as the motion of internal mobile charges at the interfaces, providing fundamental understanding for future device optimization.
