Hierarchical Multi-Mode Computing in Interlayer-Coupled 3D RRAM Crossbar Arrays

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Engineering

Background:

• Resistive random-access memory (RRAM) offers high density and low power consumption for memory applications.
• Existing RRAM architectures face limitations in operational flexibility and integration density.

Purpose of the Study:

• To present a series-stacked 2-deck RRAM crossbar array with hierarchical multi-mode operation.
• To demonstrate the functionality and performance of this novel architecture for memory and computing tasks.

Main Methods:

• Fabrication of a 2 × 16 × 16 stacked crossbar array with a shared mid electrode.
• Implementation of bias-selective operation for single-layer (1F, 2F) and serial (1F + 2F) modes.
• Utilizing an incremental step pulse with verify algorithm (ISPVA) for multilevel conductance modulation.

Main Results:

• Stable multilevel conductance modulation up to 6-bit resolution (64 states) with retention >10^4 s and endurance >100 cycles.
• Achieved high inference accuracies (93.37-93.38%) on CIFAR-10 classification using a VGG-based CNN.
• Demonstrated logic-in-memory functionality and strong physical unclonable function (PUF) characteristics.

Conclusions:

• The shared mid electrode architecture effectively expands operational space beyond density scaling.
• The proposed RRAM array is suitable for high-performance AI inference and secure PUF applications.
• This work paves the way for more efficient and versatile memory and computing systems.