Vertical Memristive Crossbar Array for Multilayer Graph Embedding and Analysis

Area of Science:

• Computer Science
• Materials Science
• Network Science

Background:

• Graph data structures model complex relationships, with metal-at-diagonal crossbar arrays (m-CBA) previously used for planar graphs.
• Conventional methods for multilayer graph embedding in high-dimensional spaces face mathematical complexity, computational burden, and information loss.

Purpose of the Study:

• To propose a novel graph embedding method for multilayer graphs using a fabricated vertical m-CBA (vm-CBA).
• To validate the functionality of the vm-CBA with a custom-built measurement system.
• To demonstrate the enhanced performance of vm-CBA for multilayer graph representation.

Main Methods:

• Fabrication of a 3D vm-CBA structure capable of directly mapping multilayer graphs.
• Development and utilization of a custom measurement system for vm-CBA validation.
• Evaluation of the vm-CBA method using link prediction and information scores on real-world datasets.

Main Results:

• The vm-CBA successfully and directly mapped multilayer graphs, representing both inter-layer and intra-layer connections.
• Practical link prediction and information scores showed enhanced accuracy for vm-CBA compared to conventional embeddings.
• The vm-CBA method achieved these results with a significantly decreased number of operations.

Conclusions:

• The proposed vm-CBA offers a more efficient and accurate method for embedding multilayer graphs.
• This 3D vm-CBA approach overcomes the limitations of conventional high-dimensional embedding techniques.
• The findings suggest vm-CBA as a promising hardware-based solution for complex network analysis.