A Production Line Level MTJ Modeling Framework: Integrating Physical Mechanisms, Experimental Data and Manufacturing Variation

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

We propose for the first time a hybrid neural network-based magnetic tunnel junction (MTJ) modeling framework. By combining graph neural networks (GNN) and deep neural networks (DNN), we implement precise approximation of cross-level coupled physical processes in MTJs, while employing fine-tuning techniques to address discrepancies between experimental data and physical models. The extensive GNN backbone guarantees the scalability for different types of MTJs. Incorporated with stable diffusion (SD) model, this model can accurately predict the MTJ parameter distribution in manufacturing process. Our model framework not only achieves accuracy and scalability but also uniquely reproduces parameter variations in the production line.

Original languageEnglish
Title of host publication2024 IEEE International Electron Devices Meeting, IEDM 2024
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798350365429
DOIs
StatePublished - 2024
Event2024 IEEE International Electron Devices Meeting, IEDM 2024 - San Francisco, United States
Duration: 7 Dec 202411 Dec 2024

Publication series

NameTechnical Digest - International Electron Devices Meeting, IEDM
ISSN (Print)0163-1918

Conference

Conference2024 IEEE International Electron Devices Meeting, IEDM 2024
Country/TerritoryUnited States
CitySan Francisco
Period7/12/2411/12/24

Fingerprint

Dive into the research topics of 'A Production Line Level MTJ Modeling Framework: Integrating Physical Mechanisms, Experimental Data and Manufacturing Variation'. Together they form a unique fingerprint.

Cite this