Yield Diagnosis and Tuning for Emerging Semiconductors During Research Stage

The process of taking a new semiconductor device from the lab to the factory involves a lot of time, funds and manpower, a large portion of which is spent on device yield improvement. In recent years new methods have been tried to rapidly improve yields and using machine learning (ML) algorithms is...

Full description

Saved in:
Bibliographic Details
Main Authors: Chunshan Wang, Zizhao Ma, Yuxuan Zhu, Chensheng Jin, Dongyu Chen, Chuxin Zhang, Yining Chen, Wenzhong Bao, Yufeng Xie
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10975036/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The process of taking a new semiconductor device from the lab to the factory involves a lot of time, funds and manpower, a large portion of which is spent on device yield improvement. In recent years new methods have been tried to rapidly improve yields and using machine learning (ML) algorithms is one option. However, they usually require a large dataset, which is often unavailable at the device research stage, emerging semiconductors (e.g., 2D materials) are extremely costly to pilot. In this paper, we propose a yield diagnosis and tuning scheme based on ensemble learning and Bayesian optimization, which demonstrate outstanding performance even with a limited data volume. We use real 2-D semiconductor device fabrication process data for scheme evaluation. Experimental results show that the algorithm for yield prediction has achieved regression fitting results whose mean absolute error (MAE) is no more than 8 points and explained variance (EVAR) is no less than 0.62, this indicates that the model fits well on this dataset. We also remanufactured a batch of devices based on the yield tuning recommendations to validate the effectiveness of our approach. The test results indicated a final yield score of 86 points, after evaluating several key indicators such as mobility and hysteresis, resulting in a 62% improvement.
ISSN:2169-3536