Chair: Prof. Yincai TANG (East China Normal University)
A Yield-reliability Relation Modeling Approach based on Random Effects Degradation Models Dr. Tao YUAN, Xiaoyan ZHU and Yue KUO (Ohio University, University of Chinese Academy of Sciences, Texas A&M University, College Station)
This paper presents a unified modeling framework for yield and reliability in micro-/nano-electronics manufacturing via spatiotemporal modeling of defects. The spatial modeling and temporal modeling of defects refer to modeling of the spatial distribution of defects in manufacturing processes and modeling of the growth of defects with time when devices are subject to stresses, respectively. The defect growth process is characterized by the random-effect degradation modeling method. The presented modeling framework will allow us to use abundantly available process control data to predict the device reliability.
On-line Parameter and RUL Updating for Degradation Processes with Three-source Variability Dr. Weiwen PENG, Yuefeng CHEN and Yuan-Jian YANG (University of Electronic Science and Technology of China, Beijing Special Vehicle Institute, Chongqing University of Science & Technology)
The capability of real-time updating of model parameters and product state based on newly observed condition monitoring observations is of critical importance for an effective degradation based on-line remaining useful life (RUL) predictions. In traditional degradation based RUL prediction, the degradation model and parameter estimation method are both tailored specifically to achieve the real-time updating capability. However, these tailored model and estimation methods can hardly deal with general situations encountered in real practice, where both historical degradation trajectories and real-time condition monitoring needed to be counted in and three-source variability exists as well. In this paper, a general stochastic process based degradation model is introduced in the form of state-space model, where temporal variability, unit-to-unit variability and measurement variability can be simultaneously characterized. The Gaussian process, gamma process, and inverse Gaussian process models with random effect and measurement errors are included as special cases. The Markov chain Monte Carlo (MCMC) and particle filter are coupled to construct a method for parameter estimation and state updating. The MCMC is adopted to fusing the historical degradation trajectories and condition monitoring observations. The posterior samples of model parameters generated from the MCMC is further used as input particles of particle filtering for real-time updating of model parameters and product state simultaneously when newly observations are available. On-line RUL prediction is then implemented through the simulation evolving of the space-state model based on the newly updated model parameters and product state. A simulation study is presented to demonstrate the proposed method.
Bayesian Analysis of Two-Phase Degradation Data Based on Change-Point Wiener Process Dr. Pingping WANG, Yincai TANG, Suk Joo BAE and Yong HE (East China Normal University, Hanyang University, Fudan University)
In degradation test of some products, such as plasma display panels (PDPs) and organic light emitting diodes (OLEDs), observed degradation paths tend to exhibit multi-phase patterns over testing period. In this paper, we propose a change-point Wiener process (CPWP) model to fit the degradation data with two-phase patterns, mainly in Bayesian framework. Hierarchical Bayesian approach is employed to estimate the parameters in this model. Considering distinct degradation behaviors between testing items, degradation rates and change-points are specified as random variables, varying from item to item in the hierarchical Bayesian CPWP. The proposed model is illustrated by a simulation study along with real application to OLED degradation data. The simulation results show that the hierarchical Bayesian approach is superior to the maximum likelihood method. From the analysis of the OLED degradation data, the CPWP model outperforms other existing models in reliability prediction.
Equivalent Accelerated Degradation Test Plans in a Nonlinear Random Coefficients Models Prof. Seong-joon KIM and Suk Joo BAE (Chosun University, Hanyang University)
Design of optimal Accelerated Degradation Testing (ADT) plan has been extensively researched over several decades. In practice, due to the rapidly changing development and assessment environment, pre-established plans often fail to meet reality. Therefore, designing a test plan that is equivalent to the target plan using an different stress-loading or a testing condition is needed to allow for more flexibility. However, there exists currently little work in the development of equivalent ADT plan. In this paper, we proposes an equivalent cost-effective accelerated degradation test (ADT) plan in the context of a nonlinear random-coefficients model. The proposed model is applied to a well-known constant-stress ADT problem in the literature.