Model construction and parameter effect for TFT-LCD process based on yield analysis by using ANNs and stepwise regression
Introduction
The market for liquid crystal displays (LCDs) is known as a growing rapidly and impacting new fields. The primary applications of LCDs include personal digital assistants (PDAs), cellular phones, digital cameras, computers, notebooks, flat panel TVs and various computer game units. During the past several years, the market for LCDs has grown at over 20% on average per annum. The speculative demand increase has driven capacity expansion, particularly in South Korea, Japan and Taiwan (Su, Yang, & Wang, 2004). The price for LCD products is significantly reduced due to both the technology maturity and ample manufacturing capacity. The downward pricing trend further promotes LCD applications. LCDs can be divided into three major products including TN (twisted nematic), STN (super twisted nematic) and TFT (thin film transistor). The most widely used LCD for high information content display is the TFT-LCD. In the TFT-LCD each picture pixel is controlled using a thin film transistor. The TFT-LCD panel has a sandwich structure (Singer, 1994) consisting of two glass plates with liquid crystal material in between. The bottom substrate is the TFT array. The top substrate is the color filter plate. Color filter glasses are usually purchased from outside vendors.
The manufacturing technology, capital investment and industrial infrastructure are key factors affecting LCD industry competition (Su et al., 2004). The ability to improve yield in the manufacturing process is an important competitiveness determinant for LCD factories due to the significant yield loss ranging from 5% to 25%. This loss is attributed to three major manufacturing sectors: the array, cell and module assembly processes. The yield loss from the cell process is one of the most critical steps. To increase cell process yield, more conforming LCD panels must be produced from one glass substrate. However, no any suitable theories were proposed to study the real yield problem and the possible opportunity of improvement to manufacturing process will be omitted. In order to survival during the competitive environment, how to mine the useful information from the “know how or domain knowledge” of manufacturing process will be an important issue to all enterprises. Hence, most manufacturers provide more resources to study such issue. Besides, a flexible model construction in TFT-LCD industry should be another consideration due to the initial development stage of yield analysis. From the systematical viewpoint, several parameters (e.g. the setting of process condition) will significantly affect the result of yield model. How to keep the knowledge about the effect on yield for those parameters will be also another importance issue. From the considerations mentioned above, we will apply the artificial neural networks (ANNs) into the yield model construction due to there were many successful applications; and the necessary statistical technique will also be applied into the analysis of parameter effect. A procedure to achieve the model construction and parameter effect for yield analysis in TFT-LCD industry was proposed in this study. In order to verify the rationality and feasibility of our approach, an illustrative example owing to TFT-LCD manufacturer at Tainan Science Park in Taiwan will be also chosen in this study.
Section snippets
Stepwise model-building technique
Stepwise model-building techniques for regression designs with a single dependent variable had been described in numerous sources (e.g., see Darlington, 1990, Hocking, 1996, Lindeman et al., 1980, Morrison, 1990, Neter et al., 1985, Pedhazur, 1982, Stevens, 1986, Younger, 1985). The basic procedures of Stepwise model-building will involve: (1) identifying an initial model; (2) repeatedly altering the model at the previous step by adding or removing a independent variable (or process parameters)
Proposed approach
Generally, a particular relationship will exist among the input variables and output variables of a system, e.g. the functional relationship or statistical relationship. From mathematical viewpoint, the logical relationship can be constructed by modeling techniques. Generally, system’s output can be viewed as a function of system’s input. Hsieh, 2006, Tong and Hsieh, 2001, Su and Heish, 1998, Ko et al., 1998 had applied the BPNN to model this logical relationship to achieve quality
Illustrative example
A TFT-LCD manufacturer’s data at Tainan Science Park in Taiwan will be taken as an example to demonstrate our proposed procedure. Basically, TFT-LCD manufacturing processes can be mainly divided into three parts (in Fig. 1) and they can be described as follows:
- 1.
Array assembly process (Array): It will grow thin films on glass substrates and produce thin film transistors (TFT). The Array manufacturing processes are similar with the one frequently seen in semiconductor processes.
- 2.
Cell assembly
Concluding remarks and recommendations
After applying a real example to demonstrate the proposed procedure, we find out that the yield analysis will make the practitioners to understand their domain knowledge or manufacturing core. The advantage of the proposed procedure can be summarized as
- (1)
The yield model can be constructed based on the real manufacturing consideration without any mathematical equation or computation. Such non-formula model will rapidly provide the related information to the engineers. Subsequently, the engineers
Acknowledgement
The author would like to thank the financial support from National Science Center with the Contract Number of NSC94-2416-H-143-001.
References (17)
- et al.
Methodology of perform design considering workability in metal forming by the artificial neural network and Taguchi method
Journal of Materials Processing Technology
(1998) Regression and linear models
(1990)Methods and applications of linear models. Regression and the analysis of variance
(1996)Process improvement in the presence of qualitative response by combining fuzzy sets and neural networks
Integrated Manufacturing Systems
(2001)- Hsieh, K.L., (2006). Parameter optimization of a multi-response process for lead frame manufacturing by employing...
- et al.
Introduction to bivariate and multivariate analysis
(1980) Multivariate statistical methods
(1990)- et al.
Applied linear statistical models: regression, analysis of variance, and experimental designs
(1985)
Cited by (21)
Effect of variable selection algorithms on model performance for predicting moisture content in biological materials using spectral data
2022, Analytica Chimica ActaCitation Excerpt :However, SWR procedures are relatively cheap computationally and easy to implement. SWR comprises three main stages: (i) identifying an initial model by a forward selection of input predictors/variables so that the selected variables at each step minimize the residual sum of squares; (ii) repeatedly altering the model at the previous step by forward inclusion or backward elimination of variables following the “stopping criteria; ” and (iii) terminating the search when the optimized model is established [42]. The stepping criteria employed for entry and removal were based on different significance levels (α).
Rapid and non-destructive detection of chicken adulteration in minced beef using visible near-infrared hyperspectral imaging and machine learning
2016, Journal of Food EngineeringCitation Excerpt :In this method, the number of predictors retained in the final model is determined by the levels of significance assumed for inclusion and exclusion of predictors from the model. Basically, stepwise regression comprises three main stages: (i) identifying an initial model; (ii) repeatedly altering the model at the previous step by forward adding or backward removing an independent variable in accordance with the “stepping criteria”; (iii) terminating the search when the optimized model is established (Hsieh and Lu, 2008). The stepping criteria employed for entry and removal were based on the significance level of 5%.
Determining manufacturing parameters to suppress system variance using linear and non-linear models
2012, Expert Systems with ApplicationsCitation Excerpt :The process of producing a TFT-LCD is roughly separated into three main steps, as shown in Fig. 1: (i) the TFT and CF are produced using a process similar to that used for making semiconductors; (ii) the cells are made by combining a TFT panel with a CF panel, and then injecting Liquid Crystal (LC) into the chamber; (iii) the modules are formed, which involves assembling polarizers and backlights. Pan and Hsieh study the cutting quality for TFT-LCD glass substrate to improve the cell yield (Hsieh, 2010; Hsieh & Lu, 2010; Li et al., 2010). In the case, we focus on the manufacturing parameters of the last station in the CF manufacturing process that is a flow shop production type process.
Predicting the development cost of TFT-LCD manufacturing equipment with artificial intelligence models
2010, International Journal of Production EconomicsCitation Excerpt :As a typical AI technique, ANNs are feasible alternatives for early cost prediction as ANNs can model complex systems given a minimal amount of data. Many studies have verified the predictive power and excellent accuracy of ANNs in process vessels (Caputo and Pelagagge, 2008), the automotive industry (Cavalieri et al., 2004), sheet metal parts (Verlinden et al., 2008), TFT-LCD yield (Hsieh and Lu, 2008), machine-tool selection (Ciurana et al., 2008), software development (De Barcelos Tronto et al., 2007), product lifecycle cost (Seo et al., 2002), vertical high speed machining centers (Ciurana et al., 2008), and building construction (Emsley et al., 2002). Case-based reasoning, another common AI technique, uses the nearest-neighbor algorithm, which is based on the rationale of characterizing projects by a number of essential attributes.
Modeling greenhouse air humidity by means of artificial neural network and principal component analysis
2010, Computers and Electronics in AgricultureComputation of magnetic-field-induced flow for targeting
2020, Proceedings of the 3rd IEEE International Conference on Knowledge Innovation and Invention 2020, ICKII 2020
- 1
Tel.: +886 6 2815208.