EMRS | alumni

Yield modeling deals with the prediction of yields for a specific process based on the estimated influences of key factors that directly or indirectly affect the process. An accurate prediction of yield would enable the efficient management of manufacturing costs, enhance the capacity of the assembly line, and assist with the procurement of material and with the on-time delivery of the product. Consequently, yield modeling is an important aspect of both New Product Introduction (NPI) and 'regular' manufacturing. The objective of this research effort was the design, development, and implementation of an accurate estimation mechanism for first-pass assembly yields in an Electronics Manufacturing Service (EMS) provider's environment. Probability based approaches - the Poisson model and the regression model - were applied to achieve this objective. However, their application in a specific industrial environment needs careful consideration. The collection and use of data should be automatic since the huge amount of production data that needs to be considered to ensure the model's accuracy renders manual work burdensome. Another important factor is that instead of getting the data at one time and constructing model that is based on 'static' data, the ideal model should retrieve the data in real time and be constructed dynamically. This type of model would represent the reality on the production floor more accurately. Therefore, the software applications designed and developed in this research, for both the models, access and utilize data in real-time from multiple databases. For some data sources, for which databases are not available at the current time, databases were constructed to simulate them. The case studies carried out to exercise the Poisson model indicates that although the actual assembly yield varies from time to time or from work order to work order because of the process variations, they tend to fluctuate around the yield estimated by the Poisson model. In another words, the yield estimated by the Poisson model can be treated as a mean value of the actual yields. For the regression model, an approach was devised so that the average defects per assembly was correlated with four factors - number of I/Os per assembly, PCB area, PCB thickness and pad coating. In addition, component related defects were further categorized based on lead type, pitch, number of leads, and pad area. Case studies indicated the data available and being currently used for the regression based model needs to be improved in terms of its quality. Also, the number of regressors being used may need to be enhanced. Regressors that are more representative may need to be selected and used. This would also increase the accuracy of the regression based.


	sitemap \| contact us