A Study On The Key Techniques Of Quality Control And Evaluation In Modern Semiconductor Manufacturing Processes

In the field of modern semiconductor and integrated circuit manufacturing, asprocess, technology and equipments are becoming more and more complicated, thefunction of electronic devices is becoming more perfect and users bring morerequirements to process level and product quality.Then, it isnecessary thatthe techniquesof quality control and process evaluation should be paied more attention. At the sametime, in order to keep abreast of time and compete with international companies, domesticcompanies begin to implement techniques of quality management and process evaluation.So, their implement is of important significance.This dissertation studies some problems of process control and evaluation inmultivariate process and multi-variety and small batch manufacturing system using self-starting technique, multivariate numerical integration and maximum likelihoodestimation and so on through theory and computer simulation analysis. The problemsinclude monitoring process mean and standard deviation of multi-variety and small batchproduction run, and the evaluation of the process and product quality of multivariateprocess. In the meanwhile, the method of evaluation on the quality of product providedby suppliers based on truncated sample is also introduced. These researches can improveprocess products quality and reduce quality defects. It is meaningful to up-grate marketcompetitiveness of enterprise. The main contents are summarized as follows:1). Fistly, the existing problem of implementing quality control and evaluation inmulti-variety and small batch production runs to monitor process mean and standarddeviation is analyzed. A quality control technique in multi-variety and small batchmanufacturing system, T-K control chart, is proposed constructively. Using this method,the quality manager needn’t collect samples as many as traditional chart. The T or Kstatistics are calculated based on each subgroup, and the statistics of each subgroup areindependent with each other and have an identical distribution. This method adopts self-startingtechnique, and does notrequire PhaseI sample aimingat estimating processmean.Especially, the T-chart does not need the estimation of variance. Even the distributionparameters are unknown, the T-K chart can be used to monitor multi-variety and smallbatch production run as long as there are no less than2subgroups.2). In the light of the theory of statistical process control, some methods to evaluatethe performance of control chart are discussed. Then, the concept, the significance andthe calculation steps of average run length are introduced. The ARL performance of Tchart and K chart is analyzed. In addition, the shortcoming of self-starting control chart is studied, and a strategy for optimizing theARLperformance of self-staring control chartis proposed. The result reveals that the ARL performance is improved dramatically afterthe optimization.3). The relationship between the index Cpkand process yield for the case with singlecharacteristic is analyzed. Then an equivalent process capability index ECpkis proposedbased on the idea of six sigma design strategy. ECpkindex has a one-to-onecorrespondence relationship with process yield. Following this idea we develop amultivariate PCI MECpkwhich can be used to evaluate process capability for processeswith multiple characteristics and we also discuss the impact of covariance matrix onprocess yield, fromwhich the authors propose a solution to improve overall process yield.Some illustrative examples are presented to verify the validity of the suggested index invarious cases.4). It is known that product yield reflects the potential product quality and reliability,which means that high yield corresponds to good quality and high reliability. From theview of consumers, the method of judging the quality of products supplied bymanufacturers based on truncated samples is introduced. This dissertation proposes analgorithm for calculating the parameters of full Gaussian distribution before truncationbased on truncated dataand estimating product yield.The algorithmis proper to both one-side and two-side truncation. The confidence interval of the yield result is derived, andthe effect of sample size on the precision of the calculation result is also analyzed. Finally,the steps for calculating the product yield are offered and the effectiveness of thisalgorithm is verified by an actual instance.5). In addition, based on the models and algorithms discussed above, the computer-aid quality control and process evaluation software is developed. The software systemincludes the function for statistical process control and process capability evaluation. Thesoftware offers many control charts,process capability indicesand especially the functionfor computing multivariate process capability index proposed in this dissertation.