應用自組性演算法構建積體電路之良率模式

Abstract

摘要 晶圓上的良率為積體電路製造業者衡量其製程能力的一個重要指標，影響良率的因素有很多，其中晶圓上缺陷點的總數以及缺陷點的群聚程度是決定晶圓良率高低的重要因素。隨著晶圓製程技術的進步，晶圓面積不斷增大，造成晶圓上缺陷點產生群聚現象，因而導致卜瓦松(Poisson)良率模式預測不準確。針對這問題，中外文獻提出一些複合卜瓦松良率模式(compound Poisson yield model)，或應用倒傳遞類神經網路(Back-propagation Network, BPN)來建構良率模式，來預測晶圓之良率，但這些良率模式均有一些缺失。因此本研究之主要目的是以群聚指標和缺陷點總數為依據，應用自組性演算法(Group Method of Data Handling, GMDH)來構建一個新的晶圓良率模式，此模式不需要任何統計假設，且可以構建出一個預測良率的數學方程式，因此實用價值甚高，在良率預測上也較業界常用的負二項良率模式和倒傳遞類神經網路良率模式更為精確。最後，本研究利用模擬與積體電路公司之實際晶圓資料來驗證本研究所發展之良率模式確實有效可行。

Abstract

For integrated circuit (IC) manufacturers, the wafer yield is a key index to evaluate their profit. There are two major factors affecting the wafer yield. One is the number of defects on a wafer and the other is the degree of defects clustering. As the wafer size increase, the defects clustering phenomenon tends to be significant. In this case, the Poisson yield model will frequently underestimate the actual wafer yield. Although many compound Poisson yield models or yield models which using Back-Propagation Neural Network (BPN) have been developed to overcome the clustering problem, these models still have some shortcomings. Therefore, the objective of this study is to develop an IC yield model using Group Method of Data Handling (GMDH). The cluster index and defect counts are the input variables for the proposed model. The model does not need any statistical assumption and can obtain a mathematical equation. A simulated data and a real-world data from an Taiwan’s IC Company are utilized to demonstrate the effectiveness of the proposed IC yield model. Comparisons are also made among the Negative Binomial yield model, Back-Propagation Neural Network yield model and the proposed model to prove that the proposed model is superior.