基於類神經網路之虛擬量測系統建構

Development of a Neural Network based Virtual Metrology system

陳志源(CHIH-YUAN CHEN)

指導教授 ： 劉益宏

中原大學/工學院/機械工程學系/碩士(2008年)

https://doi.org/10.6840/cycu200900544

摘要

台灣目前半導體(Semiconductor manufacturing)產業發展至今，在全球的半導體產業已
佔有舉足輕重的角色，因此對於產品品質的要求，會更加嚴謹。而目前半導體廠對於半
導體化學氣相沉積(Chemical Vapor Deposition, CVD)機台現階段的產品品質監控作法是
在生產機台中放置三片的監控晶圓(Monitor wafers)，並藉由量測機台去量测監控晶圓，
其量測值結果用來代表該批產品晶圓的品質，此種監控方式並無法及時性的得知每片產
品晶圓的品質，以致於當生產機台性能產生漂移(Drift)或當機時，會導致產品晶圓品質
的不良，造成極高的成本損失。
根據上述問題，本論文提出一套虛擬量測系統(Virtual Metrology System, VMS)來克服
這個問題，其不僅可達到即時量測每片晶圓之品質，亦可減少監控晶圓之使用率，並即
時監控生產機台效能將其回報給予機台工程師。本論文所提出的虛擬量測系統能夠從生
產設備所使用的製程參數資料來預測每片晶圓生產的品質，因此，機台發生漂移的問題
也能迅速的被察覺出來。並採用一個輻射基底函數類神經網路(Radial Basis Function
Neural Network, RBFN)建構虛擬量測模型，RBFN 的優點在於網路訓練速度快、調整參
數少，並搭配正交最小平方法(Orthogonal Least Squares, OLS)能自動選擇RBFN 的隱藏
層神經元的數目；此外，為了使VM 系統對新進製程參數資料有較好的預測精度，本論
文使用VM 線上學習機制，其目的是將相似度較低的新進製程參數資料納入RBFN 訓練
資料集合裡，重新訓練一個新的RBFN 模型，使得保有模型的新鮮度。因此，所提出的
虛擬量測系統既簡單又有效率，並且能實際應用建構出虛擬量測模型於半導體化學氣相
沉積(Chemical Vapor Deposition, CVD)製程上。此外，本論文發展出一套虛擬量測人機
介面系統，讓使用者方便與VM 系統做溝通。

關鍵字

輻射基底函數類神經網路 ； 化學氣相沉積 ； 虛擬量測

並列摘要

In advanced semiconductor manufacturing, the in-process wafers need to be monitored
periodically in order to obtain high stability and high yield rate, by which the products that
satisfy the specifications can be obtained. Till now, the general method is that a monitoring
wafer is inserted to three pieces of the process at a specific time for measuring the quality.
However, it not only increases the cycle time and sudden drift of the equipment can not be
observed in real time by this way. This may result in defective wafers and they will therefore
be discarded, leading to a high production-cost loss.
According to above problem, this dissertation proposes a virtual metrology (VM) system
to overcome this problem. The proposed VM system can not only achieve the goal of
real-time measuring the quality of every production wafer, but also reduce the number of
monitoring wafers. In addition, the system is able to feedback the performance of the
production equipments to the engineers, and therefore the drifting phenomena of the
equipments can be observed. In this VM system, a radial basis function neural network
(RBFN) is used to construct the virtual metrology model (VMM).The proposed VM system is
not only efficient but also simple to implement to real chemical vapor deposition (CVD)
process in real semiconductor manufacturing. In addition, this thesis also develops a VM
human-machine interface, by which users can communicate with the VM system easily.

並列關鍵字

Radial Basis Function Neural Network(RBFN) ； Chemical Vapor Deposition(CVD) ； Virtual Metrology(VM)

參考文獻

Independent Component Analysis and Support Vector Data Description,” Proceedings

[49] 張智星，MATLAB程式設計與應用，清蔚科技股份有限公司，2002年.

[1] A.C. Diebold, “Overview of metrology requirements based on the 1994 national

technology roadmap for semiconductor,” in Proc. of IEEE/SEMI Advanced

[2] K. J. Stanley, T.D. Stanley; J. Maia, “Realizing 300 mm fab productivity improvements