Photocopying – Projection printing and copying cameras – Methods
Reexamination Certificate
2002-02-12
2004-03-02
Mathews, Alan A. (Department: 2851)
Photocopying
Projection printing and copying cameras
Methods
C355S053000, C700S121000, C706S023000, C382S156000, C356S400000
Reexamination Certificate
active
06700648
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method for controlling a processing apparatus, and more particularly to a method for controlling a processing apparatus in conformity with a new Adaptive Least Mean Square Neural Network (ALMS-NN) algorithm for evaluating a correction value of a stepper during a photolithographic process in an overall process of manufacturing semiconductor devices.
2. Description of the Related Art
As one way for strengthening competitiveness in the semiconductor industry, a great number of studies have been made for constructing an effective manufacturing system capable of ensuring a high production yield. Especially in case of a photolithographic process, which is one of the important semiconductor manufacturing processes, as the processing conditions are frequently varied, necessitating development of a systematic production system to deal with the frequent variation, considerable efforts for establishing a system that reduces the frequency of sampling processes are being directed to enhance the production yield.
It is a misalignment problem in the processing that is to be primarily considered during the photolithographic process for establishing such a system. The problem chiefly occurs due to a difficulty in analyzing the physical and chemical characteristics of the process, noise interference during performing the process, and measurement errors after performing the process, which become reasons for increasing the frequency of the sampling that directly affects the production yield.
A Process Control System (PCS), widely developed and employed as a production system capable of ensuring a high production yield for the semiconductor manufacturing system, mostly involves no mathematical model with respect to the process steps, but controls the process in view of numerical values statistically obtained by processing data of the previously performed processing.
The scheme for reflecting the previous experience values of prior processing in a current process as a kind of ‘experience inheritance,’ is an algorithm that feeds back a mean of weighted values with respect to recent historical data of identical processes. However, such an algorithm is static, without considering the temporally changing characteristics of a system, having a drawback that the sampling process has to be repeatedly performed because of insufficient identical history data within a fixed period, or successive spec-out generation.
For this reason, processing controller design techniques using a neural network model have been currently suggested as a scheme for properly dealing with a process of a non-linear system without having a specific mathematical model (“Monitoring and Control of Semiconductor Manufacturing Processes,” IEEE Control System, 1998 by S. Limanond, J. Si, and K. Tsakalis; and “Artificial Neural Network Model-Based Run-to-Run Process Controller,” IEEE Trans. on Component, Packaging, and Manufacturing Technology-Part C, vol. 10, no. 1, January 1996 by X. A. Wang and R. L. Mahajan).
First, there has been suggested a method for allowing the neural network to learn by means of data from previously conducted processes, thereby predicting the manufacturing process via a pattern search with respect to a previous process. The basic premise of this method is in that the variation pattern of the non-linear system is not completely random. Accordingly, it is possible to predict that the past history data involving a pattern similar to the recent variation pattern may exist, unless the variation pattern of the system with respect to a certain stepper is completely random, so that the variation pattern of the past system is utilized to presume a current output value.
However, this method has disadvantages of being liable to induce a difficulty in using the data because it requires a lot of past history data to be effectively applied, and increasing the amount of computing required because of the continuously repeated pattern search and neural network learning.
Second, a widely utilized Exponential Weighted Moving Average (EWMA) system provides modeling and approximating methods of a system having a data variation capable of being described in a time series system, which are widely available in the processing control field of semiconductor manufacturing. This has been disclosed in literature such as “Run by Run Process Control: Combining SPC and Feedback Control,” IEEE Trans. on Semiconductor Manufacturing, vol. 8, no. 1, February 1995 by E. Sachs, A. Hu, and A. Ingolfsson; “Adaptive Optimization of Run-to-Run Controllers: The EWMA Example,” IEEE Trans. on Semiconductor Manufacturing, vol. 13, no. 1, February 2000 by N. S. Patel and S. T. Jenkins; and “A Self-Tuning EWMA Controller Utilizing Artificial Neural Network Function Approximation Techniques,” IEEE Trans. on Components, Packaging, and Manufacturing Technology-Part C, vol. 20 no. 2, April, 1997 by T. H. Smith, D. S. Boning.
The EWMA system is frequently applied to the actual operation of semiconductor devices, as its model is simple and it also is possible to apply a simple recursive formula as the equation written by:
{overscore (
x
(
x
))}=&lgr;
x
(
i−
6)+(1−&lgr;){overscore (
x
)}(
i−
1)
However, if the &lgr; value used is small, non-negligible weight is applied upon past data when applying the EWMA system, causing a drawback of requiring a lot of past data for performing accurate estimation.
Third, the system predication technique by means of the Kalman Filtering is a classical predication technique with respect to a system of which motion characteristics are fundamentally modeled by a differential equation or difference equation in a state-space form and are interrupted by a white noise.
Based on the fact that the correction value of the system is generally changed by the noise, even though it involves no change without being interrupted by the noise, the variation characteristic of the correction value is assumed to be a linear model as follows:
x
(
k+
1)=
x
(
k
)+
w
(
k
)
Here, a reference alphabet w(k) denotes a white noise term that becomes a cause of varying the correction value.
At this time, the performance is determined in accordance with the supposed system model how the model by reproduces the motion characteristics and noise characteristics of an original system. However, it is not easy to provide a supposed model similar to an original system in such a system that has the highly linear characteristics like the semiconductor process.
SUMMARY OF THE INVENTION
In order to solve the above-enumerated problems of the conventional technique, an object of the present invention is to provide a method for controlling a processing apparatus having a new ALMS-NN algorithm, capable of being effectively applied to a process that highly depends on a sampling due to a frequent replacement of work pieces subjected to the processing without depending on a number of past history data.
Another object of the present invention is to provide a method for controlling a photolithography apparatus having a new ALMS-NN algorithm for deciding an apparatus input value capable of effectively correcting an overlay alignment error by using a stepper apparatus during a photolithographic process as a target.
Accordingly, there is provided a method for controlling a processing apparatus, in which an error value between an input value of the processing apparatus for processing a work piece to be processed, and a measurement value obtained by measuring the work piece processed in the processing apparatus is obtained, a correction value for correcting the input value of the processing apparatus is computed in order to decrease the error value, and the values are managed as processing data to be utilized in computing a next correction value. The previous processing data having a history identical to the work piece loaded to the processing apparatus are searched, and a current bias correction value is predicted from the latest plurality of previous
Cho Bong Su
Hwang Jae Won
Jei Young Ho
Jun Kyoung Shik
Kang Hyun Tai
Mathews Alan A.
Samsung Electronics Co,. Ltd.
LandOfFree
Method for controlling a processing apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for controlling a processing apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for controlling a processing apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3245192