Data processing: structural design – modeling – simulation – and em – Modeling by mathematical expression
Reexamination Certificate
1999-05-27
2001-06-12
Teska, Kevin J. (Department: 2123)
Data processing: structural design, modeling, simulation, and em
Modeling by mathematical expression
C700S030000, C700S031000
Reexamination Certificate
active
06246972
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a system for optimizing manufacturing processes, and particularly, to a system for modeling and optimizing the maintenance of semiconductor processing equipment.
2. Description of the Related Art
The electronics industry's growth during the twentieth century has been driven in part by a rapid succession of revolutionary improvements in the performance and reliability of the microelectronics, accompanied by drastic reductions in cost, power consumption and size of the resulting semiconductor devices. These advances are supported by corresponding improvements in semiconductor processing technology.
The production of semiconductor devices involves three essential steps: an integrated circuit (IC) design step, a mask making step and a fabrication step. The fabrication step maps the semiconductor structures generated by the design and mask making steps onto the silicon surface in specified layers. After the features of the IC are defined and implemented as components in the first step, as specified in design files, the second step includes the photolithography step which converts the design files into a set of masking plates containing exact images of the semiconductor structures in either opaque or transparent shades. After the creation of the masks, a “fab house” or a sophisticated chemical factory makes “prints” of the masks onto silicon wafers to create semiconductor structures on the IC.
One common type of equipment in the fab houses includes plasma etch reactors for removing of layers such as dielectric films for creating capacitors and other semiconductor structures. As is common in many complex manufacturing processes, long term drift occurs in the semiconductor manufacturing equipment, including the plasma etch reactors. This drift causes plant equipment to operate outside of the specified operating ranges, leading to undesirable manufacturing defects and, if not timely corrected, ultimately to equipment failures that could affect the integrity of the plant itself.
To compensate for long term drift, manufacturing personnel traditionally vary one or more continuous process variables. For plasma etch reactors, these variables included radio frequency (RF) power variables, pressure level variables, and gas flow variables, among is others. The plant personnel also correct drift using maintenance events such as replacing or cleaning various components of the reactors. Due to their complexity, the plasma etch reactors require a significant amount of maintenance which is expensive in terms of the maintenance labor cost, the cost of anticipatory component replacement, as well as the plant throughput reduction caused by maintenance events. As the alternative of a plant shut-down is unacceptable, the industry expends significant efforts in optimizing the equipment maintenance process.
The increasing complexity of industrial processes drove process control systems toward making experience-based judgments akin to human thinking in order to cope with unknown or unanticipated events affecting the maintenance of the plant equipment. The application of expert system technology represents a step in the adaptive control of this complex IC fabrication equipment. Based on the knowledge derived from one or more experts, the expert system software typically adjusts the process control strategy after receiving inputs on changes in the system environment and control tasks. However, as the expert system depends heavily on a complete transfer of the human expert's knowledge and experience into an electronic database, it is difficult to produce an expert system capable of handling all the dynamics of a complex system.
Recently, neural network based systems were developed with powerful self-learning and adaptation capabilities to cope with uncertainties and changes in the system environment. Modeled after biological neural networks, engineered neural networks process training data and formulate a matrix of coefficients representative of the firing thresholds of biological neural networks. The matrix of coefficients are derived by repetitively circulating data through the neural network in training sessions and adjusting the weights in the coefficient matrix until the outputs of the neural networks fall within predetermined ranges of the expected outputs of the training data. Thus, after training, a generic neural network conforms to the particular task assigned to the neural network. Thus, the neural network shares common traits with a large class of flexible functional form models known as non-parametric models, which include neural networks, Fourier series, smoothing splines, and kernel estimators.
Although a neural network-based maintenance modeling system has powerful self-leaning and adaptation capabilities to cope with uncertainties and changes in its environment, the lack of a process-based internal structure represents a liability for the neural network. For instance, when training data are limited and noisy, the network outputs might not conform to known process constraints. For example, certain process variables increase monotonically as they approached their respective asymptotic limits. Both the monotonicity and the asymptotic limits constitute factors that should be enforced on the neural network when modeling these variables. However, the lack of training data may have prevented a neural network from capturing either.
In such events, insufficient data hampers the accuracy of a neural network due to the network's pure reliance on training data when inducing process behavior. A number of approaches have been utilized to exploit prior known information and to reduce the dependence on the training data alone, including the use of qualitative knowledge of a function to be modeled to overcome the sparsity of training data. One approach deploys a semi-parametric design which applies a parametric model in tandem with the neural network. As described by S. J. Qin and T. J. McAvoy in “Nonlinear PLS Modeling Using Neural Networks”,
Computers Chem. Engng
., Vol. 16, No. 4, pp. 379-391 (1992), a parametric model with a fixed structure is derived from a first principle which can be existing empirical correlations or known mathematical transformations. The neural network is used in a series approach to estimate intermediate variables to be used in the parametric model.
Alternatively, a parallel semi-parametric approach has been deployed where the outputs of the neural network and the parametric model are combined to determine the total model output. The model serves as an idealized estimator of the process or a best guess at the process model. The neural network is trained on the residual between the data and the parametric model to compensate for uncertainties that arise from the inherent process complexity.
Although the parallel semi-parametric model provides a more accurate model than either the parametric model or the neural network model alone, it requires prior knowledge, as embodied in the first principle in the form of a set of equations based on known physics or correlations of input data to outputs. The parametric model is not practical in a number of instances where the knowledge embodied in the first principle is not known or not available. In these instances, a readily adaptable framework is required to assist process engineers in creating a process model without advance knowledge of factors such as the first principle. First principle models of long-term drift from maintenance effects in IC fabrication equipment are difficult to construct. Moreover, the theory upon which the models are based is incomplete at best. Thus, they do not effectively model the events that trigger signals indicating that maintenance was needed on the IC fabrication equipment.
A second approach to modeling maintenance is to uniformly sample events back in ime and use these as inputs to a regression model. Such periodic sampling of events back in time is appropriate only for systems which sample data at regular
Aspen Technology Inc.
Bracewell & Patterson L.L.P.
Frejd Russell W.
Teska Kevin J.
LandOfFree
Analyzer for modeling and optimizing maintenance operations does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Analyzer for modeling and optimizing maintenance operations, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Analyzer for modeling and optimizing maintenance operations will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2438975