Data processing: artificial intelligence – Adaptive system
Reexamination Certificate
2002-03-07
2004-01-13
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Adaptive system
C706S015000, C706S019000, C717S156000
Reexamination Certificate
active
06678668
ABSTRACT:
FIELD AND BACKGROUND OF THE INVENTION
The present invention relates to the fields of process design, development and control. More particularly, the present invention relates to providing an enterprise wide framework for improving final product performance. The term “enterprise wide” as used herein relates to the entirety of the stages with which an enterprise is concerned with the product, typically the product design stage, the process design stage and the manufacturing stage. The present invention further relates to various techniques for performing improving product performance by mapping, modeling, optimization and control throughout the entire process stages of a product ensuring improved final outcome performance.
One example of such a process is yield in the semiconductor field. Yield improvement in semiconductor manufacturing is dominated by particle and defect inspection technology and process integration improvement in order to improve device yield. Improved process integration yields higher bin yields (faster chips) and other outputs such as electrical test (ET) and reliability. Yield loss due to inefficient process integration is becoming more critical as integrated circuits become smaller.
The control of the product parameters from the earliest stages of design and up to the final stages of manufacturing is therefore the key to yield improvement.
Engineers design and develop the ‘process spec’. The process specs are the specs of the intermediate process steps whose integration should lead to the final product, which has its own ‘product spec’.
As an unlimiting example of a process, the semiconductor industry will now be considered. In the production of semiconductor chips there can be up to as many as a few thousand ‘process specs’ that give rise to a product (chip). The product in turn may have its own ‘product specs’ with hundreds of parameter goals (electrical tests). Such is the case for example when the product is a CPU.
In complex processes such as the abovementioned semiconductor chip production, the process integration, i.e. the determining of the right sets of ‘process specs’ in order to achieve good ‘product specs’ is very intricate. A current inability to tackle correctly the process integration issue is one of the major factors in manufacturing yield losses, that today often exceed 50% for products of such a level of complexity.
Various advanced process control and supervisory control solutions have been successfully implemented in various industries. However in intricate processes such as in the semiconductor industry these solutions have been found to be insufficient due to the complexity involved in the optimization process of thousands of parameters.
In the above-mentioned Knowledge Tree application, it is explained how to logically break down intricate processes into subprocesses by mapping, each subprocess comprising Knowledge Tree Cells. The cells are then filled with models that describe the relationships between the input parameters and the outcome/s of each cell.
FIG. 1
is a diagram of KT cells involved in a process. As an example of a final target of a ‘product spec’, threshold Voltage (Vt), is used here. Vt, as indeed all of the other targets (electrical test parameters) of the final product, has its own contributing variables that have an influence on this particular target. In turn, and looking one tier back, each of these contributing variables, Gox-T (Gate Oxide Thickness) for example, is considered as a target goal for the layer beneath it and has its own contributing variables that influence the outcome and so forth until the earliest tier is reached. Following mapping is the stage of modeling, wherein models are placed in the KT cells in order to quantify the relationships between the inputs and the outputs of the KT cells.
Advanced or automatic process control has been used in the prior art for optimization of a tool or a module where all of the process or subprocess Knowledge Tree cells have had inputs and outputs defined and modeled. APC provides optimization and control solutions for the tool level. In the above-mentioned patent application on MC, APC has been taken one step higher and has been applied to Module Control on a higher level involving a number of tools.
Due to a lack of control and optimization on the whole process level, the final targets are frequently not entirely within the spec limits around the target and even when they are, they often have high variation, i.e. wide distribution that are monitored by low Cp and Cpk (process capability index) values for many of the device characteristics, such as Bin (device speed), Electrical Test and Reliability. Another major concern is the length of time taken today from the design stage through product development and then ramp up and finally to production. This can be especially lengthy and a number of years is not untypical for the semiconductor industry. This is due in part to the intricacy of the processes involved and the lack of optimization technologies, leaving nothing but trial and error in many cases for integrating the various stages.
As a further result of the intricacies of the process, the entire semiconductor process is today run on an open loop system. Having an open loop system does not allow for feed back for self-correction. There is today no comprehensive solution that takes into account the whole process from the design stage until the manufacturing that guarantees product specs.
In the patent application WO 0177872 a system and method for enterprise modeling, optimization and control to Ferguson et al, is disclosed. The application discusses such a system and method for performing modeling, prediction, optimization, and control, including an enterprise wide framework for constructing modeling, optimization, and various control solutions are disclosed. The disclosure discusses integrating a combination of batch and continuous processing frameworks and a unified hybrid modeling framework which allows encapsulation and composition of different model types, such as first principles models and empirical models. However, Ferguson et al does not describe how to solve and optimize a complex processes with thousands of process spec variables and hundreds of product specs. Furthermore Ferguson et al, in attempting to integrate numerous models into a single framework, provides a rigid structure that fails to be an organic description of the underlying processes.
There is thus a widely recognized need for, and it would be highly advantageous to have, a method and a system for the optimization of an enterprise level process that can be implemented from the very first design stage through to the final product stage and which is devoid of the above limitations.
SUMMARY OF THE INVENTION
According to the present invention there is provided apparatus for optimization of a complex process, said process being described by a plurality of input variables, a plurality of intermediate variables and a plurality of output variables having relationships therebetween such that ones of said inputs and said intermediate variables effect respectively different output variables, each of said output variables having a target, said apparatus comprising an optimizer for finding an optimum value for respective ones of said input and intermediate variables to maximize a summed convergence of said output variables to said targets.
The apparatus preferably further comprises a weight assigner associated with said optimizer, to assign each of said output variables with a weight, and wherein said summed convergence comprises a summed weighted convergence.
Preferably, at least some of said weights are user determined.
Preferably, each of said output variables comprises an upper level of a multi-level decision tree comprising said corresponding input variables and said corresponding intermediate variables, and wherein decision trees of at least two of said output variables are related by sharing ones of said input and intermediate variables such that optimization of one tree affects optimiza
Entin Efim
Fisher Yossi
Hartman Jehuda
Kokotov Yuri
Seror Jacques
G.E. Ehrlich (1995) Ltd.
Insyst Ltd.
LandOfFree
System and method for complex process optimization and control does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for complex process optimization and control, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for complex process optimization and control will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3222664