WO2009067952A1

WO2009067952A1 - An art control method and a device thereof

Info

Publication number: WO2009067952A1
Application number: PCT/CN2008/073169
Authority: WO
Inventors: Shangui Zhang
Original assignee: Beijing Nmc Co., Ltd.
Priority date: 2007-11-26
Filing date: 2008-11-24
Publication date: 2009-06-04
Also published as: CN101446804B; CN101446804A

Abstract

The invention provides an art control method. The method includes: STEP a, receiving the target output variables; STEP b, calculating the target processing variables using the preset art predictive model and the feedback control model; said art predictive model is used for describing the function relationship between the processing variables and the output variables. The target function of said feedback control model is calculating the target processing variables having the minimum distance from the preset references. The restriction term of said feedback control model is that the target output variables and the target processing variables are all up to said art predictive model. STEP c, inputting said target processing variables to real art devices and getting the real output variables; STEP d, deciding the differences between said target output variables and said real output variables; if the differences do not fulfil the preset demand, adjusting said preset art predictive model and returning to STEP b; if the differences fulfil the preset demand, then end.

Description

Process control method and device

The present invention relates to the field of industrial process control technology, and more particularly to a method and apparatus for process control. Background technique

Process feedback control, that is, control using feedback information. The actual performance of the production system is measured by the sensing system or the measurement system and then compared with the required standards. If there is a difference, it is interpreted by the control system, and the control system commands the operating device to correct the performance. Eliminate the difference. This is the simplest feedback mode.

For example, in semiconductor etch processes, process feedback control techniques are a critical technology. In recent years, with the continuous updating of process etching technology, the control requirements for etching equipment and wafer quality have become higher and higher in the etching process. For the specific etching process, how to effectively realize the feedback control of equipment and silicon performance has become a hot issue in the IC industry in the world today.

The current principle methods for process feedback control mainly include: Exponential weighted moving average method

(EWMA, including single exponential weighted moving average method (SEWMA) and double exponential weighted moving average method (DEWMA)), nonlinear programming control method (NLP), and model predictive control method (MPC).

The basic idea of this type of method is: According to the existing historical data under the normal process, according to certain criteria, the response surface analysis method (RSM, mainly linear regression, least squares or neural network method) is used to fit out. The functional relationship between the process output variable and the process input variable, that is, the prediction model; for a sample to be detected, first use the prediction model to calculate the process output variable, and then based on the difference between the process prediction output value and the actual output value, Adjust the prediction model and control parameters of the process until the difference between the predicted value and the actual value of the output variable complies with the current process regulations. Appropriate prediction models and control parameters.

Compared with other process control methods (open loop control, feedforward control), the exponential weighted moving average method, the nonlinear programming control method and the model predictive control method are all closed-loop control methods. In the feedback control process of the process, they are Generally have a good control effect.

However, as the three most commonly used feedback control methods at this stage, the exponential weighted moving average method, the nonlinear programming control method and the model predictive control method still have some shortcomings in the practical application process. as follows:

The exponential weighted moving average method is only suitable for single-input and single-output control systems. When performing multi-input single-output and multi-input and multi-output system control, the calculation process is very complicated, and a small link in the control model is not well processed. Leading to high errors;

For the nonlinear programming control method, the difficulty lies in the complexity and uncertainty of the calculation process (not all nonlinear programming problems can be solved);

The model predictive control method, because the control process involved is too complicated, its feasibility in practical process production is not so large.

In order to solve these problems, many professionals have begun to introduce various new technologies to improve existing control methods. In summary, one technical problem that needs to be solved urgently by those skilled in the art is: How can a more effective process feedback control method be proposed. Summary of the invention

The technical problem to be solved by the present invention is to provide a process feedback control method and device, which can realize process feedback control very effectively, and to some extent, improve the precision of process feedback control and avoid computational complexity, and reduce the calculation of feedback control. the amount.

In order to solve the above problem, the present invention discloses a process control method, which specifically includes: Step a: receiving a target output variable;

Step b: calculating a target process variable by using a preset process prediction model and a feedback control model; the process prediction model is used to describe a functional relationship between the process variable and the output variable; The objective function of the control model is: solving the target process variable with the smallest distance from the preset reference value; the constraint condition of the feedback control model is: the target output variable and the target process variable are in accordance with the process prediction model;

Step C: input the target process variable to the actual process device to obtain an actual output variable; Step d, determine a deviation between the target output variable and the actual output variable; if the preset requirement is not met, adjust the The preset process prediction model returns to step b; if the preset requirement is met, it ends.

Preferably, the distance between the target process variable and the preset reference value comprises: a relative distance, an absolute distance or a weighted distance.

Preferably, the manner of adjusting the preset process prediction model includes adjusting a gain coefficient and

/ or intercept.

Preferably, the process prediction model is obtained by data sample analysis.

Preferably, when there are multiple sets of target process variables that meet the preset requirements, an optimal set of target process variables is selected according to the deviation values and/or the adjustment range of the single target process variables.

According to another embodiment of the present invention, a process control apparatus is further provided, including: a process target receiving module, configured to receive a target output variable;

a feedback processing module, configured to calculate a target process variable by using a preset process prediction model and a feedback control model; the process prediction model is used to describe a functional relationship between the process variable and the output variable; and the objective function of the feedback control model is : solving a target process variable with a minimum distance from a preset reference value; the constraint condition of the feedback control model is: the target output variable and the target process variable are in accordance with a process prediction model;

a device interface module, configured to input the target process variable to an actual process device, and detect an actual output variable;

a judgment feedback module, configured to determine a deviation between the target output variable and the actual output variable; if the preset requirement is not met, adjusting the preset process prediction model, and returning to the feedback processing module; If requested, it ends. Preferably, the distance between the target process variable and the preset reference value comprises: a relative distance, an absolute distance or a weighted distance.

Preferably, the manner of adjusting the preset process prediction model may include adjusting a gain factor and/or an intercept.

Preferably, the process prediction model is obtained by data sample analysis.

Preferably, the device may further include: a screening module, configured to: when there are multiple sets of target process variables that meet preset requirements, select an optimal group according to the deviation value and/or the adjustment range of the single target process variable Target process variable. Compared with the prior art, the present invention has the following advantages:

The overall idea of the present invention still uses the principle of the nonlinear programming control method, but by introducing the technical idea of the nearest neighbor rule to realize the feedback control of the process, a better control error can be obtained, and the process drift can be better avoided; The calculation mode of searching for the most advantageous in the multi-dimensional space is changed to search for the most advantageous calculation mode on the multi-dimensional plane, thereby reducing the amount of calculation.

The present invention can be adapted to all control systems, such as single-input single-output (SISO) systems, multiple-input single-output (MISO) systems, multiple-input multiple-output (MIMO) systems, etc. Second, the present invention reduces computational feedback control calculations. The quantity and the accuracy of the process feedback control also have a significant effect. DRAWINGS

1 is a flow chart showing the steps of an embodiment of a process control method of the present invention;

2 is a comparison diagram of errors of the conventional feedback control model and the nearest neighbor rule feedback control model of the present invention;

3 is a block diagram showing the structure of an embodiment of a process control device of the present invention. detailed description The present invention will be further described in detail with reference to the drawings and specific embodiments.

The method of the present invention can be described in the general context of computer-executable instructions executed by a computer, such as a program module. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform particular tasks or implement particular abstract data types. The invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including storage devices.

The feedback control process described herein can be accomplished by a number of general purpose or special purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, distributed computing environments including any of the above systems or devices, and the like. Referring to FIG. 1, an embodiment of a process control method of the present invention is shown, which may specifically include the following steps:

Step 101: Receive a target output variable.

Step 102: Calculate a target process variable by using a preset process prediction model and a feedback control model; the process prediction model is used to describe a functional relationship between the process variable and the output variable; and the objective function of the feedback control model is: a target process variable having a minimum preset reference distance; the constraint condition of the feedback control model is: the target output variable and the target process variable conform to a process prediction model;

The calculation of the distance between the target process variable and the preset reference value in the feedback control model, wherein the distance may include a relative distance, and may also include an absolute distance or a weighted distance.

Step 103: Input the target process variable to the actual process device to obtain an actual output variable. Step 104: Determine a deviation between the target output variable and the actual output variable; If the preset requirement is not met, the preset process prediction model is adjusted, and the process returns to step 102; if the preset requirement is met, the process ends.

The process target parameters received in step 101 are determined according to the main control logic for the entire process. For different processes or process equipment, or different process moments, the process target parameters may be changed, and the present invention is based on the established The process target, to feedback adjustment to get the optimal process input parameters (Recipe, process variables, input variables). The present invention is not necessarily limited to specific process target parameters.

Step 102 is one of the core steps of the entire feedback control. One or more sets of target process variables that meet the requirements are obtained mainly by mathematical analytic operations (since the general process control may have multiple input variables at the same time).

The process prediction model is based on pre-applied multiple sets of data samples; for example, by means of response surface analysis, multiple linear regression or neural network, the functional relationship between input variables and output variables is calculated based on data samples. A process prediction model. The feedback control model is established according to the determined control target (objective function) and control conditions (constraints), and then can be calculated by programming (for example, C language or MATLAB, etc.) for the above model. Target process variable.

Step 103 is used to actually run the target process variable obtained in step 102, and obtain the actual output variable as the input variable of the actual device, and then the actual output variable obtained by the actual operation of step 103 and the required process target output variable through step 104. Compare and calculate the control error; if the control error is within the preset requirement, the overall feedback control process meets the requirements; and if the control error is not within the preset requirement, the process prediction model is adjusted according to certain rules ( The coefficients are generally adjusted) and then returned to step 102 for recalculation. The above process is executed cyclically until the calculated control error meets the preset requirements. The preset requirement may be: a control error of less than 1% or 0.5%, and the like.

The above specific operation process for the process prediction model and the feedback control model, as well as the specific process of feedback adjustment process prediction model, belong to the data principle part of the linear programming control method. It is well known to those skilled in the art and will not be described in detail herein.

It should be noted that in the actual equipment operation of step 103, the process variables that do not meet the requirements may be run, and the raw materials of these actual operations are invalidated, and these possible wastes are required as a process trial run.

Generally, for each new process target output variable, it needs to be controlled by the above several steps in the embodiment, so that the required target process variable is obtained under the premise of low calculation amount and high control precision.

In some cases, through the calculation of step 102, it is possible to obtain multiple sets of target process variables that conform to the feedback control model, but they cannot be handed over to the actual device for operation, because it causes a large waste. Therefore, preferably, when there are multiple sets of target process variables that meet the preset requirements, the embodiment shown in FIG. 1 can also select the optimal according to the deviation value (control error value) and/or the adjustment range of the single target process variable. A set of target process variables.

The adjustment range is the magnitude of the target process variable compared to the baseline (Baseline). From a process point of view, it is optimal to achieve the best control results with the smallest Recipe variation. Therefore, this parameter can be used to filter a set of optimal solutions, and in general, the adjustment is no more than 10%.

The core idea of the solution adopted by the present invention still uses the idea of the nonlinear control method, but for the nonlinear control idea, the specific solution and the specific control model are not seen in practice.

The inventor of this patent, the feedback control model originally proposed (referred to as the ordinary feedback control model) is as follows:

Min Z = (T ₀ - y(x _l , x ₂ , A , x _k )) ²

_{λ λ} ≤ χ _ι ≤Μ _ι

L, ≤ x ₂ ≤ M ₂

S.t

Μ where Min represents the minimum value for solving the objective function; st represents the constraint. Target output The expected value of the variable is T _Q ; for this specific process equipment, multiple process variables are required, and each process variable has a value range of ^e [A , , ^ = 1, 2, Λ, .

Although the above feedback control model is relatively easy to propose, since its constraint condition is a k-dimensional closed space, its calculation process is equivalent to searching for a suitable point in a space, and there may be multiple sets satisfying the objective function in this k-dimensional space. The optimal solution of (multidimensional surface), so the amount of calculation is very large.

Secondly, for the above feedback control model, a plurality of points satisfying the objective function in the k-dimensional space are calculated, and generally there are many requirements that do not satisfy the adjustment range of less than 10%. In actual process feedback, it is generally required that the adjustment range of the process input variable should not exceed 10%, otherwise it is very likely to cause process drift.

Specifically, in the above feedback control model, each process variable of the process can be freely changed within its own range of values until the deviation of the predicted value of the process output variable (control target) from the expected value satisfies the accuracy requirement of the control. Theoretically speaking, the above-mentioned process control law is indeed no problem, but in actual process production, it may have unreasonable factors, such as: etching process as a high-precision industry, if During the production process, the process recipe is greatly fluctuated due to the adjustment of the process control target, which may cause the process to drift (equipment instability), resulting in more waste.

In order to avoid this problem, the inventor of this patent, after further research, proposed a feedback control model based on the nearest neighbor rule (called the nearest neighbor rule feedback control model), as follows:

Among them, the constraint condition is a k-dimensional plane, and the calculation process is equivalent to searching for the best advantage in a k-dimensional plane, and the calculation amount is greatly reduced compared to searching in the k-dimensional space. The following is a specific example of using the nearest neighbor rule feedback control model to complete the present invention. The flow of feedback control is further explained:

Stepl: Selection of modeling sample data. According to the predetermined process output variables and input variables (process variables), a certain number of modeling samples are selected from the historical data according to the following factors: 1 consistency of sample data types; 2 rationality of sample data accuracy; 3 sample data changes Diversity. The purpose of the above selection rules is to obtain sample data that can fully reflect the actual situation. In order to more vividly illustrate the selection method of process sample data, a specific simulation example is given in the following paper: Data sample

Step2: The establishment of the process prediction model. After the input and output variables and data samples are determined, the functional relationship between the input variables and the output variables can be obtained by using multiple linear regression methods, that is, the process prediction model. Let the output variable of the process be the input variable ^, ^ , , then the prediction model of the process can be expressed by the following equation:

y(x _l , x ₂ ,A , x _k ) = a ₀ + α _ι χ _ι + α ₂ χ ₂ +Λ + a _k x _k

Based on the data samples given in Table 1, the multivariate linear regression method can be used to obtain the prediction model of the corresponding process, as follows:

7 = -0.1490 , +0.0586x, + 0.982 l , +0.6218 , +112.4246 The specific manner of feedback adjusting the preset process prediction model may be an adjustment coefficient, and the adjustment coefficient may include adjusting a gain coefficient, and may also include adjusting an intercept (eg, “ ⁰ ”).

Step3: The establishment of the feedback control model. The basic idea of the process adopted by the present invention closest to the control law is: pre-set the reference value of each process variable (also known as the Baseline of the process recipe), in the process of feedback control, the value of the process variable is To meet the error requirement between the predicted and expected values of the output variable, it must also be closest to the set reference value, ie the change between the newly generated Recipe and the reference value is minimal.

After the process control law is determined, the mathematical feedback method can be used to establish the feedback control model of the process. Set the baseline value of the process control variable Baseline to [ , ^ AA ] , and the expected value of the output variable is the feedback control model of the process:

)

Step4: Assume that the Baseline of the four process control variables are [80, 250, 65, 10.0] respectively. The output of the process feedback control in this case is shown in Table 2. The feedback control absolute error refers to the actual output. The absolute error between the value and the target output value (the values here are all absolute values, not percentage values).

Table 2 Calculation results of nonlinear control methods based on the nearest neighbor rule

190 79 250 70 10.1 0.3308

195 79 251 75 10.1 0.2999 In order to compare the further effects of the present invention, the output of the ordinary feedback control model for the feedback control model without the nearest rule of the present invention is recorded as follows:

Table 3 Calculation results of ordinary nonlinear control methods

From the comparison of the data shown in Table 2 and Table 3 above, it can be known that the nonlinear control method using the nearest neighbor law can achieve higher control precision.

For a clearer explanation, referring to Fig. 2, an error comparison diagram of the common feedback control model and the most recent rule feedback control model is shown. Comparing the error results of the process feedback control method of the nearest rule under the same conditions with the general nonlinear process control method, it can be found that: The process feedback control method using the nearest neighbor rule is more suitable for the actual process requirements, and it is controlled by feedback. Accuracy is also more in line with current process standards than existing process control methods, and this is the key to achieving advanced process automation control in the future.

The nearest feedback control model in the above specific example uses the standard of relative distance. In fact, the change between the feedback value and the reference value can be used in addition to the relative distance. Absolute distance, weighted distance and other methods are described. In this case, the process feedback control model can be expressed by the following equation:

Model one

0 Model 2 In the above two feedback control models, model one is the absolute distance control model (which can also be weighted), and model two is the weighted distance control model (weights satisfy ^W ! ^{+ W} 2 + ^3 ^{+ W} 4 ^{= 1} in actual In the control process, the weights of each process variable can be determined according to the importance of each process variable and their correlation with the control target. If the two models are given reasonable initial conditions, the feedback control of the process can be similarly realized.

Referring to Figure 3, an embodiment of a process control device is shown that can be used as a stand-alone feedback control device, typically as a stand-alone functional module integrated into an Advanced Process Control (APC) system. The device may specifically include the following components: a process target receiving module 301, configured to receive a target output variable;

The feedback processing module 302 is configured to calculate a target process variable by using a preset process prediction model and a feedback control model; the process prediction model is used to describe a functional relationship between the process variable and the output variable; and the objective function of the feedback control model For: solving a target process variable with a minimum distance from a preset reference value; the constraint condition of the feedback control model is: the target output variable and the target process variable are in accordance with a process prediction model; and the target process variable is between a preset reference value and a preset reference value Distances include: relative distance, absolute distance, or weighted distance. The process prediction model can be obtained by data sample analysis.

a device interface module 303, configured to input the target process variable to an actual process device, and detect an actual output variable;

a judgment feedback module 304, configured to determine the target output variable and the actual output variable If the preset requirement is not met, the preset process prediction model is adjusted and returned to the feedback processing module; if the preset requirement is met, the process ends. The manner of adjusting the preset process prediction model may include adjusting coefficients and/or intercepts.

The above module division is only a function division corresponding to the feedback control flow in the present invention. In fact, other module division manners may exist. For example, the judgment feedback module 304 is divided into a judgment module and a feedback adjustment module. It is within the scope of the device embodiment to split or recombine the functions of the various modules described above.

When there are multiple sets of target process variables that meet preset requirements, the apparatus shown in FIG. 3 may further include a screening module for filtering an optimal set of targets according to the deviation value and/or the adjustment range of the single target process variable. Process variable. The set of target process variables is then transferred to the actual device for operation.

Each of the embodiments in the present specification is described in a progressive manner, and each embodiment focuses on differences from other embodiments, and the same similar parts between the respective embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

The above description of the method and apparatus for process feedback control provided by the present invention is only for helping to understand the method of the present invention and its core idea; and, for a person of ordinary skill in the art, The present invention is not limited by the scope of the present invention.

Claims

UP-081624-00 request

A process control method, characterized in that it comprises:

Step a, receiving a target output variable;

Step b: calculating a target process variable by using a preset process prediction model and a feedback control model; the process prediction model is used to describe a functional relationship between the process variable and the output variable; the objective function of the feedback control model is: a target process variable having a minimum preset reference distance; the constraint condition of the feedback control model is: the target output variable and the target process variable conform to a process prediction model;

Step c, inputting the target process variable to the actual process device to obtain an actual output variable; Step d, determining a deviation between the target output variable and the actual output variable; if the preset requirement is not met, adjusting the The preset process prediction model returns to step b; if the preset requirement is met, it ends.

2. The method of claim 1, wherein the distance between the target process variable and the preset reference value comprises: a relative distance, an absolute distance, or a weighted distance.

3. The method of claim 1, wherein the adjusting the preset process prediction model comprises adjusting a gain factor and/or an intercept.

4. The method of claim 1 wherein the process prediction model is obtained by data sample analysis.

5. The method according to claim 1, wherein when there are a plurality of sets of target process variables that meet preset requirements, an optimal set is selected according to the offset value and/or the adjustment range of the single target process variable. Target process variable.

6. A process control device, comprising:

a process target receiving module, configured to receive a target output variable;

a judgment feedback module, configured to determine a deviation between the target output variable and the actual output variable; if the preset requirement is not met, adjusting the preset process prediction model, and returning to the feedback processing module; If requested, it ends.

7. The apparatus according to claim 6, wherein the distance between the target process variable and the preset reference value comprises: a relative distance, an absolute distance, or a weighted distance.

8. The apparatus of claim 6, wherein the manner of adjusting the preset process prediction model comprises adjusting a gain factor and/or an intercept.

9. Apparatus according to claim 6 wherein said process prediction model is obtained by data sample analysis.

10. The device of claim 6, further comprising:

The screening module is configured to select an optimal set of target process variables according to the deviation value and/or the adjustment range of the single target process variable when there are multiple sets of target process variables that meet the preset requirements.