WO2022244547A1 - 情報処理システムおよび処理条件決定システム - Google Patents
情報処理システムおよび処理条件決定システム Download PDFInfo
- Publication number
- WO2022244547A1 WO2022244547A1 PCT/JP2022/017395 JP2022017395W WO2022244547A1 WO 2022244547 A1 WO2022244547 A1 WO 2022244547A1 JP 2022017395 W JP2022017395 W JP 2022017395W WO 2022244547 A1 WO2022244547 A1 WO 2022244547A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- function
- unit
- processing
- kernel
- objective
- Prior art date
Links
- 230000010365 information processing Effects 0.000 title claims abstract description 38
- 238000012545 processing Methods 0.000 title claims description 184
- 230000006870 function Effects 0.000 claims abstract description 280
- 238000000034 method Methods 0.000 claims abstract description 110
- 238000006243 chemical reaction Methods 0.000 claims abstract description 46
- 238000010801 machine learning Methods 0.000 claims abstract description 46
- 238000000137 annealing Methods 0.000 claims abstract description 18
- 238000009795 derivation Methods 0.000 claims abstract description 13
- 238000004458 analytical method Methods 0.000 claims description 53
- 238000012887 quadratic function Methods 0.000 claims description 36
- 238000005457 optimization Methods 0.000 claims description 29
- 238000012886 linear function Methods 0.000 claims description 23
- 239000013598 vector Substances 0.000 claims description 21
- 230000009977 dual effect Effects 0.000 claims description 11
- 230000001131 transforming effect Effects 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 6
- 238000012905 input function Methods 0.000 claims description 2
- 238000004148 unit process Methods 0.000 claims 1
- 230000005366 Ising model Effects 0.000 abstract description 18
- 239000011159 matrix material Substances 0.000 description 10
- 238000010586 diagram Methods 0.000 description 7
- 238000004519 manufacturing process Methods 0.000 description 5
- 239000000654 additive Substances 0.000 description 4
- 230000000996 additive effect Effects 0.000 description 4
- 238000010438 heat treatment Methods 0.000 description 4
- 239000000243 solution Substances 0.000 description 4
- 239000004065 semiconductor Substances 0.000 description 3
- 238000005229 chemical vapour deposition Methods 0.000 description 2
- 238000000151 deposition Methods 0.000 description 2
- 230000008021 deposition Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005240 physical vapour deposition Methods 0.000 description 2
- 238000003825 pressing Methods 0.000 description 2
- 240000006927 Foeniculum vulgare Species 0.000 description 1
- 238000001015 X-ray lithography Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000011230 binding agent Substances 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013400 design of experiment Methods 0.000 description 1
- 238000001312 dry etching Methods 0.000 description 1
- 238000010894 electron beam technology Methods 0.000 description 1
- 238000000609 electron-beam lithography Methods 0.000 description 1
- 238000001125 extrusion Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000005468 ion implantation Methods 0.000 description 1
- 238000010884 ion-beam technique Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000004093 laser heating Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000001459 lithography Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000000465 moulding Methods 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 238000000059 patterning Methods 0.000 description 1
- 239000000843 powder Substances 0.000 description 1
- -1 sheet lamination Substances 0.000 description 1
- 238000002922 simulated annealing Methods 0.000 description 1
- 238000004544 sputter deposition Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000007740 vapor deposition Methods 0.000 description 1
- 238000001039 wet etching Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N99/00—Subject matter not provided for in other groups of this subclass
Definitions
- the present invention relates to an information processing system and a processing condition determination system.
- annealing machine As an effective analysis device for efficiently solving combinatorial optimization problems, there is an annealing machine (or Ising machine) that converts the objective function into an Ising model and searches for a global solution using the annealing method.
- annealing methods mainly include simulated annealing and quantum annealing.
- An Ising model is a model that considers up to a first-order term and a second-order term for multiple spin variables that take the value of -1 or 1, and is part of combinatorial optimization problems such as the traveling salesman problem. It is known that the objective function of can be represented by an Ising model. However, the objective function in many actual combinatorial optimization problems is generally not formulated in advance, and the Ising model is not defined.
- Patent document 1 is a conventional technique for finding an optimum combination using an annealing machine in such a case.
- Patent Document 1 describes formulating an objective function from data and optimizing conditions for minimizing or maximizing it using annealing such as quantum annealing.
- Cited Document 1 does not disclose a specific mapping method for converting the objective function into the Ising model.
- the purpose of the present invention is to provide an information processing system that enables the search for optimal solutions by means of an arraying, etc., by converting a strongly nonlinear objective function derived by machine learning into an Ising model.
- the present invention analyzes a learning database consisting of sample data regarding one or more explanatory variables and one or more objective variables, and generates an unconstrained quadratic function or a linear function with linear constraints.
- an objective function deriving system for deriving an objective function from the learning database by machine learning; a function conversion system for converting to a linear formal function, wherein the objective function derivation system includes a machine learning setting unit for setting details of a machine learning method, and a machine learning method set by the machine learning setting unit.
- a learning unit for deriving the objective function using the A dummy variable generation unit that generates the dummy variables based on the generation method set by the dummy variable setting unit; , the dimension of the nonlinear term of the explanatory variable higher than the second order is reduced to the second order or less, and the objective function is converted into the unconstrained quadratic function or the linearly constrained linear function with respect to the dummy variable and the objective variable. and a function conversion unit for converting.
- an objective function with strong nonlinearity higher than quadratic is transformed into an unconstrained quadratic function or a linearly constrained linear function, and then subjected to annealing, linear programming, integer programming, or the like. It becomes possible to search for the optimum solution.
- FIG. 1 is a configuration example of an information processing system in Example 1;
- FIG. 4 is a diagram illustrating a typical objective function derived using machine learning;
- 2B is a diagram illustrating an unconstrained quadratic function obtained by generating dummy variables for the objective function shown in FIG. 2A;
- FIG. 4 is a flow chart of the information processing system according to the first embodiment; It is an example of a learning database. It is an example of a learning database. This is an example of true regression with explanatory and objective variables.
- FIG. 10 is a diagram showing how to estimate a regression and search for the value of an explanatory variable that gives the maximum value.
- FIG. 10 is a diagram showing how to estimate an acquisition function using Bayesian optimization and search for the value of an explanatory variable that gives the maximum value.
- FIG. 6B shows an example output of an unconstrained quadratic function for the variables in FIG. 6A.
- FIG. 6B shows an example of the output result (coefficient vector) of the linear-constrained linear function for the variables in FIG. 6A.
- FIG. 6B shows an example of the output result (constraint matrix, constraint constant vector) of the linearly constrained linear function with respect to the variables in FIG. 6A.
- FIG. 11 is a configuration example of an information processing system according to a second embodiment;
- FIG. 10 is a flowchart of an information processing system in Example 2;
- FIG. 11 is a configuration example of a processing condition determination system in Example 3;
- FIG. 11 is a flow chart of a processing condition determination system in Example 3;
- It is an example of GUI for input.
- GUI for output It is an example of GUI for output.
- the Ising model is a model that considers up to the quadratic terms of variables, in order to convert a general combinatorial optimization problem to the Ising model, terms higher than quadratic are defined as additional spin variables. Therefore, a transformation that drops the dimension is required.
- the cubic term X 1 X 2 X 3 can be transformed into the quadratic term Y 12 X 3 by putting the product of the two spin variables X 1 X 2 with the new spin variable Y 12 .
- the objective function obtained using machine learning generally has many high-order nonlinear terms, if it is converted to an Ising model by the above method, a large-scale additional spin variable is required. For example, consider the case where the highest degree of the objective function is 10.
- the present embodiment provides a technique for converting a strongly nonlinear objective function derived by machine learning into an Ising model at high speed and with high accuracy so that the number of dummy variables does not become enormous. As a result, it becomes possible to perform optimization using an annealing machine or the like by converting a complex problem in the real world into an objective function by machine learning and further converting it into an Ising model.
- This embodiment converts an objective function obtained using machine learning, especially the kernel method, into an Ising model.
- the Ising model is known to be equivalent to an unconstrained quadratic function (Quadratic Unconstrained Binary Optimization model) for variables that can align binary variables that take only 0 or 1 by a predetermined transformation. Therefore, a method of converting an objective function f(X) related to a binary explanatory variable X into an unconstrained quadratic function by appropriately generating dummy variables will be described below.
- the unconstrained quadratic function on the variable vector x is a function of at most quadratic on x, and using a symmetric matrix Q with the same number of rows and columns as the number of dimensions of x, the following (Equation 1 ) is a function expressed as
- the superscript T indicates a transpose operation on a matrix.
- This matrix Q is hereinafter referred to as a coefficient matrix.
- the objective function is represented by a linear sum of kernel functions according to the Representer theorem. Since the objective function itself is generally a highly nonlinear function, the above method requires a large number of additional binary variables, ie dummy variables, to convert to an unconstrained quadratic function. However, the kernel function is weaker in nonlinearity than the objective function, and can be converted or approximated to an unconstrained quadratic function using a small number of dummy variables as described later.
- a linearly constrained linear function with respect to the variable vector x is a vector a whose dimension is equal to the number of dimensions of x, a matrix A whose number of columns is equal to the number of dimensions of x, and a number of dimensions which is equal to the number of rows of A. It is a function represented by the following (Equation 2) using a vector c having
- the vector a is called a coefficient vector
- the matrix A is called a constraint matrix
- the vector c is called a constraint constant vector.
- FIG. 1 is a diagram showing a configuration example of an information processing system according to the first embodiment.
- the information processing system of the first embodiment uses machine learning to derive an objective function from data of explanatory variables and objective variables. Convert to linear form function and output.
- the information processing system 100 includes an objective function derivation system 200 that derives an objective function from sample data relating to one or more explanatory variables X and one or more objective variables Y using machine learning, and an additional a function transformation system 300 that generates a dummy variable X' and transforms the objective function into an unconstrained quadratic form function or a linearly constrained linear form function with respect to X and X'.
- FIG. 2A illustrates an example of the objective function obtained by the objective function derivation system 200
- FIG. 2B illustrates an unconstrained quadratic function as an example of the Ising model obtained by the function conversion system 300. is.
- the objective function derivation system 200 includes a learning database 210 that stores sample data of the explanatory variable X and the objective variable Y, a machine learning setting unit 220 that sets the details of the machine learning method (type, specification, etc.), and a machine learning setting and a learning unit 230 that derives an objective function using the machine learning method set in the unit 220 .
- the learning database 210 stores, as structured data, values of explanatory variables and objective variables for the number of samples.
- the number of objective variables may be two or more as shown in FIG. 4B.
- FIG. 5A shows an example of a true regression function
- FIG. 5B shows a regression function obtained by estimating this true regression from the sample data shown in FIG. 4 .
- an acquisition function by Bayesian optimization shown in FIG. 5C is also conceivable.
- the acquisition function is the regression function estimated in FIG. 5B modified using the prediction variance.
- the function conversion system 300 includes a dummy variable setting unit 310 that sets a dummy variable generation method, a dummy variable generation unit 320 that generates dummy variables based on the generation method set in the dummy variable setting unit 310, and a learning unit 230. and a function conversion unit 330 that converts the objective function obtained in (1) into an unconstrained quadratic function or a linear constraint linear function and outputs the result.
- the function transforming unit 330 eliminates one or more explanatory variables that explicitly appear in the objective function using dummy variables, so that the The nonlinear term of is reduced to the second or lower dimension.
- FIG. 6A shows a list of variables possessed by the quadratic form function without constraints or the linear form function with linear constraints, in which the original explanatory variables and the dummy variables generated by the dummy variable generation unit 320 are displayed.
- FIG. 6B When converted into an unrestricted quadratic function, each element of the coefficient matrix of (Formula 1) is output as shown in FIG. 6B.
- FIGS. 6C and 6D When converted to a linear form function with linear constraints, each element of the coefficient vector, constraint matrix, and constraint constant vector of (Equation 2) is output as shown in FIGS. 6C and 6D.
- FIG. 3 is a flowchart for outputting a quadratic function without constraints or a linear function with linear constraints by the information processing system 100 from a state in which sample data of the explanatory variable X and the objective variable Y are stored in the learning database 210.
- a method for the information processing system 100 of this embodiment to output a quadratic function without constraints or a linear function with linear constraints will be described below with reference to FIG.
- the machine learning setting unit 220 sets the details of the machine learning method for deriving the objective function (step S101). For example, a learning type such as a kernel method, neural network, or decision tree is selected. In step S101, learning hyperparameters such as the type of kernel function and activation function, the depth of the tree, and the learning rate during error backpropagation are also set.
- a learning type such as a kernel method, neural network, or decision tree is selected.
- learning hyperparameters such as the type of kernel function and activation function, the depth of the tree, and the learning rate during error backpropagation are also set.
- the learning unit 230 uses the data stored in the learning database 210 to learn the objective function by machine learning under various conditions set by the machine learning setting unit 220, and outputs the objective function to the function conversion unit 330. (step S102).
- the user determines the dummy variable generation method and inputs it to the dummy variable setting unit 310 (step S103).
- the generation method can be set by giving a constraint formula such as (Equation 3) below that some function related to the dummy variable X' and the explanatory variable X is identically established.
- the dummy variable generation unit 320 generates dummy variables by the dummy variable generation interpolation method set in step S103 (step S104). That is, the dummy variable generator 320 generates a dummy variable X' that satisfies (Formula 3).
- K By setting K to be smaller than the order of the objective function, it is possible to suppress the number of dummy variables, and even if the objective function obtained by machine learning is highly nonlinear, the objective function can be converted into an Ising model. become.
- a dummy variable generation method described in step S204 of the second embodiment may be set.
- the function conversion unit 330 converts the objective function derived by the learning unit 230 into an unconstrained quadratic function or a linear constraint linear function and outputs it (step S105).
- the unconstrained quadratic function or the linear function with linear constraints is a function related to explanatory variables and dummy variables.
- the function conversion unit 330 performs the above conversion by eliminating one or more explanatory variables appearing in one or more terms of the objective function with the dummy variables generated by the dummy variable generation unit 320. do.
- the function conversion unit 330 when the function conversion unit 330 outputs a linear-form function with linear constraints, it is limited only when the constraint equation of (Equation 3) is linear, and there is a linear constraint shown in (Equation 2).
- the constraints that the linear form function has are given by (Equation 3).
- the converted objective function when outputting a non-restricted quadratic function from the function transforming unit 330, the converted objective function is output by adding a penalty term related to the constraint expression of (Equation 3). However, the penalty term is limited to quadratic form.
- FIG. 7 is a diagram showing a configuration example of an information processing system according to the second embodiment.
- the information processing system 100 of the second embodiment derives an objective function from the data of explanatory variables and objective variables using the kernel method, converts the objective function into an unconstrained quadratic function or a linearly constrained linear function, and outputs it. do.
- Information processing system 100, objective function derivation system 200, learning database 210, machine learning setting unit 220, learning unit 230, function conversion system 300, dummy variable setting unit 310, dummy variable generation unit 320, and function conversion unit 330 are defined as follows: It is common with Example 1.
- the machine learning setting unit 220 in this embodiment includes a kernel method selection unit 221 and a kernel function selection unit 222 .
- a kernel method selection unit 221 selects a kernel method to be used for deriving the objective function.
- the kernel function selection unit 222 selects the type of kernel function used in the kernel method.
- FIG. 8 is a flowchart for the information processing system 100 to output a quadratic function without constraints or a linear function with linear constraints from the state in which the learning database 210 stores sample data of the explanatory variable X and the objective variable Y.
- a method of outputting an unconstrained quadratic function or a linearly constrained linear function will be described below with reference to FIG.
- the user selects one of kernel regression, Bayesian optimization, and multi-objective optimization by kernel regression in the kernel method selection unit 221 (step S201).
- the Bayesian optimization method is selected to efficiently determine the next search data when there is a sparse learning area with little data.
- a method of multi-objective optimization using kernel regression is selected.
- kernel function type in the kernel method selected in step S201 in the kernel function selection unit 222 (step S202).
- Types of kernel functions include functions such as RBF (Radial Basis Function) kernels, polynomial kernels, and sigmoid kernels.
- the learning unit 230 uses the kernel method selected by the kernel method selection unit 221 and the kernel function selected by the kernel function selection unit 222 to learn and derive the objective function (step S203).
- the derived objective function is the regression function if kernel regression is selected, the acquisition function if Bayesian optimization is selected, or the linear regression function for each objective variable for multi-objective optimization with kernel regression. Harmony.
- the kernel method selection unit 221 selects the kernel regression
- the kernel function selected by the kernel function selection unit 222 is approximated by adding one or more basis functions to obtain a new kernel function for learning.
- the objective function is derived by kernel regression using the new kernel function.
- the user determines the dummy variable generation method and inputs it to the dummy variable setting unit 310 (step S204).
- the following two generation methods are particularly exemplified.
- the first generation method is to generate dummy variables by applying one-hot encoding to possible values of the kernel function.
- the one-hot encoding for the variable x with M-level values ⁇ 1 , ⁇ 2 ,.., ⁇ M ⁇ means that several levels are is a method of generating dummy variables x'1 , x'2 , .. , x'M .
- This generation method assumes a vector of binary variables as explanatory variables. Since the number of possible values of the kernel function obtained by substituting the explanatory variables, ie, the level, is proportional to the dimension of the explanatory variables, it is possible to prevent the number of dummy variables from becoming enormous. As a result, even if the objective function obtained by machine learning is highly nonlinear, the objective function can be converted into an Ising model, and the objective function can be optimized using an annealing machine or the like. (Equation 5) and (Equation 6) can be easily transformed into the form of (Equation 3). From (Formula 5) and (Formula 6), the kernel function can be represented by a linear form function with linear constraints on explanatory variables and dummy variables.
- the second generation method is a method of approximating the kernel function by performing fitting with the addition of one or more basis functions, and defining the conjugate variable in the dual transformation of this basis function as a dummy variable.
- a dual problem consider one of Lagrange dual problem, Fenchel dual problem, Wolfe dual problem, and Legendre dual problem.
- a vector of binary variables was assumed as explanatory variables, but the second generation method is not limited to this assumption.
- the basis functions used here are expressed in quadratic form with respect to the conjugate variable of the dual problem and the explanatory variables.
- the kernel function can be approximated by a quadratic function with respect to the explanatory variables and the dummy variables by using dummy variables as the conjugate variables of the above dual problem.
- the constraint condition of the conjugate variable required in the dual problem may be the constraint expression (Equation 3) satisfied by the dummy variable.
- the kernel function is expressed as an unconstrained quadratic form function.
- Steps S205 and S206 that follow have the same definitions as steps S104 and S105 in the first embodiment, respectively.
- step S201 when the kernel method selection unit 221 selects kernel regression or multi-objective optimization by kernel regression, the function conversion unit 330 converts the objective function into a linear function related to dummy variables, By imposing restrictions on the dummy variables generated by the variable generation unit 320, a constrained linear form function is derived. At this time, the function conversion unit 330 can also derive an unconstrained quadratic function by adding a quadratic form penalty term for the constraint to the converted linear form function. Also, in step S201, when the kernel method selection unit 221 selects Bayesian optimization, the function conversion unit 330 converts the objective function into a quadratic function related to dummy variables, and derives an unconstrained quadratic function. .
- FIG. 9 is a diagram showing a configuration example of a processing condition determination system according to the third embodiment.
- Information processing system 100 objective function derivation system 200, learning database 210, machine learning setting unit 220, kernel method selection unit 221, kernel function selection unit 222, learning unit 230, function conversion system 300, dummy variable setting unit 310, dummy variables Definitions of the generator 320 and the function converter 330 are common to the second embodiment.
- the processing condition determination system 400 includes an objective function derivation system 200, a function conversion system 300, a processing device 500, a learning data generation unit 600, and a processing condition analysis system 700. It is a system that decides.
- the processing device 500 is a device that processes a target sample by some process.
- the processing equipment 500 includes semiconductor processing equipment.
- Semiconductor processing equipment includes lithography equipment, film deposition equipment, patterning equipment, ion implantation equipment, heating equipment, cleaning equipment, and the like.
- the lithographic apparatus includes an exposure apparatus, an electron beam lithography apparatus, an X-ray lithography apparatus, and the like.
- the film forming apparatus includes a CVD (Chemical Vapor Deposition) apparatus, a PVD (Physical Vapor Deposition) apparatus, a vapor deposition apparatus, a sputtering apparatus, a thermal oxidation apparatus, and the like.
- Pattern processing devices include wet etching devices, dry etching devices, electron beam processing devices, laser processing devices, and the like.
- Ion implanters include plasma doping devices and ion beam doping devices.
- Heating devices include resistance heating devices, lamp heating devices, and laser heating devices.
- the processing device 500 may also be an additive manufacturing device.
- Additive manufacturing equipment includes each type of additive manufacturing equipment such as liquid bath photopolymerization, material extrusion, powder bed fusion bonding, binder injection, sheet lamination, material injection, and directed energy deposition. Note that the processing apparatus 500 is not limited to a semiconductor processing apparatus or an additive manufacturing apparatus.
- the processing device 500 includes a processing condition input unit 510 for inputting processing conditions output from the processing condition analysis system 700, and a processing unit 520 for performing processing of the processing device 500 using the processing conditions input by the processing condition input unit 510. and have In FIG. 9, the processing result acquisition unit 530 that acquires the processing result of the processing unit 520 is installed in the processing device 500, but it may be a stand-alone device separate from the processing device 500. In the processing unit 520, a sample is placed inside and processed.
- the learning data generation unit 600 processes (converts) the processing conditions input to the processing condition input unit 510 into explanatory variable data and the processing results acquired by the processing result acquisition unit 530 into target variable data, and then Store in learning database 210 .
- the processing condition analysis system 700 uses an analysis method selection unit 710 that selects an analysis method for the value of the explanatory variable according to the type of function derived by the function conversion system 300, and the analysis method selected by the analysis method selection unit. Then, the processing condition analysis unit 720 calculates the value of the explanatory variable that gives the minimum value or maximum value of the input function, and the value of the explanatory variable obtained by the processing condition analysis unit 720 is processed (converted) into the processing condition. and a processing condition output unit 730 for outputting to the processing device 500.
- FIG. 10 is a flowchart for determining the processing conditions of the processing device 500.
- FIG. A method for determining the processing conditions of the processing device 500 by the processing condition determining system 400 will be described below with reference to FIG.
- the user inputs arbitrary processing conditions in the processing condition input unit 510 (step S301).
- the processing conditions input here are called initial processing conditions, and there may be a plurality of them.
- the processing unit 520 processes the sample (step S302). However, if the processed sample remains in the processing section 520, it is removed and a new unprocessed sample is placed in the processing section 520 before processing. When there are multiple processing conditions, the sample is replaced each time and processed.
- the processing result acquisition unit 530 acquires the processing result (step S303). If the processing result satisfies the user, the process ends; otherwise, the process proceeds to step S305 (step S304).
- the learning data generation unit 600 converts the processing conditions input by the processing condition input unit 510 into explanatory variable data, and converts the processing results acquired by the processing result acquisition unit 530 into objective variable data. Then, it is stored in the learning database 210 and the learning database is updated (step S305).
- the learning data generating unit 600 can convert the processing condition data into explanatory variable data represented by binary variables by performing binary conversion or one-hot encoding.
- the learning data generation unit 600 may perform normalization or the like, or may perform a combination of two or more of these conversion methods.
- step S306, step S307, step S308, step S309, step S310 and step S311 are common to step S201, step S202, step S203, step S204, step S205 and step S206 of the second embodiment, respectively. be.
- the function conversion unit 330 of the function conversion system 300 converts the objective function for the explanatory variable and the objective variable into an unconstrained quadratic function or a linear constraint linear function, and outputs the result. 710 selects how to parse this function (step S312). That is, if the function output in S311 is an unconstrained quadratic function, annealing is selected. If the function output in S311 is a linear-form function with linear constraints, integer programming or linear programming is selected.
- the processing condition determination system 400 of this embodiment can handle not only unconstrained quadratic functions but also linear functions with linear constraints by selecting an appropriate analysis method according to the objective function. , can be used properly by the user.
- the processing condition analysis unit 720 analyzes the function output in S311 using the analysis method selected by the analysis method selection unit 710 (step S313).
- the analysis in step S313 it is possible to search for the value x opt of the variable x that maximizes or minimizes this function.
- such X opt is searched for and output.
- the processing condition output unit 730 converts the explanatory variable value data obtained in step S313 into processing conditions (step S314). A plurality of processing conditions may be obtained. Next, based on the processing conditions obtained in step S314, the user determines whether or not the processing of the processing device 500 can be executed. Input to the processing condition input unit 510 . If execution is not possible, the process returns to step S305 and subsequent steps, and the kernel method, kernel function, dummy variable generation method, and analysis method are reset. By repeating a series of steps from step S302 to step S315 until a processing result satisfying the user is obtained in step S304, it is possible to determine good processing conditions.
- FIG. 11 shows an input GUI 1200, which is an example of an input screen for inputting settings for the processing condition determination system 400 of the third embodiment. It is assumed that this input screen is presented during the procedure of step S301.
- the input GUI 1200 includes an initial processing condition setting box 1210, a learning data setting box 1220, a machine learning setting box 1230, a dummy variable setting box 1240, an analysis method setting box 1250, a valid/invalid display section 1260, A determination button 1270 is provided.
- the initial processing condition setting box 1210 has a condition input section 1211.
- the condition input unit 1211 for example, the data number, the name of each factor of the processing condition, and the value of each factor of each data can be input to a structure such as a csv file as the initial processing condition.
- These factors are the control factors of the processing device 500, and in the example of FIG. 11 the factors are power and pressure.
- the initial processing conditions can be input to the processing condition input unit 510 of the processing device 500 .
- the learning data setting box 1220 has a conversion method input section 1221.
- the conversion method input unit 1221 for example, one or more of one-hot encoding, binary conversion, and normalization are used to select a method of converting the processing conditions into explanatory variable data. Although only one method is selected in FIG. 11, multiple methods may be selected. Learning data generation in the learning data generation unit 600 is performed using the input method.
- the machine learning setting box 1230 has a kernel method input section 1231 and a kernel function input section 1232.
- Kernel method input unit 1231 selects, for example, kernel regression, Bayesian optimization, or multi-objective optimization by kernel regression. With the above inputs, the kernel method selection unit 221 selects a kernel method. In FIG. 11, multi-objective optimization by kernel regression is simply abbreviated as multi-objective optimization.
- a kernel function input unit 1232 selects an RBF kernel, a polynomial kernel, a Sigmoid kernel, or the like. Kernel function selection in the kernel function selection unit 222 is performed by the input here.
- the dummy variable setting box 1240 has a generation method input section 1241.
- the generation method input unit 1241 can select, for example, three dummy variable generation methods (abbreviated as one-hot, basis function expansion, and approximation up to K-th order). It's like One-hot is the first generation method described in Example 2, and is a method by one-hot encoding for kernel functions.
- Basis function expansion is the second generation method described in the second embodiment, and is a method of approximating a kernel function by summing basis functions and defining conjugate variables in the dual problem of these basis functions as dummy variables.
- the approximation up to the K order is the generation method described in Example 1.
- the analysis method setting box 1250 has an analysis method input section 1251. Annealing, integer programming, linear programming, etc. are selected as analysis methods. By the input here, the analysis method setting in the analysis method selection unit 710 is performed.
- Whether or not the above inputs are valid is displayed by the valid/invalid display section 1260 provided in each of the above setting boxes.
- the decision button 1270 of the input GUI 1200 is pressed, the procedure of FIG. 10 is executed, and the processing result output GUI 1300 is shown in FIG. 12A as an output GUI presented after step S303.
- This GUI displays the current status and allows the user to choose whether or not to proceed to the next step.
- the processing result output GUI 1300 includes a processing result display section 1310 , a completion/continue selection section 1320 and an enter button 1330 .
- the processing result display unit 1310 has a sample display unit 1311 and a processing result display unit 1312.
- the sample display section 1311 shows the state of the sample after the processing in step S302 is completed.
- the processing result display unit 1312 displays the processing result obtained in step S303.
- FIG. 12A shows the processing result output GUI 1300 when assuming an additional manufacturing device as the processing device 500 and a screw-shaped molded object as the target sample.
- the sample display area 1311 shows the appearance of the screw-shaped molded object after the molding process
- the processing result display area 1312 shows the height of the molded object and the defect rate.
- the user can select whether to complete or continue in the completion/continuation selection section 1320 based on the information displayed in the processing result display section 1310. That is, the user can perform the work of step S304 on this GUI. If the user is satisfied with the results of the processing, he selects Done and presses the OK button 1330, thereby ending the steps as shown in FIG. If the user decides that he/she is not satisfied, he/she selects continuation and presses the determination button 1330, and the process proceeds to step S305.
- the decision button 1330 of the processing result output GUI 1300 is pressed, the procedure of FIG. 10 is executed, and the analysis result output GUI 1400 is shown in FIG. 12B as the output GUI presented after step S314.
- This GUI displays the current status and allows the user to choose whether or not to proceed to the next step.
- the analysis result output GUI 1400 includes an analysis result display section 1410 , a continuation/reset selection section 1420 and an enter button 1430 .
- the analysis result display unit 1410 has an objective function display unit 1411, a function conversion result display unit 1412, and a processing condition analysis result display unit 1413.
- the objective function display unit 1411 displays information on the objective function derived in step S308.
- the function conversion result display unit 1412 displays information on the unconstrained quadratic function or the linearly constrained linear function derived in step S311.
- the processing condition analysis result display unit 1413 displays the processing conditions obtained in step S314.
- the objective function display section 1411 displays values of hyperparameters, training errors, generalization errors, etc. when deriving the objective function in the learning section 230 .
- the function conversion result display section 1412 the information shown in FIG. 7 output from the function conversion section 330 is displayed.
- items such as explanatory variables and dummy variables, coefficient vectors, constraint matrices, and constrained constant vectors in linear form with linear constraints, coefficient matrices in quadratic form functions without constraints, and the like are displayed.
- the processing condition analysis result display section 1413 the processing conditions output from the processing condition output section 730 are displayed.
- the user can select continuation or resetting in the continuation/resetting selection section 1420. That is, the user can perform the work of step S315 on this GUI.
- the user determines that the processing conditions displayed in the processing condition analysis result display unit 1413 can be used to perform the processing of the processing device 500, continuation is selected, and by pressing the decision button 1430, the A processing condition is input to the processing condition input unit 510, and the process proceeds to step S302.
- the process proceeds to step S306.
- resetting may be selected when it is determined that the numerical value of a specific factor of the processing condition displayed in the processing condition analysis result display unit 1413 is not preferable for the operation of the processing apparatus 500 . Further, when it is expected that the objective function is over-learning based on the objective function information displayed in the objective function display section 1411, resetting may be selected. Furthermore, resetting may be selected when the number of dummy variables of the function displayed in the function conversion result display section 1412 exceeds the user's reference value.
- 100 information processing system, 210: learning database, 220: machine learning setting unit, 221: kernel method selection unit, 222: kernel function selection unit, 230: learning unit, 300: function conversion system, 310: dummy variable setting unit, 320: dummy variable generation unit, 330: function conversion unit, 400: processing condition determination system, 500: processing device, 510: processing condition input unit, 520: processing unit, 530: processing result acquisition unit, 600: learning data generation unit , 700: processing condition analysis system, 710: analysis method selection unit, 720: processing condition analysis unit, 730: processing condition output unit, 1200: input GUI, 1210: initial processing condition setting box, 1211: condition input unit, 1220 : learning data setting box, 1221: conversion method input section, 1230: machine learning setting box, 1231: kernel method input section, 1232: kernel function input section, 1240: dummy variable setting box, 1241: generation method input section, 1250: Analysis method setting box, 1251: analysis method input section, 1260: valid/invalid display section,
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Complex Calculations (AREA)
Abstract
Description
ここで、変数ベクトルxに関する線形制約あり一次形式関数とは、xの次元数と等しい次元のベクトルaと、xの次元数と等しい列数を持つ行列Aと、Aの行数と等しい次元数を持つベクトルcとを用いて以下の(式2)ように表される関数である。
図2Aは、目的関数導出システム200により得られた目的関数の一例を図示したものであり、図2Bは関数変換システム300により得られたイジングモデルの一例として制約なし二次形式関数を図示したものである。
さらに、特に目的変数が2つ以上ある場合、すなわち多目的最適化の場合は、目的関数として目的変数ごとの回帰関数の線形和を選択することも可能である。
リソグラフィ装置としては、露光装置、電子線描画装置およびX線描画装置などがある。
製膜装置としては、CVD(Chemical Vapor Deposition)装置、PVD(Physical Vapor Deposition)装置、蒸着装置、スパッタリング装置および熱酸化装置などがある。パターン加工装置としては、ウェットエッチング装置、ドライエッチング装置、電子ビーム加工装置およびレーザー加工装置などがある。イオン注入装置としては、プラズマドーピング装置およびイオンビームドーピング装置などがある。加熱装置としては、抵抗加熱装置、ランプ加熱装置およびレーザー加熱装置などがある。また、処理装置500は、付加製造装置であってもよい。付加製造装置には、液槽光重合、材料押出、粉末床溶融結合、結合剤噴射、シート積層、材料噴射、指向性エネルギー堆積など、各方式の付加製造装置が含まれる。なお、処理装置500は、半導体処理装置や付加製造装置に限定されない。
処理結果出力用GUI 1300は、処理結果表示部1310と、完了/継続選択部1320、決定ボタン1330と、備える。
Claims (15)
- 1つ以上の説明変数および1つ以上の目的変数に関する標本データからなる学習データベースを解析し、制約なし二次形式関数または線形制約あり一次形式関数を導出する情報処理システムであって、
前記情報処理システムは、
前記学習データベースに対して機械学習により目的関数を導出する目的関数導出システムと、
前記目的関数を前記制約なし二次形式関数または前記制約あり一次形式関数に変換する関数変換システムと、を備え、
前記目的関数導出システムは、
機械学習の手法の詳細を設定する機械学習設定部と、
前記機械学習設定部で設定された機械学習の手法を用いて前記目的関数を導出する学習部と、を有し、
前記関数変換システムは、
0か1の値のみを成分とするベクトルであるダミー変数の生成方法を設定するダミー変数設定部と、
前記ダミー変数設定部で設定された生成方法に基づいて前記ダミー変数を生成するダミー変数生成部と、
前記目的関数に陽に現れる1つ以上の前記説明変数を前記ダミー変数を用いて消去することで、前記説明変数の二次より高次の非線形項を二次以下に次元を落とし、前記目的関数を前記ダミー変数および前記目的変数に関する前記制約なし二次形式関数または前記線形制約あり一次形式関数へ変換する関数変換部と、
を有することを特徴とする情報処理システム。 - 請求項1において、
前記機械学習設定部で設定される機械学習の手法はカーネル法であることを特徴とする情報処理システム。 - 請求項2において、
前記機械学習設定部は、
カーネル回帰、ベイズ最適化、カーネル回帰による多目的最適化、のいずれかを選択するカーネル手法選択部と、
カーネル関数の種類を選択するカーネル関数選択部と、を有し、
前記カーネル手法選択部で前記カーネル回帰が選択された場合には、前記目的関数はカーネル回帰の回帰関数であり、
前記カーネル手法選択部で前記ベイズ最適化が選択された場合には、前記目的関数はベイズ最適化の獲得関数であり、
前記カーネル手法選択部で前記カーネル回帰による多目的最適化が選択された場合には、前記目的関数はカーネル回帰の1つ以上の回帰関数の線形和で与えられる関数であることを特徴とする情報処理システム。 - 請求項3において、
前記説明変数は0か1の値のみを成分とするベクトルであることを特徴とする情報処理システム。 - 請求項4において、
前記ダミー変数設定部で設定される生成方法として、
前記カーネル関数選択部で選択されたカーネル関数の取り得る値に対してone-hotエンコーディングを施すことで、前記ダミー変数を生成する生成方法を含むことを特徴とする情報処理システム。 - 請求項4において、
前記カーネル手法選択部で、前記カーネル回帰、または、前記カーネル回帰による多目的最適化、が選択された場合に、
前記関数変換部は、
前記目的関数を前記ダミー変数に関する一次形式関数に変換し、
前記ダミー変数生成部で生成された前記ダミー変数に関する制約を課すことで、
前記線形制約あり一次形式関数を導出することを特徴とする情報処理システム。 - 請求項4において、
前記カーネル手法選択部で、前記ベイズ最適化が選択された場合に、
前記関数変換部は、
前記目的関数を前記ダミー変数に関する二次形式関数に変換し、
前記制約なし二次形式関数を導出することを特徴とする情報処理システム。 - 請求項4において、
前記カーネル手法選択部で、前記カーネル回帰、または、前記カーネル回帰による多目的最適化、が選択された場合に、
前記関数変換部は、
前記目的関数を前記ダミー変数に関する一次形式関数に変換し、
前記ダミー変数生成部で生成された前記ダミー変数に関する制約に対する二次形式のペナルティ項を付与することで、
前記制約なし二次形式関数を導出することを特徴とする情報処理システム。 - 請求項4において、
前記カーネル手法選択部で、前記カーネル回帰が選択された場合に、
前記カーネル関数選択部で選択されたカーネル関数を1つ以上の基底関数の足し合わせで近似することにより新たなカーネル関数とし、
前記学習部では、前記新たなカーネル関数を用いたカーネル回帰により前記目的関数を導出することを特徴とする情報処理システム。 - 請求項9において、
前記基底関数の足し合わせで近似する際に、
前記カーネル関数と前記基底関数の足し合わせとの誤差に関する最小二乗法または最小絶対値法を用いることを特徴とする情報処理システム。 - 請求項10において、
前記ダミー変数生成部は、
前記基底関数に対する双対変換における共役変数を、前記ダミー変数の一部として生成することを特徴とする情報処理システム。 - 請求項11において、
前記基底関数は、前記説明変数の一次形式または二次形式を入力とする、ReLU(Rectified Linear Unit)関数、絶対値を返す数値演算関数、指示関数、のいずれかであることを特徴とする情報処理システム。 - 請求項3に記載の情報処理システムと、
処理装置と、
前記処理装置の処理条件を出力する処理条件解析システムと、
前記処理条件のデータ、および、得られる処理結果のデータ、を加工して出力する学習データ生成部と、
を備え、前記処理装置の前記処理条件を決定する処理条件決定システムであって、
前記処理装置は、
前記処理条件解析システムから出力された前記処理条件を入力する処理条件入力部と、
前記処理条件入力部で入力された前記処理条件を用いて前記処理装置の処理を行う処理部と、
前記処理部の処理結果を取得する処理結果取得部と、を有し、
前記処理条件解析システムは、
前記情報処理システムによって導出された関数の種類に応じて前記説明変数の値の解析手法を選択する解析手法選択部と、
前記解析手法選択部で選択された解析手法を用いて、入力された関数の最小値または最大値を与える前記説明変数の値を算出する処理条件解析部と、
前記処理条件解析部で得られた前記説明変数の値を前記処理条件に加工して出力する処理条件出力部と、を有し、
前記処理条件入力部で入力された前記処理条件と、前記処理結果取得部で取得された処理結果は、前記学習データ生成部に入力され、
前記学習データ生成部は、前記処理条件入力部で入力された前記処理条件を前記説明変数のデータに加工し、前記処理結果取得部から出力された前記処理結果を前記目的変数のデータに加工して、前記学習データベースに格納し、
前記情報処理システムによって導出された関数は、前記解析手法選択部に入力され、
前記処理条件出力部によって出力された前記処理条件は、前記処理条件入力部に入力され、
所望の処理結果が得られるまで、前記情報処理システムによる関数導出、前記処理条件解析システムによる処理条件の出力、前記処理装置による処理、前記学習データ生成部による前記学習データベースへの前記説明変数のデータおよび前記目的変数のデータの格納、を繰り返すことを特徴とする処理条件決定システム。 - 請求項13において、
前記解析手法選択部では、
前記情報処理システムによって導出された関数が制約なし二次形式関数の場合はアニーリングが選択され、
前記情報処理システムによって導出された関数が制約あり一次形式関数の場合は整数計画法または線形計画法が選択されることを特徴とする処理条件決定システム。 - 請求項13において、
前記学習データ生成部では、
入力された前記処理条件のデータに対して二進数変換またはone-hotエンコーディングを施すことで、前記説明変数のデータを生成することを特徴とする処理条件決定システム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020237036991A KR20230162065A (ko) | 2021-05-18 | 2022-04-08 | 정보 처리 시스템 및 처리 조건 결정 시스템 |
CN202280029029.5A CN117178277A (zh) | 2021-05-18 | 2022-04-08 | 信息处理系统以及处理条件决定系统 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2021083895A JP2022177549A (ja) | 2021-05-18 | 2021-05-18 | 情報処理システムおよび処理条件決定システム |
JP2021-083895 | 2021-05-18 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022244547A1 true WO2022244547A1 (ja) | 2022-11-24 |
Family
ID=84141305
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2022/017395 WO2022244547A1 (ja) | 2021-05-18 | 2022-04-08 | 情報処理システムおよび処理条件決定システム |
Country Status (4)
Country | Link |
---|---|
JP (1) | JP2022177549A (ja) |
KR (1) | KR20230162065A (ja) |
CN (1) | CN117178277A (ja) |
WO (1) | WO2022244547A1 (ja) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017056367A1 (ja) * | 2015-09-30 | 2017-04-06 | 日本電気株式会社 | 情報処理システム、情報処理方法および情報処理用プログラム |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9727824B2 (en) | 2013-06-28 | 2017-08-08 | D-Wave Systems Inc. | Systems and methods for quantum processing of data |
-
2021
- 2021-05-18 JP JP2021083895A patent/JP2022177549A/ja active Pending
-
2022
- 2022-04-08 WO PCT/JP2022/017395 patent/WO2022244547A1/ja active Application Filing
- 2022-04-08 CN CN202280029029.5A patent/CN117178277A/zh active Pending
- 2022-04-08 KR KR1020237036991A patent/KR20230162065A/ko unknown
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017056367A1 (ja) * | 2015-09-30 | 2017-04-06 | 日本電気株式会社 | 情報処理システム、情報処理方法および情報処理用プログラム |
Non-Patent Citations (1)
Title |
---|
Yasuo Ozu, Performance Evaluation of New Dimensionality Reduction Method in Annealing, Proceedings of the 83rd Annual Conference (2021) Vol.1, 04 March 2021, pp. 1-289 to 1-290 * |
Also Published As
Publication number | Publication date |
---|---|
JP2022177549A (ja) | 2022-12-01 |
CN117178277A (zh) | 2023-12-05 |
KR20230162065A (ko) | 2023-11-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11250308B2 (en) | Apparatus and method for generating prediction model based on artificial neural network | |
Wang et al. | Adaptive MLS-HDMR metamodeling techniques for high dimensional problems | |
Miche et al. | OP-ELM: theory, experiments and a toolbox | |
JP7061536B2 (ja) | 最適化装置、シミュレーションシステム及び最適化方法 | |
Li et al. | A novel evolutionary algorithm for determining unified creep damage constitutive equations | |
CN113112013A (zh) | 针对分辨率降低的神经网络的优化量化 | |
US20210089832A1 (en) | Loss Function Optimization Using Taylor Series Expansion | |
JP2020052543A (ja) | 形状補完装置、形状補完学習装置、方法、及びプログラム | |
CN112578644B (zh) | 自适应全芯片光源优化方法及系统 | |
JP2017097803A (ja) | 予測装置、方法及びプログラム | |
Bartz-Beielstein et al. | High-performance simulation-based optimization | |
Zhao et al. | Entropy rate of nonequilibrium growing networks | |
Benala et al. | Genetic algorithm for optimizing functional link artificial neural network based software cost estimation | |
WO2022244547A1 (ja) | 情報処理システムおよび処理条件決定システム | |
KR102554791B1 (ko) | 데이터 세트로부터의 피쳐의 추출 | |
JP2021018677A (ja) | 情報処理システム、ニューラルネットワーク構造の生成方法および情報処理プログラム | |
Zhao et al. | Achieving real-time lidar 3d object detection on a mobile device | |
US20240362526A1 (en) | Information processing system and processing condition determination system | |
CN111191339A (zh) | 求解天线阵列综合问题的约束超多目标智能优化转换方法 | |
GIUSTOLISI et al. | A novel genetic programming strategy: evolutionary polynomial regression | |
JP2018190251A (ja) | Lle計算装置、lle計算方法及びlle計算プログラム | |
Halle et al. | Bayesian dropout approximation in deep learning neural networks: analysis of self-aligned quadruple patterning | |
JP2017207987A (ja) | 目的変数予測装置、方法及びプログラム | |
Li et al. | Parameter optimization of sub-35 nm contact-hole fabrication using particle swarm optimization approach | |
Shiely | Machine learning for compact lithographic process models |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22804461 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 20237036991 Country of ref document: KR Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1020237036991 Country of ref document: KR |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 22804461 Country of ref document: EP Kind code of ref document: A1 |