JP6754878B2 - Search device and search method - Google Patents

Description

The present invention relates to a search device and a search method for searching for a solution.

In order to improve the performance of semiconductor devices, new materials that make up semiconductor devices will be introduced, and at the same time, the structure of semiconductor devices will become complicated. Further, in the processing of semiconductor devices, nanometer-level accuracy is required. Further, in order to improve the productivity of the semiconductor device, it is required to continue mass production processing while maintaining the accuracy as much as possible. To meet these requirements, semiconductor devices need to be able to process a wide variety of materials and structures with extremely high precision. Therefore, the control range of the semiconductor processing apparatus that processes the semiconductor device is expanded, and a large number of control parameters are added. By using a semiconductor processing device, a high-performance semiconductor device is produced while maintaining high productivity.

On the other hand, in order to fully bring out the performance of the semiconductor processing apparatus, it is necessary to determine several to several tens of types of input parameters for each semiconductor processing apparatus. Furthermore, there are many steps in one process, and it is necessary to change the input parameter for each step. Therefore, it is extremely difficult to find out the combination of input parameters that can obtain the target machining result.

In order to maintain and improve productivity, it is necessary to acquire data on the state and processing results of semiconductor processing equipment that is mass-producing semiconductor devices. A plurality of sensors and monitors for the purpose of acquiring these data are mounted on the semiconductor processing apparatus. In order to perform control to correct data fluctuations in the state of semiconductor processing equipment and processing results during mass production of semiconductor devices, the relationship between sensor data and monitor data and processing results is analyzed, and control parameters are determined. Need to find out. In order to control processing at the nanometer level, the number of sensors mounted on semiconductor processing equipment and monitors of manufacturing status is increasing, and the frequency of data acquisition is also increasing. This increases the amount of data acquired. Therefore, it is extremely difficult to develop a control method for a high-precision semiconductor processing device, which requires analysis of a huge amount of data and verification of control performance.

The manufacture of advanced devices such as semiconductor devices requires the development of aging methods for semiconductor processing equipment in order to ensure productivity. The aging method of semiconductor processing equipment includes suppressing the performance difference between semiconductor processing equipment, correcting the change over time in processing characteristics during mass production, and correcting the performance difference between semiconductor processing equipment that cannot be corrected by maintenance of the semiconductor processing equipment. This is a method for shrinking. The development of aging methods for semiconductor processing equipment is carried out by top engineers with advanced knowledge and technology. However, the number of wafers processed and the number of difficult processes in the manufacture of semiconductor devices continues to increase, and the shortage of top engineers is becoming more serious. Therefore, the number of processes in which data is only acquired and analysis is not available increases. Therefore, the semiconductor processing apparatus is required to have a function of automatically drawing out the performance of the semiconductor processing apparatus by itself and a function of supporting an engineer who draws out the performance of the semiconductor processing apparatus.

Patent Document 1 discloses a technique of learning data when a recipe of a manufacturing tool is incrementally or randomly changed by using an autonomous learning system based on biology, and using the result to generate an adjusted recipe. To do.

Special Table 2013-518449

For example, the input parameters of a semiconductor processing device are parameters that determine the operation of the semiconductor processing device, such as gas type, gas flow rate, pressure, input power, voltage, current, processing time, heating temperature, cooling temperature, dose amount, and light amount. There are input parameters such as. The output parameters of the semiconductor processing device are parameters obtained by monitoring or measuring the processing object (processing result) being processed or processed by the semiconductor processing device, and are CD (Critical Dimensions), deposition film thickness, and ER. There are output parameters indicating the machining results such as (Etch Rate), machining shape, mask selection ratio, and in-plane distribution and uniformity of these wafers. Further, as sensor data and monitor data related to these processing results, light reflection spectrum, plasma light spectrum, wafer incident current, wafer voltage, wafer temperature, equipment component temperature, and their spatial distribution and uniformity are shown. There is also data. Sensor data and monitor data are also output parameters.

In order to analyze the input / output relationship of the semiconductor processing device, it is necessary to be able to analyze the input / output data from 1 input 1 output to multiple inputs and multiple outputs. Then, in order to obtain a combination of input parameters that satisfy the desired output result, it is necessary to search a vast device parameter space consisting of the input parameters and the output parameters.

For example, consider the case where five basic input parameters such as each flow rate of two types of gas used, gas pressure, discharge power, and wafer application bias power are selected as the input parameters to be searched. The control range of each input parameter is as follows. The control range of both gas flow rates is 100 to 200 [sccm], the control range of gas pressure is 1 to 10 [Pa], the control range of discharge power is 500 to 1500 [W], and the control range of bias power is 0 to 100 [ W] is a typical range. The typical value of the minimum width when changing each parameter is as follows. The value of both gas flow rates is 1 [sccm], the value of gas pressure is 0.1 [Pa], the value of discharge power is 1 [W], and the value of bias power is 1 [W].

In this case, the total combination of the control ranges of the input parameters in the entire device parameter space, that is, the number of search conditions is estimated to be 100 × 100 × 100 × 1000 × 100 = 10 ¹¹ . If the time required for one search is about 1 minute, it takes 100,000 years or more to search the total number of search conditions, which is impossible.

Also, if ten set number of values of each input parameter in the search one set, the combination of the input parameters in the search is 10 ^5. If the time for one search is about 1 minute, it takes more than 2 months to search for one set. In order to reach the target solution by repeating the search and the analysis of the search result, it is necessary that the time required for one set of search is several days or less, preferably one hour or less even if it is long. Therefore, in order to reach the target solution, it is extremely important to set the input parameters in the search, that is, to determine the search area.

When the number of search conditions in one set of search is 100 and the search time is 1 hour, and 2000 conditions are searched in one day by repeating this search, one day out of the number of conditions 10 ^{11 in} the device parameter space. 0.000002% of the region is searched for. If this is continued for one month, that is, when 60,000 sets of searches are executed, 0.00006% of the device parameter space is searched. Therefore, when the search area in one set of search is narrow, it is extremely unlikely that the optimum solution can be reached when the search area is randomly changed. Moreover, when there are overlapping search areas, the time required to reach the optimum solution becomes even longer.

Further, the input / output relationship of the semiconductor processing device is non-linear in most cases, and there are many local solutions in the device parameter space. Therefore, it is extremely rare to find the value of the input parameter that satisfies the value of the output parameter by one data analysis and estimation. Considering the case where about one local solution exists in 1% of the search area in the device parameter space, if the search area is a narrow area and the search area is randomly selected, the best solution in or near the search area is selected. Even if it can reach, there is a 99% chance that it will reach the local solution. Therefore, it is necessary to determine the search region so that the local solution can be avoided efficiently or the possibility that the solution can be reached after the local solution is reached is high.

However, the technique of Patent Document 1 described above has a problem that it is extremely unlikely that the best solution, which is an input parameter to be a solution, can be reached because the recipe is only incrementally or randomly changed during data learning. In other words, there is the problem that it is extremely likely that a local solution will be reached that will result in inferior results compared to the best solution.

An object of the present invention is to improve the efficiency of operation in a semiconductor processing apparatus.

The search device, which is one aspect of the invention disclosed in the present application, includes a condition set in the semiconductor processing device for processing the semiconductor or a target value indicating the result of processing the semiconductor by the semiconductor processing device, and the condition and the above. An input unit that accepts the input of the condition in the search area defined by the range of the result or the reference value of the result indicated by the target value, and the set value of the condition in the search area. Based on the actual measurement value of the result when the set value is given to the semiconductor processing apparatus, a generation unit that generates a prediction model showing the relationship between the condition and the result, and a generation unit generated by the generation unit. By giving the target value input to the input unit to the prediction model, the prediction value is acquired from the prediction model, and the region where the prediction value exists is specified from the search area, and the prediction value corresponds to the specific unit. A determination unit that determines whether or not the actual measurement value of the result is closer to the target value than the reference value input to the input unit, and the actual measurement value of the result corresponding to the predicted value by the determination unit. When it is determined that is close to the target value, the measured value of the result corresponding to the predicted value is set to the reference value, and the existing region of the predicted value specified by the specific unit is set to the search region. It is characterized in that it has a setting unit to be used, and an output unit that outputs a predicted value satisfying the achievement condition when the measured value of the result corresponding to the predicted value satisfies the achievement condition of the target value. ..

According to a typical embodiment of the present invention, it is possible to improve the efficiency of operation and optimize the processing in the semiconductor processing apparatus. Issues, configurations and effects other than those described above will be clarified by the description of the following examples.

FIG. 1 is an explanatory diagram showing an example of searching for input parameters. FIG. 2 is an explanatory diagram showing a system configuration example of a semiconductor manufacturing system. FIG. 3 is a block diagram showing a hardware configuration example of the search device. FIG. 4 is a block diagram showing a functional configuration example of the search device. FIG. 5 is a flowchart 1 showing an example of a control processing procedure of the semiconductor processing apparatus. FIG. 6 is a flowchart 2 showing an example of a control processing procedure of the semiconductor processing apparatus. FIG. 7 is a flowchart 3 showing an example of a control processing procedure of the semiconductor processing apparatus. FIG. 8 is a flowchart 4 showing an example of a control processing procedure of the semiconductor processing apparatus. FIG. 9 is a flowchart showing a method for suppressing machine error. FIG. 10 is a flowchart showing a time-dependent change correction method.

<Example of input parameter search>
FIG. 1 is an explanatory diagram showing an example of searching for input parameters. FIG. 1 shows input / output data input data and output for a semiconductor processing device when searching for input data (values of the above-mentioned input parameters) for obtaining output data (values of the above-mentioned output parameters) satisfying the target in the semiconductor processing device. An example of combination with data is shown.

Here, the semiconductor processing device is a device that processes a semiconductor or a semiconductor device including a semiconductor. Specifically, the semiconductor processing apparatus includes, for example, a lithography apparatus, a film forming apparatus, a pattern processing apparatus, an ion implantation apparatus, a heating apparatus, and a cleaning apparatus. The lithography apparatus includes an exposure apparatus, an electron beam drawing apparatus, and an X-ray drawing apparatus. The film forming apparatus includes, for example, CVD (Chemical Vapor Deposition), PVD (Physical Vapor Deposition), a vapor deposition apparatus, a sputtering apparatus, and a thermal oxidation apparatus. The pattern processing apparatus includes, for example, a wet etching apparatus, a dry etching apparatus, an electron beam processing apparatus, and a laser processing apparatus. The ion implantation device includes a plasma doping device and an ion beam doping device. The heating device includes, for example, a resistance heating device, a lamp heating device, and a laser heating device. The cleaning device includes a liquid cleaning device and an ultrasonic cleaning device.

FIG. 1 shows a graph 100 having a two-dimensional coordinate plane on two axes, the X1 axis and the X2 axis. The axis orthogonal to the coordinate plane is defined as the Y axis. The X1 axis is the coordinate axis of the input parameter X1 that is input to the semiconductor processing device, and the X2 axis is the coordinate axis of the input parameter X2 that is input to the semiconductor processing device. Input data x1n (n = 1, 2, 3, ...), Which is the value of the input parameter X1, is plotted on the X1 axis, and input data x2n, which is the value of the input parameter X2, is plotted on the X2 axis. In the graph 100, the output data of the Y axis in each region determined by the input data x1n and x2n is represented by contour lines. As an example, the region including the output data y1 is set to the maximum value (that is, the best solution), and the region of y13 is set to the minimum value. Further, in the graph 100, the X1 axis is divided into three regions, which are regions x1a, x1b, and x1c. Similarly, in the graph 100, the X2 axis is divided into three regions, which are regions x2a, x2b, and x2c.

For example, as a target, when searching for a region where the output data is the highest, that is, a combination of input data x1n and x2n from which the output data y1 is obtained, the inclination of the output data yn in the device parameter space is acquired as an analysis method. , The search may proceed in the direction in which the output data yn becomes larger.

However, if the output data at that time does not include the output data of the region determined by x1a and x2b, the search goes to the output data y4'of the vertex which is the local solution. Further, when the input parameter to be searched is determined according to this result, the output data in the vicinity of y4'is intensively acquired, and the input parameter for obtaining the output data in the vicinity of y4' itself or very close to it is discovered. That is, even if the analysis using the input parameter value acquired in this search and the acquisition of the output parameter value by further search are repeated, the input data of the best solution corresponding to the output data y1 cannot be found.

Further, when a large number of local solutions exist around the output data y1 or in other areas, if the acquired input parameter value is small with respect to the size of the device parameter space, the search falls into the local solution. There is a high possibility that the output data y1 cannot be found. Further, it is assumed that a region of x13 and x23 is estimated as a region in which the estimated solution exists. Even after that, if the search area is limited to a minute area such as x13 and a part of x23 as in Patent Document 1, the slope of the output data yn is very small, so the number of searches until the output data y1 is reached is enormous. There is a possibility that the search will fail. Further, the slope of the output data yn is buried in the noise included in the output data yn, so that the search may fail. If the quality of the output data yn to be analyzed is poor, the estimation of the solution will inevitably result in poor results. Therefore, in order for the search to be successful, a search for obtaining high-quality data that can approach the solution is required. You need to specify the space.

In FIG. 1, three parameters of x1n, x2n and yn are used. Since an actual semiconductor processing device has a large number of input / output parameters, a vast device parameter space in which the indexes of each axis in FIG. 1 are multidimensionally vectorized is searched. Therefore, it is necessary to specify the device parameter space in order to acquire high-quality data that can efficiently approach the solution from such a vast device parameter space. In addition, the input / output relationship of the semiconductor processing device is non-linear in most cases, and there are many local solutions in the device parameter space.

In this embodiment, the experimental conditions for searching the device parameter space are automatically determined in consideration of the analysis results of the input / output data of the semiconductor processing device, the experimental results are automatically verified, and these automatic operations are performed. repeat. As a result, the device parameter space is searched and the best solution is efficiently obtained. That is, the value of the input parameter that gives the value of the output parameter indicating the state of the target semiconductor processing apparatus and the processing result is efficiently discovered.

<System configuration example>
FIG. 2 is an explanatory diagram showing a system configuration example of a semiconductor manufacturing system. The semiconductor manufacturing system 200 includes a semiconductor processing device 201, a device control system 202, a monitor system 203, a sensor system 204, a database 205, and an automatic control system (search device) 206.

As described above, the semiconductor processing device 201 is a device that processes a substrate such as a wafer or a semiconductor device. The semiconductor processing device 201 is connected to the device control system 202, the monitor system 203, and the sensor system 204.

The device control system 202 is a system that controls the semiconductor processing device 201 when operating and processing the semiconductor processing device 201. The device control system 202 has an input interface such as a GUI, and controls the execution of the semiconductor processing device 201 by the value of the input parameter input via the input interface. In addition, the device control system 202 has a network interface, and acquires input parameter values from an external computer and a database 205 via the network interface.

The device control system 202 includes a recipe setting controller 221, a device basic setting controller 222, and a setting error detection system 223. The recipe setting controller 221 sets the input parameters and their values that determine the operation of the semiconductor processing device 201 during processing in the semiconductor processing device 201. The device basic setting controller 222 sets the input parameters and their values for operating the semiconductor processing device 201 in the semiconductor processing device 201.

The setting error detection system 223 determines whether or not the input parameters can actually be set in the semiconductor processing device 201 when the input parameters are set in the device basic setting controller 222. Specifically, for example, in the setting error detection system 223, is the input input parameter within the inputable range, or is it a combination of input parameter values that enables the operation of the semiconductor processing apparatus 201? To judge. When a value of an input parameter that cannot be set or a combination thereof is detected, the setting error detection system 223 reports as a setting error to an engineer or a higher-level system to which the semiconductor processing apparatus 201 is connected. When a setting error occurs, it is recorded as log data that the change of the input input parameter is stopped or the processing using the input parameter value is stopped.

The monitor system 203 is a system for acquiring monitor data by monitoring or measuring a processing object (processing result) being processed or processed by the semiconductor processing apparatus 201. The monitor system 203 includes an optical monitor, a processing dimension measuring device using an electron microscope, a temperature measuring device using infrared light, a defect detecting device using a Kelvin probe force microscope, and a prober for evaluating the electrical characteristics of the object to be processed. Includes equipment. The monitoring system 203 measures, for example, the reflection, transmission, absorption, and polarization spectra of light, laser light, and X-rays when they are incident on the object to be processed, thereby measuring the processed shape of the object to be processed and the thickness of the film to be processed. Acquires the light and processing defects as monitor data. The monitor system 203 does not need to be directly connected to the semiconductor processing device 201, and the measurement result may be acquired by transporting the processing target to the monitor system 203, and the result may be stored in the database 205.

The monitoring system 203 monitors the medium used for processing plasma, gas, liquid, etc. acting on the object to be processed at the time of processing, and the products generated by the processing. These media and products are those that act directly on or result from the action on the object to be treated. The monitor system 203 uses a plasma emission monitor using optical spectrum measurement, a deposit monitor in a processing room using infrared spectroscopic measurement, an atom and molecule monitor emitted from a processing target using a mass spectrometer, and a probe. Includes an electrical characteristic monitor in the processing room that was located. Monitoring by these monitors can measure monitor data that can indirectly evaluate the processing result in real time and on the spot during processing.

The sensor system 204 is a system that acquires sensor data indicating the device state of the semiconductor processing device 201. The sensor system 204 is an assembly of sensors. The sensor data includes power output values such as voltage, current, and power, values of variable electric elements such as capacitors and coils in the matching machine, flow rates of various gases used, temperatures of the equipment frame and equipment components, pressure in the processing chamber, and so on. Includes pressure control valve opening, valve open / closed state, gas exhaust rate, processing and device operating timing and operating time.

The database 205 stores various input parameter values set by the device control system 202, output parameter values which are processing results from the semiconductor processing device 201, monitor data acquired by the monitor system 203 and the sensor system 204, and sensor data. To do. The database 205 also stores the learning data. The learning data is input / output data that is a set of an input parameter value (input data) given to the semiconductor processing device 201 in the past and an output parameter value (output data) output from the semiconductor processing device 201. The database 205 may be a distributed database 205 that stores various types of data to be stored in different storage devices 302. A distributed database may be constructed in which the information handled by each system is stored in each system.

The automatic control system 206 searches for a solution that satisfies the target by using the data stored in the database 205. The solution that satisfies the target is the value of at least one of the input parameter value used for the operation of the semiconductor processing apparatus 201 and the input parameter value used for the operation of the semiconductor processing apparatus 201 during processing. is there. The automatic control system 206 includes a target setting controller 261, an autonomous search system 262, and an unstable motion detection system 263.

The target setting controller 261 accepts input of a target input parameter value, an output parameter value, and an allowable value of the difference or deviation between the search result and the target as initial values before the start of the search. Further, the target setting controller 261 includes an upper limit of the time for executing one condition in the search, the number of searches, an upper limit of the total time of one set of searches, an upper limit of the total time of the entire search, an upper limit of the analysis time of the search results, and analysis data. It can accept the input of the upper limit of the number. Further, the target setting controller 261 can set the searchability for each input parameter, the upper limit value and the lower limit value of the control range of the input parameter to be searched, and the value for limiting the control range of the input parameter to be searched. .. Further, the goal setting controller 261 is a model function that explains the best solution before the start of the search including the past results, the analysis target data used to obtain the solution, and the relationship between the goal and the input parameter obtained from the analysis. Input can be accepted.

The autonomous search system 262 acquires the contents input to the target setting controller 261 and sets a divided area in which the control range of the searchable input parameter is divided into two or more areas for one or more parameters. .. As described above, in order to reach the target solution by repeating the search and the analysis of the search result, the time required for one set of search needs to be several days or less, preferably one hour or less even if it is long. That is, assuming that the number of search conditions in one set of search is Ns, the search time in one set of search is Ts [min], and the time required for one search condition is t1 [min], the number of searches is given by the equation (1.1).

Ts = t1 · Ns ... (1.1)

When one set of search is within one day, Ts ≦ 1440, and when it is within one hour, the number of search conditions Ns may be determined so that Ts ≦ 60.

In order to increase the number of search conditions Ns, it is effective to shorten the measurement time of the sensor and monitor for evaluating the search result. In particular, it is effective to use sensors and monitors that can measure in real time during the search experiment.

Further, as described above, by measuring the characteristics of the medium acting on the processing object and the product generated by the processing, which are data that can indirectly evaluate the processing result, the characteristics of the product are measured in real time and during the processing. Can be measured in the field.

As the types of input parameters to be changed in the search increase, the search time Ts increases sharply. For example, when the number Da of the input parameter types is Da = 10 and the area division number A of each parameter is A = 3, the search condition number Ns when searching for all parameter combinations is given by the equation (1.2). ).

Ns = A ^Da ... (1.2)

The search condition number Ns increases to 59049. In such a case, it is preferable to predict the value of each input parameter for which a solution is predicted to exist, and limit the number of input parameters that can be changed at the same time during the search, using the predicted value as a central condition. As a result, the search condition number Ns and the search time Ts become feasible values. Past search results or engineer knowledge can be used to predict central conditions. Alternatively, it is also possible to start the search by giving an appropriate central condition as the initial value of the search.

For example, for five of the 10 kinds of input parameters, if the divided region is estimated to solution exists is specified, by changing the other five, the next search condition number Ns, Ns = 3 ⁵ It can be reduced to 243. Further, among the 10 types of input parameters, the number of types of input parameters that can be changed at one time is defined as Dc. The search condition number Ns can be obtained from Eq. (1.3) using the symbol C of combinatorics.

By limiting the input parameters that can be changed at one time, the number of search conditions Ns can be reduced. For example, if the number Dc of the types of input parameters that can be changed is Dc = 1, the search condition number Ns can be reduced to Ns = 21, and similarly, if Dc is Dc = 2, Ns can be reduced to Ns = 201. Further, among the types of input parameters, a method of specifying a division region in which a solution is presumed to exist and a method of specifying the types of parameters that can be changed at one time can be combined for some input parameters.

The unstable operation detection system 263 detects a case where the semiconductor processing device 201 can continue the processing operation at the time of executing the search, but the processing becomes unstable. Before executing the input of the input parameter, the setting error detection system 223 confirms whether the input parameter can be input. However, the media and components controlled by the equipment in the semiconductor processing apparatus 201 have non-linearity, and processing is executed by further combining them. Therefore, the setting error detection system 223 cannot detect the setting error (input parameter cannot be input), and there is a possibility that the input parameter whose operation becomes unstable is found only when the process is actually executed.

Further, as the number of input parameters increases, the device parameter space expands, so that if a local unstable operation region exists in the device parameter space, there is a high possibility that it cannot be detected in advance.

Therefore, when the unstable operation detection system 263 detects the unstable operation of the semiconductor processing device 201 during the processing of the semiconductor processing device 201, the semiconductor processing device 201 sets the input parameter and its value at the time of the unstable operation. Save and report to the engineer or the higher system to which the semiconductor processing apparatus 201 is connected. Thereby, it is possible to determine or predict the malfunction in the processing and the search due to the unstable operation of the semiconductor processing apparatus 201.

When unstable operation is detected, a sequence for restoring the semiconductor processing apparatus 201 to the steady operating state is executed after the processing is completed, or the processing is immediately stopped and the semiconductor processing apparatus 201 is restored to the steady operating state. The search is continued by executing the sequence for.

Such unstable operations include local abnormal discharges and vibrations of discharge strength during processing, sudden fluctuations in film forming speed and film quality, vibrations in gas pressure, and momentary fluctuations and vibrations in input power. .. Monitors capable of real-time and in-situ measurement during processing, such as the emission spectrum monitor, deposit monitor, mass spectrometry monitor, electrical characteristic monitor, and pressure monitor described above, detect unstable operation.

<Hardware configuration example of search device>
FIG. 3 is a block diagram showing a hardware configuration example of the search device 300. The search device 300 searches for the value of the input parameter to be the solution from the search area. The automatic control system 206 is an example of the search device 300. The search device 300 includes a processor 301, a storage device 302, an input device 303, an output device 304, and a communication interface (communication IF305). The processor 301, the storage device 302, the input device 303, the output device 304, and the communication IF 305 are connected by a bus. The processor 301 controls the search device 300. The storage device 302 serves as a work area for the processor 301. The storage device 302 is a non-temporary or temporary recording medium that stores various programs and data. Examples of the storage device 302 include a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), and a flash memory. The input device 303 inputs data. The input device 303 includes, for example, a keyboard, a mouse, a touch panel, a numeric keypad, and a scanner. The output device 304 outputs data. Examples of the output device 304 include a display and a printer. The communication IF305 connects to the network and transmits / receives data.

<Example of mechanical configuration of search device 300>
FIG. 4 is a block diagram showing a functional configuration example of the search device 300. The search device 300 includes an input unit 401, a generation unit 402, a specific unit 403, a determination unit 404, a setting unit 405, an output unit 406, a determination unit 407, a division unit 408, and a detection unit 409. Has. Specifically, each of the configurations 401 to 409 is a function realized by causing the processor 301 to execute a program stored in the storage device 302 shown in FIG. 3, for example.

The input unit 401 accepts input of various data by user operation or reading from the database 205. Specifically, for example, the input unit 401 receives an input of a condition value set in the semiconductor processing device 201 that processes the semiconductor device. The conditions set in the semiconductor processing apparatus 201 are the above-mentioned input parameters. Specifically, the input parameters include, for example, gas type, gas flow rate, pressure, input power, voltage, current, processing time, heating temperature, cooling temperature, dose amount, and light amount.

Further, the input unit 401 receives an input of a target value indicating the result of processing the semiconductor by the semiconductor processing device 201. The result of processing the semiconductor by the semiconductor processing apparatus 201 is the output parameter described above. Specifically, the output parameters include, for example, the processing result of the semiconductor by the semiconductor processing apparatus 201 and the data regarding the apparatus state of the semiconductor processing apparatus 201. Specifically, the processing results of the semiconductor by the semiconductor processing apparatus 201 include, for example, CD (Critical Dimensions), deposition film thickness, ER (Etching Rate), processing shape, mask selection ratio, and in-plane distribution and uniformity of these. Including sex. Specifically, the processing result of the semiconductor processing apparatus 201 and the data regarding the apparatus state of the semiconductor processing apparatus 201 include, for example, a light reflection spectrum, a plasma light spectrum, a wafer incident current, a wafer voltage, a wafer temperature, an apparatus component temperature, and further. Includes data (sensor data and monitor data) showing these spatial distributions and uniformity. The target value indicating the result of processing the semiconductor by the semiconductor processing device 201 is the value of the output parameter of the semiconductor processing device 201 requested by the user.

Further, the input unit 401 accepts the input of the reference value of the condition in the search area defined by the range of the condition (input parameter) and the result (output parameter). The search area is defined by the control range of the input parameter and the control range of the output parameter of the semiconductor processing apparatus 201, and is an area for searching the value of the input parameter. Specifically, for example, it is the search area A shown in FIG. The reference value of the condition is the reference value of the input parameter. Specifically, for example, it is a value of an input parameter obtained in the past.

Further, the input unit 401 accepts the input of the reference value of the result in the search area. The reference value of the result is the value of the output parameter of the semiconductor processing apparatus 201 when the reference value of the input parameter is given to the semiconductor processing apparatus 201.

The generation unit 402 generates a prediction model showing the relationship between the condition and the result based on the set value of the condition in the search region and the actually measured value of the result when the set value is given to the semiconductor processing apparatus 201. To do. The set value of the condition is, for example, the value of the input parameter prepared as the training data. The measured value of the result is the value of the output parameter of the semiconductor processing device 201 when the value of the input parameter (setting value of the condition) prepared as the learning data is given to the semiconductor processing device 201. The prediction model is a function that shows the relationship between input parameters and output parameters. The generation unit 402 sets the conditions in the search area by performing regression analysis such as neural network, support vector machine, etc. that can handle multiple inputs and multiple outputs, and statistical analysis such as correlation analysis, principal component analysis, and multiple regression analysis. Generate a prediction model that shows the relationship with the measured value of the output.

The specific unit 403 obtains the predicted value corresponding to the target value from the predicted model by giving the target value input by the input unit 401 to the predicted model generated by the generated unit 402, and determines the region where the predicted value exists. Identify from the search area. Further, when the output parameter of the region where the predicted value exists has not been acquired, the generation unit 402 sets the measured value of the output when the set value of the condition in the divided region is given to the semiconductor processing apparatus 201 for each divided region. get.

When the target value is the value of the output parameter of the semiconductor processing apparatus 201, the specific unit 403 gives the value of the output parameter to the prediction model, and thereby obtains the value of the input parameter from the prediction model as the prediction value corresponding to the target value. get. Then, the identification unit 403 specifies the region where the input parameter value, which is the predicted value, exists from the search region. Specifically, for example, in FIG. 1, when the target value is the output parameter value y12, the predicted value corresponding to the target value y12 is the input parameter X1 specified by the contour line of the target value y12 in FIG. It is a value of X2. Therefore, the identification unit 403 specifies the existing regions A1, A2, A3, A8, and A9 of the values of the input parameters X1 and X2 specified by the contour lines of the target value y12 from the search region A.

The determination unit 404 determines whether or not the target value corresponding to the predicted value is closer to the target value than the reference value of the result input by the input unit 401. Specifically, for example, in the search area A, the determination unit 404 sets the distance between the target value corresponding to the predicted value and the target value (first distance) and the reference value of the result and the target value. Find the distance (second distance). The distance is, for example, the Euclidean distance. When the first distance is shorter than the second distance, the determination unit 404 determines that the target value corresponding to the predicted value is closer to the target value than the reference value of the result. If the first distance is not shorter than the second distance, the determination unit 404 determines that the target value corresponding to the predicted value is not closer to the target value than the reference value of the result.

When the determination unit 404 determines that the target value corresponding to the predicted value is closer to the target value than the reference value of the result, the setting unit 405 uses the predicted value and the target value corresponding to the predicted value as the reference of the condition. The value and the reference value of the result are set, and the existing area of the predicted value specified by the specific unit 403 is set in the search area. As a result, the reference value approaches the target value, and the search area is narrowed down to the area where the predicted value exists.

The output unit 406 outputs a predicted value that satisfies the achievement condition when the predicted value satisfies the achievement condition of the target value. The achievement condition is, for example, an allowable range of the target value. The output unit 406 may display a predicted value satisfying the achievement condition on a display which is an example of the output device 304, may transmit the predicted value to an external device via the communication IF 305, and store the predicted value in the storage device 302 or the database 205. You may.

When the determination unit 407 determines that the target value corresponding to the predicted value is not closer to the target value than the reference value of the result, the determination unit 407 excludes the predicted value and the target value corresponding to the predicted value. (Described later in step S601 of FIG. 6). Exclusion data is the value of an input parameter that should not be given to the prediction model.

Further, in this case, the setting unit 405 sets the data excluding the exclusion data determined by the determination unit 407 as the learning data. As a result, the generation unit 402 can use the learning data in which the exclusion data does not exist. Therefore, it is possible to improve the search speed of the value of the input parameter which is the solution.

Further, when the determination unit 407 determines that the target value corresponding to the predicted value is not closer to the target value than the reference value of the result, the determination unit 407 determines the existing area of the predicted value as the exclusion area. It may be (described later in step S602 of FIG. 6). The exclusion area is an area of input parameter values that the prediction model should not output.

Further, in this case, the setting unit 405 sets the remaining area excluding the exclusion area determined by the determination unit 407 from the search area as the search area. As a result, the search area can be narrowed down by excluding the range of predicted values that can obtain results that are not close to the target value. Therefore, it is possible to improve the search speed of the value of the input parameter which is the solution.

The division unit 408 divides the search area into a plurality of areas. Specifically, for example, the division unit 408 divides the search area A into nine areas (division areas) A1 to A9, as shown in FIG. In this case, the generation unit 402 acquires the actually measured value of the output when the set value of the condition in the divided region is given to the semiconductor processing apparatus 201 for each divided region. Then, the generation unit 402 generates a prediction model based on the set value of the condition in each division region and the measured value of the output. By using the plurality of semiconductor processing devices 201, the measured values can be acquired in parallel for each division region, and the generation speed of the prediction model can be improved.

Further, when the determination unit 404 determines that the target value corresponding to the predicted value is not closer to the target value than the reference value of the result, the setting unit 405 sets the predicted value and the target value corresponding to the predicted value. , The existing region of the predicted value may be set in the search region without being set in the reference value of the condition and the reference value of the result (described later in steps S604 to S606 of FIG. 6). As a result, by further subdividing the search area, it is possible to execute a more detailed search for the solution and confirm that the solution is not overlooked.

Further, as described above, when the determination unit 404 determines that the target value corresponding to the predicted value is not closer to the target value than the reference value of the result, the determination unit 407 determines the predicted value and the predicted value. The corresponding target value may be determined as exclusion data. In this case, the generation unit 402 creates a prediction model based on the measured value of the residual obtained by removing the exclusion data from the measured value and the set value of the residual obtained by removing the set value corresponding to the measured value of the residual from the set value. It may be generated (described later in FIG. 7). As a result, it is possible to improve the search speed of the value of the input parameter which is the solution.

Further, as described above, when the determination unit 404 determines that the target value corresponding to the predicted value is not closer to the target value than the reference value of the result, the determination unit 407 excludes the existing region of the predicted value. The area may be determined. In this case, the generation unit 402 is based on the set value of the condition in the remaining search area excluding the exclusion area from the search area and the actually measured value of the output when the set value is given to the semiconductor processing apparatus 201. , A prediction model may be generated (described later in FIG. 8). As a result, it is possible to improve the search speed of the value of the input parameter which is the solution.

The detection unit 409 detects the unstable operation of the semiconductor processing device 201 based on the output of the semiconductor processing device 201 and a predetermined output threshold value. The detection unit 409 corresponds to the unstable operation detection system 263. In this case, the output unit 406 outputs the detection result by the detection unit 409.

<Example of control processing procedure of semiconductor processing device 201>
FIG. 5 is a flowchart 1 showing an example of a control processing procedure of the semiconductor processing apparatus 201. The search device 300 receives the input of the output value (target value of the output parameter) from the target semiconductor processing device 201 and the search setting (step S501). The search settings are, for example, the permissible value of the difference or deviation between the search result and the target value, the upper limit of the time for executing one condition in the search, the number of searches, the upper limit of the total time of one set of searches, and the total time of the entire search. Upper limit of search result analysis time, upper limit of number of analysis data, threshold of acquisition frequency of output data from semiconductor processing device 201, acquisition time of output data (value of output parameter) from semiconductor processing device 201 Is the upper limit of, and the lower limit of the number of output data from the semiconductor processing apparatus 201.

Next, the search device 300 accepts the input of the base solution and the input of information about the solution (step S502). Specifically, for example, the search device 300 uses the input parameters that have been practically used in the past, the output parameters when the input parameters are used, the best solution (value of the input parameter) before the start of the search, and the best solution. It accepts the input of the output parameter, the target value of the output parameter before the start of the search, and the input of the model function explaining the relationship between the input parameter and the output parameter.

Next, the search device 300 saves the best solution input in step S502 as the best solution OS1 (step S503). If there is no solution, a symbol or value is set to indicate that it is farthest from the solution.

Next, the search device 300 sets the basic search area as the search area (step S504). Specifically, for example, the search device 300 sets searchability for each input parameter, upper and lower limits of the control range of the input parameter to be searched, and a value for limiting the control range of the input parameter to be searched (for example). , Upper limit value and lower limit value) are set. Further, the search device 300 determines the search area by referring to the control range of the input parameter determined by the setting error detection system 223 as a setting error. Further, the search device 300 determines the search area with reference to the parameters detected by the unstable motion detection system 263. Further, the unstable operation detection system 263 holds a combination of input parameters or a value of each input parameter range in which the semiconductor processing apparatus 201 is inoperable or unstable in the past, and the search apparatus 300 uses this combination. Can determine the search area.

For example, in FIG. 1, when two input parameters X1 and X2 are selected as the input parameter types, the control range of the input parameter X1 is [x11, x1n] and the control range of the input parameter X2 is [x21, x2n]. Then, the area A of the entire range shown in FIG. 1 is the search area.

The input and setting contents of the target setting controller 261 in steps S501 to S504 are passed to the autonomous search system 262, and the automatic search is executed by the procedure of steps S505 to S510 described below.

The search device 300 divides the search area (step S505). Specifically, for example, the search device 300 divides the control range of the searchable input parameter into two or more regions for one or more input parameters. The divided area is called a divided area. When the number of search conditions increases and it is predicted that the search cannot be completed within the desired time, a method of specifying a division region in which a solution is presumed to exist in some input parameters among the types of input parameters, And by using the method of specifying the types of input parameters that can be changed at once, the search area can be limited and the number of search conditions can be reduced. Further, by combining these two methods, the search area can be limited and the number of search conditions can be reduced.

For example, in FIG. 1, when two input parameters X1 and X2 are selected as the input parameter types, the control range [x11, x1n] of the input parameter X1 is divided into x1a, x1b, x1c, and the input parameter X2 Assuming that the control range [x21, x2n] is divided into x2a, x2b, and x2c, the nine division regions A1 to A9 shown in FIG. 1 can be obtained.

The search device 300 executes an autonomous search for each divided region (step S506). Specifically, for example, the search device 300 acquires the input / output data of the semiconductor processing device 201 under each search condition as a search result by autonomous search using the divided region and the search conditions. The input / output data is a set of an input parameter value (input data) given to the semiconductor processing device 201 and an output parameter value (output data) obtained from the semiconductor processing device 201.

Specifically, for example, the search device 300 selects the value of the input parameter satisfying the search condition for each division region, and gives the value of the selected input parameter to the semiconductor processing device 201. Then, the search device 300 acquires the output data (value of the output parameter) from the semiconductor processing device 201. The combination of the value of this input parameter and the value of the output parameter corresponding to the value is the search result.

Further, in this case, the unstable operation detection system 263 detects a case where the semiconductor processing device 201 can continue the processing operation at the time of executing the autonomous search, but the processing of the semiconductor processing device 201 becomes unstable. When unstable operation is detected, a sequence for restoring the semiconductor processing device 201 to the steady operation state is executed after the processing is completed, or the processing is immediately stopped and the semiconductor processing device 201 is restored to the steady operation state. By executing the sequence of, the autonomous search by the target setting controller 261 is continued.

The search device 300 stores the search result for each divided area in the database 205 (step S507). Specifically, for example, the search device 300 uses the value of the input parameter used in the autonomous search (step S506) and the output of the semiconductor processing device 201 acquired by using the value of the input parameter for each divided region. The input / output data, which is a set of the parameter values, is stored in the database 205 as the search result for each divided area.

The search device 300 generates a prediction model for predicting a solution (input parameter) that satisfies the target (target output) (step S508). Specifically, for example, the search device 300 uses the learning data stored in the database 205 in step S507 to generate a function indicating the relationship between the input / output data of the semiconductor processing device 201 as a prediction model. As a method for analyzing the relationship between input / output data, regression analysis capable of responding to multiple inputs and multiple outputs such as neural network, support vector regression, and regression using the kernel method can be used. In addition, statistical analysis such as correlation analysis, principal component analysis, and multiple regression analysis can be used.

In the generation of the prediction model, for example, sensor data and monitor data for acquiring indirect measured values for the processing result of the semiconductor processing apparatus 201 are used as output data. The acquisition frequency of output data is lower than the frequency specified in the search settings, or longer than the acquisition time specified in the search settings, and the number of output data that can be acquired by the search is less than the number of output data specified in the search settings. In some cases. In this case, it is sufficient to acquire the sensor data and the monitor data that can acquire a large number of data as compared with the number of acquired output data. This makes it possible to analyze the relationship between the sensor data and the monitor data with respect to the output data and the relationship between the input data with respect to the sensor data and the monitor data. Further, by using both of these analysis results, the relationship between the input data and the output data can be obtained.

The search device 300 estimates the divided region in which the solution exists (step S509). When the output data in the region estimated to have a solution has been acquired (step S5090A: Yes), the search device 300 identifies the best solution OS2 from the estimated divided regions, and the identified best solution OS2 is stored in the database 205. Save to (step S510). Further, when the output data in the region where the solution is estimated to exist has not been acquired (step S5090A: No), the search device 300 executes the autonomous search in the estimated region (step 5060A), and the solution exists. The output data in the estimated area, that is, the search result is acquired and stored in the database 205 (step 5070A). These autonomous searches and storage of results are the same processes as in steps 506 and 507. There are two methods for processing in step S509. One is a method (first estimation method) of giving the target value of the output parameter given in step S501 to the prediction model.

In the first estimation method, specifically, for example, the search device 300 substitutes the target value of the output parameter given in step S501 into the prediction model obtained in step S508 to target the output parameter. Estimate the input data (input parameter value) that is the solution that satisfies the value. Then, the search device 300 identifies the divided region in which the input data to be the solution exists. For example, when the prediction model is a function showing the input / output relationship of FIG. 1, when y6 is given as the target value of the output parameter, among the division areas A1 to A9, the division areas A1 and A4, where the solution exists, Three divided regions of A7 are estimated. If the output data has not been acquired in each of the divided regions A1, A4, and A7, steps 5060A and 5070A are executed as described above.

In step S510, the search device 300 has a division area in which output data equal to the target value y6 exists from the acquired output data, and the difference or deviation between the output data and the target value is from the permissible value (permissible value given in S501). The divided area in which the small output data exists or the divided area in which the output data closest to the target value y6 exists is specified, and this divided area is determined as the area in which the best solution OS2 exists (hereinafter, a specific divided area). When a plurality of divided regions that can be specific divided regions are specified, the search device 300 determines all of them as the divided regions in which the best solution OS2 exists. Alternatively, the search device 300 selects the divided region in which the number of output data for which the difference or deviation between the output data and the target value is smaller than the allowable value (allowable value given in S501) is the largest among the plurality of divided regions. , Determine in the divided area where the best solution OS2 exists. In the case of the above example, if the acquired output data is similar to FIG. 1 shown by the prediction model, all of the divided regions A1, A4, A7, or the divided regions A1, A4, A7 in which a value equal to or close to y6 is obtained. Area A4 is determined to be a specific divided area. The search device 300 has the output data equal to the target value of the output parameter in a specific divided region, the difference or deviation between the output data and the target value is smaller than the permissible value (the permissible value given in S501), or the output data closest to the target value. The value of the input parameter when is obtained is determined as the best solution OS2.

The process of another step S509 can be applied when the input parameter that is the solution satisfying the target value of the output parameter cannot be directly obtained by substituting the target value of the output parameter given in step S501 into the prediction model. This is the method (second estimation method). The second estimation method is to give a set of input parameters to the prediction model at a time, give a large number of sets of input parameters to the prediction model, calculate the estimated output parameters, and obtain the result closest to the target output. This is a method to get a set of parameters. For example, the search device 300 can create one or more sets of input parameters included in each divided area for each divided area, give them to the prediction model, and calculate the output parameters at that time. As a method of obtaining a typical estimated output for each divided area, a value that becomes the center coordinates in the divided area can be used as a set of input parameters included in the divided area.

When calculating the estimated output by giving an extremely large number of sets of input parameters to the prediction model, if the calculation time becomes enormous, as described using the above equations (1.1) to (1.3). The search device 300 determines the central condition of the input parameter given to the prediction model, and limits the types of input parameters that can be changed from the central condition, so that the calculation time can be suppressed. In addition, the calculation time can be suppressed by limiting the number of parameters that can be changed at one time from the central condition. As a result, the search device 300 can obtain an input parameter that gives a result closer to the target output while reducing the number of sets of input parameters given to the prediction model. Past search results or engineer knowledge can be used when setting the central condition. Alternatively, the center coordinates of the entire search area can be used as the center condition.

In the second estimation method, specifically, the search device 300 acquires the value of the output parameter to be the predicted value by substituting the value of the set of input parameters into the prediction model obtained in step S508. .. For example, when the prediction model is a function showing the input / output relationship of FIG. 1, the input parameter values to the prediction model are (x11, x21), (x11, x22), (x11, x23), (x12, x21). , (X12, x22), (x12, x23), (x13, x21), (x13, x22), (x13, x23), (x1n, x21), (x1n, x22), (x1n, x23) Then, the estimated output parameters corresponding to each input parameter can be obtained.

When the output parameters corresponding to each of these input parameters have not been acquired (step S5090A: Yes), the search device 300 executes an autonomous search using each input parameter (step 5060A), and if a solution exists. The output data in the estimated area, that is, the search result is acquired and stored in the database 205 (step 5070A). These autonomous searches and storage of results are the same processes as in steps 506 and 507. If the output parameters corresponding to each input parameter have already been acquired (step S5090A: No), the process proceeds to step S510.

Then, in step S510, the search device 300 determines, for example, the value of the input parameter obtained by obtaining the value of the output parameter which is the closest predicted value from the target value of the output parameter in the divided region as the best solution OS2. For example, if the acquired output data is the same as in FIG. 1 shown by the prediction model, and if the target value of the output parameter is y10, the value of the output parameter corresponding to the value of the input parameter (x13, x23). y10 is the closest value. Therefore, the best solution OS2 is (x13, x23). Which of the first estimation method and the second estimation method is applied is set in advance.

The search device 300 determines whether or not the value of the best solution OS2 is a solution that can obtain an output parameter closer to the target than the value of the best solution OS1 (step S511). Then, if the value of the best solution OS2 is a solution that can obtain an output parameter closer to the target than the value of the best solution OS1 (step S511: Yes), the process proceeds to step S512, and if not (step S511: No), the step Transition to S601 (FIG. 6).

In the case of step S511: Yes, the search device 300 updates the best solution OS1 by setting the best solution OS2 to the best solution OS1 (step S512). Further, the search device 300 updates the search area by setting the divided area of the updated best solution OS1 in the search area (step S514).

After that, the search device 300 determines whether or not the target has been achieved (step S514). When the target is achieved (step S514: Yes), the search device 300 ends the control process. On the other hand, when the target is not achieved (step S514: No), the process proceeds to step S505 or step S601 (FIG. 6). Whether to move to step S505 or step S601 (FIG. 6) may be preset, or may be selected by the user each time. If the target has not been achieved (step S514: No) and the process proceeds to step S505, the search device 300 divides the search area after the update of step S513 (step S505).

In step S514, specifically, for example, when the output parameter corresponding to the updated best solution OS1 is equal to the target value or the difference from the target value is within the allowable range, the search device 300 determines that the target has been achieved. (Step S514: Yes). Further, even if the output parameter corresponding to the updated best solution OS1 is equal to the target value or the difference from the target value is not within the permissible range, if the search time set in step S501 elapses, the target is achieved. Judgment (step S514: Yes). On the other hand, if the output parameter corresponding to the updated best solution OS1 is equal to the target value or the difference from the target value is not within the permissible range and the search time set in step S501 has not elapsed, the target has not been achieved. (Step S514: No).

FIG. 6 is a flowchart 2 showing an example of a control processing procedure of the semiconductor processing apparatus 201. The search device 300 determines the search data in the divided region obtained in step S509 as exclusion data (step S601). Specifically, for example, the search device 300 determines the input data (value of the input parameter) in the estimation area and the output data (output parameter) as the exclusion data. The excluded data is data that may be excluded in future processing (exclusion from the data to be actually analyzed is carried out in S703 of FIG. 7 or S608 of FIGS. 6 to 8). Similarly, the search device 300 determines the divided region obtained in step S510 as the exclusion region (step S602). The exclusion area is an area that may be excluded in future processing (exclusion from the data to be actually analyzed is carried out in S803 of FIG. 8 or S608 of FIGS. 6 to 8).

After that, the search device 300 determines whether or not to divide the divided region obtained in step S509 (step S603). When dividing (step S603: Yes), the process proceeds to step S604, and when not dividing (step S603: No), the process proceeds to step S608.

Specifically, in step S603, for example, the search device 300 determines whether or not to divide the division area depending on the presence or absence of the division instruction input from the user. Further, the search device 300 may forcibly divide the division area (step S603: Yes).

After that, the search device 300 generates a prediction model as in step S508 (step S604). Next, the search device 300 estimates the divided region in which the solution exists from the divided region group divided in step S603, as in step S509 (step S605). The estimated division area is referred to as an estimated division area. Then, the search device 300 executes the same processing as in steps S5090A, S5060A, and S5070A in steps S6090A, S6060A, and S6070A. Specifically, for example, when the output data in the region where the solution is estimated to exist has already been acquired (step S6090: Yes), the search device 300 identifies the best solution OS3 from the estimated divided regions. , The identified best solution OS3 is stored in the database 205 (step S606). Similarly, when the output data in the region where the solution is estimated to exist has not been acquired (step S6090: No), the search device 300 executes an autonomous search in the estimated region (step S6060A). The output data in the region where the solution is estimated to exist, that is, the search result is acquired and stored in the database 205 (step S6070A).

After that, as in step S511, the search device 300 determines whether or not the value of the output parameter corresponding to the best solution OS3 is closer to the target than the value of the output parameter corresponding to the best solution OS1 (step S607). ). Then, when the value of the output parameter corresponding to the best solution OS3 is closer to the target than the value of the output parameter corresponding to the best solution OS1 (step S607: Yes), the process proceeds to step S512, and when not (step S607). : No), the process proceeds to step S608. In step S608, the search device 300 updates the search area by excluding the exclusion area from the search area or excluding the exclusion and exclusion data of the exclusion area (step S608), and proceeds to step S505.

FIG. 7 is a flowchart 3 showing an example of a control processing procedure of the semiconductor processing apparatus 201. The flowchart 3 is another processing example of the flowchart 2 shown in FIG. For the same processing as in FIG. 6, the same step numbers are assigned and the description thereof will be omitted. After step S601, the search device 300 determines whether or not to exclude the exclusion data determined in step S601 (step S703). When the analysis is performed with exclusion (step S703: Yes), the process proceeds to step S604, and when the analysis is performed without exclusion (step S703: No), the process proceeds to step S608.

Specifically, in step S703, for example, the search device 300 determines whether or not to exclude the excluded data for analysis depending on the presence or absence of the exclusion instruction input from the user. Further, the search device 300 may forcibly exclude the excluded data for analysis (step S703: Yes).

After that, the search device 300 generates a prediction model of the divided region (step S604) without using the exclusion data, and executes steps S605, S609A, S6060A, S6070A, S606, S607, and S608.

FIG. 8 is a flowchart 4 showing an example of a control processing procedure of the semiconductor processing apparatus 201. The flowchart 4 is another processing example of the flowchart 2 shown in FIG. For the same processing as in FIG. 6, the same step numbers are assigned and the description thereof will be omitted. After step S602, the search device 300 determines whether or not to perform analysis by excluding the exclusion region determined in step S602 (step S803). When the analysis is performed with exclusion (step S803: Yes), the process proceeds to step S604, and when the analysis is performed without exclusion (step S803: No), the process proceeds to step S608.

Specifically, in step S803, for example, the search device 300 determines whether or not to perform analysis by excluding the solution (value of the input parameter) in the exclusion region depending on the presence or absence of the exclusion instruction input from the user. Further, the search device 300 may forcibly exclude the exclusion region for analysis (step S803: Yes).

After that, the search device 300 generates a prediction model of the divided region as in step S604 (step S604). Next, the search device 300 estimates the divided region in which the solution exists from the divided region group divided in step S603 without using the solution (value of the input parameter) in the excluded region (step S605). .. The estimated division area is referred to as an estimated division area. Then, the search device 300 executes steps S6090A, S6060A, S6070A, S606, S607, and S608.

In this control process, if the scale of data analysis, storage, transfer, etc. becomes large and the execution time is longer than the time to search for one condition, the search should be continued in parallel with those executions. Is possible. At that time, one or more of the number of input parameters to be changed in the search condition, the number of input parameters to be changed at the same time, and the number of divisions of the search area are increased. As a result, the number of search conditions is increased, and by searching for these conditions, the search results can be increased by utilizing the time for executing analysis or the like. In particular, the time required for data analysis may be several minutes to several hours or more, and the search speed can be improved by continuing the search during the analysis.

<Application Example 1 of Control Processing of Semiconductor Processing Device 201>
Next, an application example of the control process for correcting the device difference of the semiconductor processing device 201 in the maintenance of the semiconductor processing device 201 before the mass production of the semiconductor device will be shown. Here, as an example for explaining the procedure for suppressing the machine difference, the semiconductor processing apparatus 201 is used as an etching apparatus for performing discharge processing. Further, here, the input parameter when executing the discharge process is referred to as a recipe. In the etching apparatus that performs the discharge treatment, the output to be corrected includes the treatment result or the characteristics of the discharge used in the treatment. As a correction method, there is a method of correcting other etching devices so that the same output as the reference etching device can be obtained, or a method of correcting so that the outputs of a plurality of etching devices are uniform.

FIG. 9 is a flowchart showing a method for suppressing machine error. Specifically, FIG. 9 shows, for example, the search device 300 starts acquiring discharge data after maintenance in order to correct a device difference showing an example of a maintenance processing procedure of the semiconductor processing device 201 before mass production of the semiconductor device. ..

First, the search device 300 performs a reference recipe discharge using a recipe for performing a basic discharge process, and acquires input / output data at that time. In addition, discharge is performed according to the recipe used in mass production, and output data (value of output parameter) at that time is acquired (step S901). Step S901 is a process corresponding to steps S501 to S504.

Next, the search device 300 searches for the device difference correction recipe (step S902). Step S902 is a process corresponding to steps S505 to S507. Then, the search device 300 executes the device difference correction using the device difference correction recipe searched in step S902 (step S903). Step S903 is a process corresponding to the processes of steps S508 to S513 and FIGS. 6 to 8. If the correction target has not been achieved (step S904: No), the process returns to step S902, and if the correction target has been achieved (step S904: Yes), the process ends. Step S904 is a process corresponding to step S514.

The search speed may be improved by proceeding with the search in parallel using a plurality of the same etching devices. In that case, by using a plurality of etching devices in which the device difference is corrected by the procedure of FIG. 9, the possibility of searching for a solution satisfying the target can be increased. Furthermore, it becomes possible to expand the searched solution to those plurality of devices and perform correction.

In this way, by executing the search method by the search device 300 after the maintenance of the semiconductor processing device 201, the value of the output parameter of the semiconductor processing device 201 can be brought close to the value of the reference output parameter (automatic machine error correction). function).

<Application example 2 of control processing of semiconductor processing apparatus 201>
Next, in the mass production process of the semiconductor device, an application example of the control process for correcting the change with time will be shown.

FIG. 10 is a flowchart showing a time-dependent change correction method. In FIG. 10, as in FIG. 9, as an example for explaining the procedure for correcting the change with time, the semiconductor processing apparatus 201 is an etching apparatus that uses an electric discharge for processing. In the pre-mass production discharge data acquisition, in order to correct the change over time during mass production, the search device 300 first gives the etching device a reference recipe for performing basic discharge processing, and performs the reference recipe discharge. The output data (value of the output parameter) of is acquired (step S1011). Step S1011 is a process corresponding to steps S501 to S504.

Next, the search device 300 searches for a device difference correction candidate recipe (step S1012). In step S1012, output data or sensor data and monitor data that can predict the time-dependent change of the correction target from the analysis results of the input / output data at the time of mass production in the past are used. Then, the search device 300 searches the etching device before the start of mass production for a device difference correction candidate recipe for changing the data that can predict the change with time to the value after the change with time. Step S1012 is a process corresponding to steps S505 to S507.

Next, the search device 300 executes device difference correction (step S1013). In step S1013, the search device 300 compares the basic mass production recipe used in a state where no change with time has occurred with the device difference correction candidate recipe searched in step S1012, and changes the device difference correction candidate recipe. Clarify the input parameters that have been made. As a result, it is possible to generate a function explaining the relationship between the input parameter to be corrected and the output parameter at the stage before mass production, and it is possible to generate a recipe as a correction candidate from the relationship. Step S1013 is a process corresponding to the processes of steps S508 to S513 and FIGS. 6 to 8. After that, mass production processing is started.

When the mass production process of the semiconductor device is started after step S1013, the wafer is introduced into the processing chamber (step S1021), and the etching apparatus etches the wafer (step S1022). Etching (step S1022) is composed of one step or a plurality of steps. When the etching (step S1022) is a plurality of steps, the etching of each step is executed by changing the discharge conditions. After the etching (step S1022) is completed, the wafer is carried out of the processing chamber (step S1023). Then, plasma cleaning is performed in order to remove the reactive organisms generated during etching and deposited on the surface of the treatment chamber (step S1024). If there is a next wafer (step S1025: Yes), the process returns to step S1021, and if there is no next wafer (step S1025: No), the process proceeds to post-mass production discharge data acquisition.

The input / output data of the etching apparatus being mass-produced is stored in the database 205, and the search apparatus 300 continues to analyze the input / output data in parallel in real time. As a result, it is possible to continue estimating the output data or the sensor data and the monitor data that can predict the change with time of the correction target during mass production. Further, the search device 300 can increase the number of data by aggregating the input / output data of a plurality of etching devices that execute the same mass production process.

When the discharge data is acquired after the mass production or when the designated time has elapsed from the start of the mass production, the search device 300 executes the search for the recipe for correcting the change with time (step S1031). Specifically, for example, the search device 300 searches the time-dependent change correction recipe by using the data acquired in steps S1011 and S1012, the relationship between the analyzed input / output data, and the correction candidate recipe as the input of step S502. .. Step S1031 is a process corresponding to steps S505 to S507.

Then, the search device 300 verifies the correction result by using the time-dependent change correction recipe which is the search result (step S1032). Step S1032 is a process corresponding to the processes of steps S508 to S513 and FIGS. 6 to 8.

Further, before the execution of steps S1031 and S1032, the search device 300 analyzes the output data acquired during the mass production of the semiconductor device, estimates the output data or the sensor data and the monitor data that can predict the change with time of the correction target, and estimates the output data or the sensor data and the monitor data. Step S1012 may be executed. As a result, a function explaining the relationship between the input parameter to be corrected and the output parameter can be generated, and a recipe as a correction candidate can be generated from the relationship. By using these results in step S502, the search device 300 can execute steps S1031 and S1032.

Further, as the input of step S502 in the above search, a recipe in which the input parameters frequently used for correction are changed by using the knowledge of the engineer, input / output data when discharging is performed using the recipe, and the analysis result thereof. You may use it.

Then, when the correction target is not achieved (step S1033: No), the process returns to step S1031 and when the correction target is achieved (step S1033: Yes), the process ends. Step S1033 is a process corresponding to step S514. When executing the search shown in FIG. 10, the search speed can be improved by proceeding in parallel with the search using a plurality of post-mass production etching devices that have performed the same mass production process. Further, the searched solution (input parameter) can be expanded to a plurality of etching devices that perform the same mass production process to perform correction.

In this way, by executing the search method by the search device 300 after mass production of the semiconductor, it is possible to correct the change with time of the output parameter value of the semiconductor processing device 201 being mass-produced (automatic change with time correction function).

In this way, the search device 300 automatically analyzes the value of the input parameter and the value of the output parameter of the semiconductor processing device 201, and after considering the analysis result, automatically sets the experimental conditions for searching the value of the input parameter. To decide with. Then, the search device 300 automatically verifies the experimental result, and by repeating these automatic operations, automatically obtains the target device state and the processing result (output parameter value) of the input parameter value. Can be searched for. As a result, the semiconductor processing device 201 itself can automatically draw out the device performance, and in addition, an engineer who develops a control model and selects device parameters (combination of input parameters and output parameters) for drawing out the device performance Can help.

As described above, the search device 300 according to the present embodiment has a condition set in the semiconductor processing device 201 for processing the semiconductor or a target value indicating the result of processing the semiconductor by the semiconductor processing device 201, and the condition and the result. The input unit 401 that accepts the input of the reference value indicated by the target value among the conditions or results in the search area defined by the range of, the set value of the condition in the search area, and the set value are semiconductors. The generation unit 402 that generates a prediction model showing the relationship between the condition and the result based on the measured value of the result when given to the processing device 201, and the input unit 401 to the prediction model generated by the generation unit 402. By giving the target value input by, the prediction value is acquired from the prediction model, and the specific unit 403 that identifies the existing area of the prediction value from the search area and the measured value of the result corresponding to the prediction value are the input unit 401. When the judgment unit 404 that determines whether or not the value is closer to the target value than the reference value input by, and the judgment unit 404 determines that the measured value of the result corresponding to the predicted value is closer to the target value. The setting unit 405 that sets the predicted value as the reference value and sets the existing area of the predicted value specified by the specific unit as the search area, and the actually measured value of the result corresponding to the predicted value satisfy the achievement condition of the target value. It has an output unit 406 that outputs a predicted value that satisfies the achievement condition in the case.

As a result, it is possible to improve the accuracy of reaching the best solution regarding the input / output of the semiconductor processing apparatus 201. Therefore, it is possible to improve the efficiency of operation and optimize the processing of the semiconductor processing apparatus 201.

Further, when the search device 300 determines that the actually measured value of the result corresponding to the predicted value obtained from the predicted model by the determination unit 404 is not closer to the target value, the determination unit 407 determines the region where the predicted value exists. The predicted value and the measured value of the result corresponding to the predicted value are determined as exclusion data, and the target given to the semiconductor processing apparatus 201 when the exclusion data and the exclusion data are obtained from the search area by the setting unit 405. The remaining area excluding the exclusion area specified by the value is set in the search area. As a result, the exclusion region in which the combination of the predicted value and the target value that is not close to the target value exists can be excluded from the latest search region, and the accuracy of reaching the best solution can be improved.

Further, when the search device 300 divides the search area into a plurality of areas by the division unit 408 and the determination unit 404 determines that the measured value of the result corresponding to the predicted value is closer to the target value, the search device 300 is a specific unit. According to 403, the predicted value is set as the reference value, and the region where the predicted value exists is specified from the plurality of divided regions. As a result, the existing region of the predicted value can be easily specified, and the search speed can be improved.

Further, the search device 300 divides the search area into a plurality of areas by the division unit 408, and the generation unit 402 sets the actual measurement value of the result when the set value of the condition in the division area is given to the semiconductor processing device 201. It is acquired for each divided region, and the generation unit 402 generates a prediction model based on the set value of the condition in each divided region and the actually measured value of the result. As a result, by using the plurality of semiconductor processing devices 201, the measured values can be acquired in parallel for each divided region, and the generation speed of the prediction model can be increased.

Further, when the determination unit 404 determines that the measured value of the result corresponding to the predicted value is not closer to the target value, the search device 300 obtains the data acquired by the determination unit 407 in the existing region of the predicted value. The exclusion data is determined, and the generation unit 402 excludes the specific actual measurement value obtained by excluding the exclusion data from the actual measurement value, and the specific measurement value excluding the setting value given to the semiconductor processing apparatus 201 when the exclusion data is obtained from the set value. A prediction model is generated based on the set value of. In addition, the exclusion region in which a combination of the predicted value and the set value that is not close to the target value exists can be excluded from the candidates for the predicted value, and the accuracy of the predicted model can be improved. Therefore, better prediction values can be obtained from the generated prediction model.

Further, the search device 300 detects the unstable operation of the semiconductor processing device 201 based on the actually measured value of the result and the predetermined output threshold value by the detection unit 409, and the detection unit 409 by the output unit 406. Outputs the detection result by. As a result, the user can be prompted whether or not the search can be continued.

The present invention is not limited to the above-mentioned examples, but includes various modifications and equivalent configurations within the scope of the attached claims. For example, the above-described examples have been described in detail in order to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to those having all the described configurations. Further, a part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Further, the configuration of another embodiment may be added to the configuration of one embodiment. In addition, other configurations may be added, deleted, or replaced with respect to a part of the configurations of each embodiment.

Further, each of the above-described configurations, functions, processing units, processing means, etc. may be realized by hardware by designing a part or all of them by, for example, an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing the program to be processed.

Information such as programs, tables, and files that realize each function is recorded in a storage device such as a memory, a hard disk, an SSD (Solid State Drive), or an IC (Integrated Circuit) card, an SD card, or a DVD (Digital Versaille Disc). It can be stored in a medium.

In addition, the control lines and information lines indicate those that are considered necessary for explanation, and do not necessarily indicate all the control lines and information lines necessary for implementation. In practice, it can be considered that almost all configurations are interconnected.

200 Semiconductor manufacturing system 201 Semiconductor processing device 202 Device control system 203 Monitor system 204 Sensor system 205 Database 206 Automatic control system 221 Recipe setting controller 222 Device basic setting controller 223 Setting error detection system 261 Target setting controller 262 Autonomous search system 263 Unstable operation Detection system 300 Search device 401 Input unit 402 Generation unit 403 Specific unit 404 Judgment unit 405 Setting unit 406 Output unit 407 Determination unit 408 Division unit 409 Detection unit

Claims (18)

The condition set in the semiconductor processing apparatus for processing the semiconductor or the target value indicating the result of processing the semiconductor by the semiconductor processing apparatus, and the condition or the condition within the search region defined by the range of the condition and the result. An input unit that accepts input of the reference value of the result indicated by the target value, and
Based on the set value of the condition in the search region and the actually measured value of the result when the set value is given to the semiconductor processing apparatus, a prediction model showing the relationship between the condition and the result is generated. Generation part and
By giving the target value input to the input unit to the prediction model generated by the generation unit, the prediction value is acquired from the prediction model, and the region where the prediction value exists is specified from the search area. When,
A determination unit for determining whether or not the actually measured value of the result corresponding to the predicted value is closer to the target value than the reference value input to the input unit.
When the determination unit determines that the measured value of the result corresponding to the predicted value is closer to the target value, the measured value of the result corresponding to the predicted value is set as the reference value, and the identification is performed. A setting unit that sets the existing area of the predicted value specified by the unit in the search area, and
An output unit that outputs a predicted value that satisfies the achievement condition when the measured value of the result corresponding to the predicted value satisfies the achievement condition of the target value.
A search device characterized by having. The search device according to claim 1.
When the determination unit determines that the measured value of the result corresponding to the predicted value is not closer to the target value, the predicted value in the region where the predicted value exists and the measured result of the result corresponding to the predicted value are measured. It has a decision part that determines the value as exclusion data
The setting unit is a remaining area excluding the exclusion area specified by the exclusion data determined by the determination unit from the search area and the target value given to the semiconductor processing apparatus when the exclusion data is obtained. Is set in the search area. The search device according to claim 1.
It has a dividing portion that divides the search area into a plurality of areas.
When the determination unit determines that the actually measured value of the result corresponding to the predicted value is closer to the target value, the specific unit sets the predicted value as the reference value and specifies the predicted value by the specific unit. A search device characterized in that the existing region of the predicted value is specified from a plurality of divided regions obtained by the divided portion. The search device according to claim 1.
It has a dividing portion that divides the search area into a plurality of areas.
The generation unit acquires the measured value of the result when the set value of the condition in the division region obtained by the division unit is given to the semiconductor processing apparatus for each division region, and obtains each of the division regions. A search device characterized by generating the prediction model based on the set value of the condition and the measured value of the result. The search device according to claim 1.
When the determination unit determines that the measured value of the result corresponding to the predicted value is not closer to the target value, the setting unit sets the measured value of the result corresponding to the predicted value as the reference value. A search device characterized in that the existing region of the predicted value is set in the search region without being set to. The search device according to claim 1.
When the determination unit determines that the measured value of the result corresponding to the predicted value is not closer to the target value, the predicted value in the region where the predicted value exists and the result corresponding to the predicted value It has a decision unit that determines the measured value as exclusion data.
The generator has a specific measured value obtained by removing the excluded data from the measured value and a specific setting excluding the set value given to the semiconductor processing apparatus when the excluded data is obtained from the set value. A search device comprising generating the prediction model based on a value. The search device according to claim 1.
It has a detection unit that detects unstable operation of the semiconductor processing apparatus based on the actually measured value of the result and a predetermined output threshold value.
The output unit is a search device characterized by outputting a detection result by the detection unit. The search device according to claim 1.
The setting unit sets the central condition of the condition to be given to the prediction model, and sets the central condition.
The specific section, the prediction model generated by the generating unit, by may grant the set centered condition by the setting unit obtains the predicted value from the prediction model, the presence region of the predicted value Identify from the search area,
A search device characterized by that. The search device according to claim 8.
The setting unit specifies the number of types of the condition that can be changed at one time from the central condition.
A search device characterized by that. It is a search method executed by the search device.
The search device is
The condition set in the semiconductor processing apparatus for processing the semiconductor or the target value indicating the result of processing the semiconductor by the semiconductor processing apparatus, and the condition or the condition within the search region defined by the range of the condition and the result. An input process that accepts the input of the reference value of the result indicated by the target value, and
Based on the set value of the condition in the search region and the actually measured value of the result when the set value is given to the semiconductor processing apparatus, a prediction model showing the relationship between the condition and the result is generated. Generation processing and
By giving the target value input by the input process to the prediction model generated by the generation process, the prediction value is acquired from the prediction model, and the region where the prediction value exists is specified from the search area. When,
Judgment processing for determining whether or not the actually measured value of the result corresponding to the predicted value is closer to the target value than the reference value input by the input processing.
When it is determined by the determination process that the predicted value is closer to the target value, the measured value of the result corresponding to the predicted value is set as the reference value, and the predicted value specified by the specific process is set. And the setting process to set the existing area of
Output processing that outputs a predicted value that satisfies the achievement condition when the predicted value satisfies the achievement condition of the target value, and
A search method characterized by executing. The search method according to claim 10.
The search device is
When it is determined by the determination process that the measured value of the result corresponding to the predicted value is not closer to the target value, the predicted value in the existing region and the measured value of the result corresponding to the predicted value are excluded. Execute the decision process to determine the data,
In the setting process, the search device uses the exclusion area specified by the exclusion data determined by the determination process from the search area and the target value given to the semiconductor processing device when the exclusion data is obtained. A search method characterized in that the remaining area excluding the above is set in the search area. The search method according to claim 10.
The search device is
A division process for dividing the search area into a plurality of areas is executed.
In the specific process, when the search device determines by the determination process that the actually measured value of the result corresponding to the predicted value is closer to the target value, the search device sets the predicted value to the reference value. A search method characterized in that an existing region of the predicted value specified by the specific processing is specified from a plurality of division regions obtained by the division processing. The search method according to claim 10.
The search device is
A division process for dividing the search area into a plurality of areas is executed.
In the generation process, the search device acquires the actually measured value of the result when the set value of the condition in the divided region obtained by the divided process is given to the semiconductor processing device for each divided region. , A search method characterized in that the prediction model is generated based on the set value of the condition in each of the divided regions and the actually measured value of the result. The search method according to claim 10.
The search device is
In the setting process, when the search device determines by the determination process that the actually measured value of the result is not closer to the target value, the actually measured value of the result corresponding to the predicted value is used as the reference value. A search method characterized in that the existing region of the predicted value is set in the search region without setting. The search method according to claim 10.
The search device is
When it is determined by the determination process that the actually measured value of the result corresponding to the predicted value is not closer to the target value, the predicted value in the region where the predicted value exists and the result corresponding to the predicted value Execute the decision process to determine the measured value as exclusion data,
In the generation process, the search device uses a specific measured value obtained by excluding the excluded data from the measured value and a set value given to the semiconductor processing device when the excluded data is obtained from the set value. A search method characterized in that the prediction model is generated based on a specific set value excluded. The search method according to claim 10.
The search device is
Based on the actually measured value of the result and the predetermined output threshold value, the detection process for detecting the unstable operation of the semiconductor processing apparatus is executed.
In the output process, the search device is a search method characterized in that the search device outputs a detection result by the detection process. The search method according to claim 10.
In the setting process, the search device sets the central condition of the condition to be given to the prediction model.
In the specific process, the search device obtains the predicted value from the predicted model by giving the central condition set by the setting process to the predicted model generated by the generation process , and obtains the predicted value of the predicted value. The existing area is specified from the search area.
A search method characterized by that. The search method according to claim 17.
In the setting process, the search device specifies the number of types of the condition that can be changed at one time from the central condition.
A search method characterized by that.

Images