JP2021012931A - Alternative proposal device, program and method - Google Patents

Abstract

To provide an alternative proposal device and an alternative proposal method for proposing an alternative which can eliminate a defect by using a computer with respect to a cause of the defect estimated by using machine learning.SOLUTION: Defect cause data related to causes of defects in workpieces processed by a semiconductor manufacturing apparatus and alternative data related to alternatives devised by an inventor in order to reduce or eliminate the defects are stored as teacher data in association with each other in a storage device. By executing machine learning based on the teacher data, an alternative proposal model showing the correspondence relationship between the defect cause data and the alternative data is constructed. Alternative data is output by inputting the data of the cause of a defect of a target work piece into the alternative proposal model.SELECTED DRAWING: Figure 3

Description

The present invention relates to the manufacture of semiconductors. The present invention particularly relates to estimating the state of a work piece processed by a semiconductor manufacturing apparatus.

When manufacturing a semiconductor device, various semiconductor manufacturing devices are used. For example, the Japan Semiconductor Manufacturing Equipment Association classifies semiconductor manufacturing equipment into seven major categories: semiconductor design equipment, mask / reticle manufacturing equipment, wafer manufacturing equipment, wafer process processing equipment, and assembly equipment. It is classified into inspection equipment and related equipment for semiconductor manufacturing equipment (http://www.seaj.or.jp/statistics/semi_classification.html).

These semiconductor manufacturing devices perform a predetermined process on the work piece. For example, in the so-called pre-process, the semiconductor manufacturing apparatus executes the following steps (1)-(9), respectively.
(1) Circuit design and pattern design are performed using a semiconductor design device.
(2) A photomask is created using a mask / reticle manufacturing apparatus based on the design data prepared in (1).
(3) A wafer is manufactured using a wafer manufacturing apparatus.
(4) Using a photoresist coating device (wafer process processing device), the photoresist is applied to the wafer manufactured in (3).
(5) Using the exposure / drawing apparatus (wafer process processing apparatus) and the photomask created in (2), the photomask pattern is printed on the wafer coated with the photoresist in (4).
(6) Etching is performed using a dry etching apparatus (wafer process processing apparatus), and the oxide film and resist are removed from the wafer on which the pattern is baked in (5).
(7) Using an ion implantation device (wafer process processing device), ions are implanted into the wafer etched in (6).
(8) Using a CMP (chemical mechanical polishing) device (wafer process processing device), the surface of the wafer that has been ion-implanted in (7) is polished.
(9) Using a sputtering apparatus (wafer process processing apparatus), a metal film for electrode wiring is formed on the surface of the wafer to which the ions have been implanted in (7).
(10) Using an inspection evaluation device (wafer process processing device), the wafer on which the metal film is formed in (9) is inspected.
The steps (4)-(8) are repeated a plurality of times to form an element on the wafer, and then the steps (7) to (10) are continued.

In each of steps (1)-(9), the operator inputs appropriate data into the semiconductor manufacturing apparatus. These data shall be called setting data. Depending on the operation of the operator, the semiconductor manufacturing apparatus executes a predetermined process on a product (hereinafter, simply referred to as a workpiece) that is the target of the process of the semiconductor manufacturing apparatus, such as a wafer or a processed product of a wafer. To do. Here, it is assumed that a process in which the same setting data is set in the same semiconductor manufacturing apparatus is executed for each of two workpieces having the same quality to the extent that they are substantially the same. At this time, since the conditions for executing the steps are almost the same, it is ideal to obtain two results having the same quality to the extent that they are substantially the same. However, in practice, a non-uniform result is obtained in which the defects present on one side are not present on the other side.

For example, assume that two wafers are manufactured in step (3) and neither of them has a defect. Step (4) is executed on each of these wafers to apply a photoresist. The same setting data of the photoresist coating device is used. At this time, if one of the photoresist-coated wafers has no defects, it is ideal that the other has no defects. However, in reality, there is a possibility that one of the photoresist-coated wafers has no defects, but the other has defects.

As described above, although the cause of the variation in the results of the semiconductor manufacturing process is known to some extent, it has not been completely clarified at present. It is considered that various factors such as not only the state of the semiconductor manufacturing equipment itself and the state of the work piece but also the state of the environment in which the semiconductor manufacturing equipment is installed are complicatedly involved in the generation of defects. It is not clear how the elements are related to the location and type of defects.

Therefore, when some kind of defect is found in the work piece after being processed by the semiconductor manufacturing apparatus, it is difficult to identify the cause. In some cases, a skilled worker may be able to estimate the cause of the defect based on his own experience and the condition of the work piece. However, the abilities of such skilled workers are beyond the scope of personal skills and skills. These abilities are not technical in the sense that they do not have the objectivity that can be transmitted to others as knowledge.

Further, in order to improve the quality of the manufactured semiconductor device, it is necessary to propose an alternative that can remove a part or all of the defects after estimating the cause of the defects. In the past, only some skilled workers could propose such alternatives.

In connection with the present invention, Patent Document 1 discloses a data generation method for converting position information of defective chips on a wafer into feature information suitable for classification. According to the same document, it is described that machine learning is performed based on the position information of defects actually occurring and the feature information extracted from the position information of defects (paragraph 0066 of the same document). Further, it is described that the problematic process is specified among the processes in the manufacture of the IC chip based on the obtained machine learning data (paragraph 0071 of the same document).

JP-A-2018-147978

According to Patent Document 1, it is described that a problematic process is estimated based on machine learning data. However, there is no description about estimating defects in products such as wafers.

Further, according to Patent Document 1, estimation is performed using discrimination information such as measurement data related to defects on the wafer, defect pattern images obtained from the above measurement data, and information (labels) of distribution pattern names of defect chips. It is stated that. In Patent Document 1, machine learning is performed based on data derived from the state of a semiconductor manufacturing apparatus or wafer and data derived from a name or the like given in advance to a distribution pattern, and based on a model obtained as a result. We are making an estimate. This estimation does not take into account the state of the environment in which the semiconductor manufacturing equipment is installed.

The present invention has been made in view of such a situation, and the problem to be solved by the present invention is to remove defects by using a computer for the causes of defects estimated by using machine learning. The purpose is to provide an alternative proposal device and an alternative proposal method for proposing possible alternatives.

In order to solve the above-mentioned problems, the present invention has, as one aspect, the defect cause data regarding the cause of the defect found in the workpiece processed by the semiconductor manufacturing apparatus, and the inventor in order to reduce or eliminate the defect. By performing machine learning based on a first storage means that stores teacher data that correlates with alternative data related to the alternative devised by the first storage means and the teacher data stored in the first storage means. By inputting the first machine learning means for constructing an alternative proposal model showing the correspondence between the defect cause data and the alternative data and the data on the cause of the defect into the alternative proposal model, the defect is introduced. Provided is an alternative proposal device including an alternative proposal means for outputting alternative data regarding alternatives for processing the work piece that can be reduced or removed.

Further, as another aspect, the present invention includes defect cause data regarding the cause of defects found in a work piece processed by a semiconductor manufacturing apparatus, and an alternative devised by the inventor in order to reduce or eliminate the defects. By executing machine learning based on the first storage means that stores the teacher data that is associated with the alternative data related to the above and the teacher data stored in the first storage means, the defect cause data. The defect can be reduced or eliminated by inputting the first machine learning means for constructing the alternative proposal model showing the correspondence between the data and the alternative data and the data on the cause of the defect into the alternative proposal model. A program for operating a computer is provided as an alternative proposal means for outputting alternative data regarding alternatives for processing the workpiece.

Further, as another aspect, the present invention includes defect cause data regarding the cause of defects found in a work piece processed by a semiconductor manufacturing apparatus, and an alternative devised by the inventor in order to reduce or eliminate the defects. By performing machine learning based on the first storage step, which stores the teacher data in association with the alternative data relating to the above, and the teacher data stored in the first storage means. The defect can be mitigated or reduced by inputting data on the cause of the defect into the alternative proposal model and the first machine learning stage for constructing an alternative proposal model showing the correspondence between the defect cause data and the alternative data. Provided is an alternative proposal method including an alternative proposal stage, which outputs alternative data regarding alternatives for processing the work piece that can be removed.

According to the present invention, in addition to manufacturing equipment setting data, manufacturing measurement data, workpiece measurement data, defect data, and defect cause data, the defect is reduced based on the cause of the defect estimated based on the environmental measurement data. Suggest alternatives that may be removed. Therefore, according to the present invention, when a work piece having a defect is manufactured on the production line, it is possible to change the production line necessary for reducing or removing the defect without requiring the judgment of a skilled worker.

It is a block diagram for demonstrating the semiconductor manufacturing system which is one Embodiment of this invention. It is a block diagram for explaining an alternative proposal apparatus. It is a block diagram for explaining an alternative proposal part. It is a flowchart for demonstrating the operation at the time of constructing the cause estimation model among the operation of the alternative proposal apparatus. It is a flowchart for demonstrating the operation at the time of estimating the cause of a defect among the operation of the alternative proposal apparatus. It is a block diagram for demonstrating the operation at the time of constructing an alternative proposal model among the operations of an alternative proposal apparatus. It is a block diagram for demonstrating the operation at the time of proposing an alternative among the operations of the alternative proposal apparatus. It is a block diagram for demonstrating one modification of an alternative proposal part. It is a block diagram for demonstrating one modification of an alternative proposal part. It is a block diagram for demonstrating one modification of an alternative proposal part. It is a block diagram for demonstrating the semiconductor manufacturing system which is one modification of one Embodiment of this invention. It is a table for demonstrating the example of the workpiece before and after processing, the input manufacturing apparatus setting data, and the output manufacturing apparatus measurement data in the semiconductor manufacturing apparatus which carries out a circuit design / pattern design process. It is a block diagram for demonstrating the semiconductor manufacturing system which is one Example of this invention. It is a block diagram for demonstrating the work piece inspection apparatus of the semiconductor manufacturing system of FIG.

A semiconductor manufacturing system 1 according to an embodiment of the present invention will be described with reference to FIG. The semiconductor manufacturing system 1 includes a semiconductor manufacturing device 2, a semiconductor manufacturing device measuring device 3, a workpiece inspection device 4, a clean room environment measuring device 5, a clean room environment control device 6, a factory environment measuring device 7, and a factory environment control device. 8. The alternative proposal device 40 is provided. The alternative proposal device 40 is connected to the off-factory environment measurement device 9 via the data communication network 10.

Among the semiconductor manufacturing systems 1, the semiconductor manufacturing device 2, the semiconductor manufacturing device measuring device 3, the workpiece inspection device 4, the clean room environment measuring device 5, and the clean room environment control device 6 are installed in the clean room 21. The clean room (hereinafter, also referred to as CR) 21 is an indoor space controlled so that suspended fine particles and suspended microorganisms in the air are kept below a predetermined cleanliness level. CR21 is located in the building of the factory (Fabricating lab, also referred to as Fab) 22. The factory 22 is a factory for semiconductor manufacturers and the like to manufacture semiconductors.

The semiconductor manufacturing equipment 2 is classified into seven major categories by the Japan Semiconductor Manufacturing Equipment Association, that is, semiconductor design equipment, mask / reticle manufacturing equipment, wafer manufacturing equipment, wafer process processing equipment, and assembly equipment. , Inspection equipment, semiconductor manufacturing equipment related equipment (http://www.seaj.or.jp/statistics/semi_classification.html).

Typical examples are given in each major classification. Devices for semiconductor design specifically include pattern input devices. The mask / reticle manufacturing apparatus includes an inspection evaluation apparatus, and particularly includes a defect inspection apparatus, a defect correction apparatus, a mask foreign matter inspection apparatus, and a visual inspection apparatus. Wafer manufacturing equipment includes inspection and evaluation equipment, and in particular, includes lifetime measuring equipment, crystal defect measuring equipment, processing inspection equipment, and various measuring equipment. Wafer process processing equipment includes inspection and evaluation equipment, and in particular, visual inspection equipment, length measurement SEM (Scanning Electron Microscope), and foreign matter inspection equipment. Assembling equipment includes inspection and evaluation equipment. It should be noted that these are merely examples, and other types of devices and devices may be used as the semiconductor manufacturing device 2.

The semiconductor manufacturing equipment (Semiconductor Production Equipment, hereinafter also referred to as SPE) 2 executes a predetermined process according to the manufacturing equipment setting data Vset input from the outside. When the SPE2 is a semiconductor design device, the manufacturing device setting data Vset is data generated by an input operation performed by the operator using a keyboard or a mouse when designing the semiconductor. When the SPE2 is another device, the manufacturing device setting data Vset is data representing the setting when the semiconductor manufacturing device operates. In this case, the manufacturing device setting data Vset is input from the numeric keypad, keyboard, or the like of the device, or from a terminal device that performs wired or wireless data communication with the device. While the manufacturing device setting data Vset is input to SPE2, it is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like. The manufacturing device setting data Vset may be directly input to the alternative proposal device 40 by using the input device (keyboard, mouse, etc.) of the alternative proposal device 40.

An object to be processed by SPE2 is called a processed object. For example, when SPE2 is an exposure / drawing apparatus for forming a pattern on a wafer surface, the work piece before processing is a wafer coated with a photoresist in advance. The processed product is a wafer on which a photomask pattern is printed.

The SPE2 includes a sensor that measures its own state and the state of the workpiece being machined by the SPE2, and outputs the measurement data as manufacturing measurement data Vman. The manufacturing measurement data Vman is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like.

The semiconductor manufacturing apparatus measuring apparatus (hereinafter, also referred to as SPE measuring apparatus) 3 is an apparatus for measuring the state of SPE2 and the state of a workpiece being machined by SPE2. The measurement data by the SPE measuring device 3 shall be referred to as manufacturing measurement data Vo. The manufacturing measurement data Vo is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like.

Generally, in a semiconductor manufacturing factory, various state values related to a semiconductor manufacturing apparatus are measured. Any of these various state values may be used as a state value to be measured by the SPE 2 and the SPE measuring device 3, that is, a state value to be included in the manufacturing measurement data Vman and Vo. For example, when SPE2 is a sputtering apparatus, the degree of vacuum in the vacuum chamber of the sputtering apparatus may be included in the manufacturing measurement data Vman or the manufacturing measurement data Vo.

Either one of the production measurement data Vman and Vo may be used. When only the manufacturing measurement data Vman is used, the SPE measuring device 3 may not be provided. On the contrary, when only the manufacturing measurement data Vo is used, the SPE2 does not have to be provided with a sensor for measuring the state of itself or the state of the workpiece being machined by the SPE2.

The SPE2 may include a vacuum transfer system. The vacuum transfer system is a system in which a substrate or the like is transferred in a vacuum, and includes a vacuum chamber through which an object to be transferred is passed. When the vacuum transfer system is provided, the SPE 2 or the SPE measuring device 3 may measure the degree of vacuum in the vacuum chamber and output it as a part of the manufacturing measurement data Vman or Vo. Further, the degree of vacuum in the vacuum chamber of the vacuum transfer system may be included in the setting device setting data Vset of SPE2.

The work piece inspection device 4 is a device that measures and inspects the state of the work piece after being processed by SPE2. The work piece inspection device 4 measures the work piece and outputs the measurement result as the work piece measurement data Vw. The work piece measurement data Vw is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like.

For example, in the case of a scanning electron microscope (SEM) in which the workpiece inspection device 4 scans the surface of the workpiece to generate and output an electron microscope image, the workpiece measurement data Vw is an electron microscope image. is there. Further, when the work piece inspection device 4 is an application device of SEM and is a length measuring SEM for measuring the size of the fine shape of the work piece surface, the work piece measurement data Vw is the size of the fine shape of the work piece surface. It is data.

The workpiece inspection device 4 may include a vacuum transfer system. In this case, the workpiece inspection device 4 may measure the degree of vacuum in the vacuum chamber of the vacuum transfer system and output it as a part of the workpiece measurement data Vw. Further, the degree of vacuum in the vacuum chamber of the vacuum transfer system may be included as the setting data of the workpiece inspection device 4.

Also, in general, the SEM sample chamber is evacuated. When the workpiece inspection device 4 includes the SEM, the degree of vacuum in the sample chamber of the SEM may be measured and output as a part of the workpiece measurement data Vw. Further, the degree of vacuum in the sample chamber of the SEM may be included as the setting data of the workpiece inspection device 4.

The clean room environment measuring device (hereinafter, also referred to as a CR environment measuring device) 5 measures the state of the environment in the clean room 21 and outputs the measured value as the CR environment measurement data Vcr. The environment measurement data Vcr in the CR is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like.

The CR internal environment measuring device 5 includes one or a plurality of measuring instruments according to the state value of the environment in the CR 21 to be measured. Examples of the state values to be measured by the environment measuring device 5 in the CR include air temperature, humidity, atmospheric pressure, amount of light, sound, electromagnetic waves, degree of vacuum, etc. in the CR21. In order to measure these state values, for example, various types such as thermometers, hygrometers, barometers, photometers, vibration meters (accelerometers), sound level meters (acoustic measuring instruments), electromagnetic wave measuring instruments, vacuum degree meters, etc. The CR in-CR environment measuring device 5 includes one or a plurality of types of measuring instruments. The measured values by one or more of these measuring instruments are output as the environment measurement data Vcr in CR.

The environment measuring device 5 in the CR has not only a state value known to have an influence on semiconductor manufacturing, but also a state value whose presence or absence is unknown at the present time and a state value which is considered to have no influence at the present time. May be measured and output as measurement data. By measuring a state value in which such an influence is considered to be unknown or not and using it in model construction, the alternative proposal device 40 may newly discover a state affecting semiconductor manufacturing.

The clean room environment control device (hereinafter, also referred to as a CR environment control device) 6 includes one or more devices for controlling the state of the environment in the clean room 21. The CR internal environment control device 6 controls the state of the environment in the CR 21 according to the CR internal environment setting data Vset_cr input from the outside. The environment setting data Vset_cr in the CR is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like. The input device (keyboard, mouse, etc.) of the alternative proposal device 40 may be used to directly input the CR environment setting data Vset_cr to the alternative proposal device 40.

Examples of the devices included in the CR internal environment control device 6 include a heating device, a cooling device, and an air conditioner as devices for controlling the temperature. In addition, as a device for controlling humidity, there are a humidifier and a dehumidifier. Further, as a device for controlling the amount of light, there is a lighting device capable of adjusting the amount of light. In addition, there is an ionizer as a device for removing static electricity. Further, as a device for removing suspended fine particles from the air in CR21, there is an air purifier. Further, there is an oxygen supply device as a device for controlling the oxygen concentration in CR21.

The SPE 2, the SPE measuring device 3, the workpiece inspection device 4, the CR internal environment measuring device 5, and the CR internal environment control device 6 are installed in the CR 21. On the other hand, the factory environment measuring device 7 and the factory environment control device 8 described below are installed outside the clean room 21 and inside the factory 22.

The factory environment measuring device (hereinafter, also referred to as a Fab environment measuring device) 7 measures the state of the environment outside the clean room 21 and inside the factory 22, and outputs the measured value as the Fab environment measurement data Vfab. .. The environment measurement data Vfab in the Fab is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like.

The in-fab environment measuring device 7 includes one or a plurality of measuring instruments according to the state value of the environment in the Fab 22 to be measured. Examples of the state value to be measured by the environment measuring device 7 in the Fab include the air temperature, humidity, atmospheric pressure, light intensity, sound, electromagnetic wave, and vacuum degree in the Fab 22. In order to measure these state values, for example, various types such as thermometers, hygrometers, barometers, photometers, vibration meters (accelerometers), sound level meters (acoustic measuring instruments), electromagnetic wave measuring instruments, vacuum degree meters, etc. The Fab internal environment measuring device 7 includes one or a plurality of types of measuring instruments. The measured values by one or more of these measuring instruments are output as the in-fab environment measurement data Vfab.

The in-Fab environment measuring device 7 has not only a state value that is known to have an effect on semiconductor manufacturing, but also a state value whose presence or absence is unknown at the present time and a state value that is considered to have no effect at the present time. May be measured and output as measurement data. By measuring a state value in which such an influence is considered to be unknown or not and using it in model construction, the alternative proposal device 40 may newly discover a state affecting semiconductor manufacturing.

The factory environment control device (hereinafter, also referred to as a Fab environment control device) 8 includes one or more devices for controlling the state of the environment outside the clean room 21 and inside the factory 22. Examples of such a device include a heating device, a cooling device, and an air conditioner as a device for controlling the temperature. In addition, as a device for controlling humidity, there are a humidifier and a dehumidifier. Further, as a device for controlling the amount of light, there is a lighting device capable of adjusting the amount of light. The in-fab environment control device 8 controls the state of the environment outside the CR 21 and in the factory 22 according to the in-fab environment setting data Vset_fab input from the outside. The environment setting data Vset_fab in the Fab is input to the alternative proposal device 40 described later via a data bus (not shown), a data communication line, or the like. The input device (keyboard, mouse, etc.) of the alternative proposal device 40 may be used to directly input the environment setting data Vset_fab in the Fab to the alternative proposal device 40.

As described above, in addition to the factory 22 and the one installed inside the factory 22, the semiconductor manufacturing system 1 is assumed to be installed outside the factory 22 and is also referred to as an outside factory environment measuring device (hereinafter, also referred to as a Fab outside environment measuring device). (Note) 9 is provided.

The Fab external environment measuring device 9 measures the state of the external environment of the factory 22, and outputs the measured value as the Fab external environment measurement data Vex.

Specifically, the Fab external environment measuring device 9 is outside the factory 22, and is provided on the site where the factory 22 is installed, such as a thermometer, a hygrometer, a barometer, a light meter, a vibration meter (accelerometer), and the like. Various measuring instruments and sensors such as hygrometers (acoustic measuring instruments) and electromagnetic wave measuring instruments.

Alternatively, the Fab outside environment measuring device 9 may be a measuring device installed outside the site of the factory 22. For example, various measuring instruments and sensors used by the Japan Meteorological Agency for collecting meteorological information may also be regarded as the Fab external environment measuring device 9.

The Fab external environment measuring device 9 has not only state values that are known to have an effect on semiconductor manufacturing, but also state values whose presence or absence is unknown at this time and state values that are considered to have no effect at this time. May be measured and output as measurement data. By measuring a state value in which such an influence is considered to be unknown or not and using it in model construction, the alternative proposal device 40 may newly discover a state affecting semiconductor manufacturing.

The Fab external environment measurement data Vex output by the Fab external environment measurement device 9 is transmitted to the alternative proposal device 40 described later via the data communication line 10. When the Fab outside environment measuring device 9 is a measuring instrument installed on the premises of the factory 22, it is conceivable to use a wired or wireless LAN line or a data bus such as USB (Universal Social Bus) as the data communication line 10. Be done. When the Fab external environment measuring device 9 is various measuring devices of the Japan Meteorological Agency, it is conceivable to use the Internet as the data bus line 10.

The alternative proposal device 40 includes SPE2, SPE measuring device 3, workpiece inspection device 4, CR internal environment measuring device 5, CR internal environment control device 6, Fab internal environment measuring device 7, Fab internal environment control device 8, Fab external environment. Various setting data and measurement data are received from the measuring device 9, and based on these, estimation processing related to the workpiece is performed.

Here, the manufacturing apparatus setting data Vset, the CR internal environment setting data Vset_cr, and the CR external environment setting data Vset_fab are collectively referred to as setting data. Further, among the setting data, the environment setting data Vset_cr inside the CR and the environment setting data Vset_fab outside the CR are particularly referred to as environment setting data. Further, the production measurement data Vo, the work piece measurement data Vw, the CR internal environment measurement data Vcr, the CR external environment measurement data Vfab, and the Fab external environment measurement data Vex are collectively referred to as measurement data.

Further, the setting data (manufacturing equipment setting data Vset, CR internal environment setting data Vset_cr, CR external environment setting data Vset_fab) and measurement data (manufacturing measurement data Vo, work piece measurement data Vw, CR) associated with one work piece. A set of internal environment measurement data Vcr, CR external environment measurement data Vfab, and Fab external environment measurement data Vex) shall be referred to as a data set. The data set associated with the workpiece to be estimated by the alternative proposal device 40 is referred to as a target data set.

(Alternative proposal device 40)
Next, the alternative proposal device 40 will be described. The alternative proposal device 40 estimates the cause of the defect in the work piece (wafer Wx) to be estimated, and proposes an alternative alternative to reduce or eliminate the estimated cause. Output the proposal data. In the following, the wafer Wx will also be referred to as a target workpiece.

Referring to FIG. 2, the alternative proposal device 40 includes a data input unit 41, a storage device 42, a cause machine learning unit 43, a cause estimation processing unit 44, and an alternative proposal unit 50.

The data input unit 41 is an interface device for receiving input of various data such as setting data, measurement data, defect data, and defect cause data. The data input unit 41 is, for example, an input device such as a keyboard and a mouse, a data bus interface such as USB, and a data communication interface device such as a LAN card. A plurality of types of these interface devices may be provided.

The storage device 42 is a readable and writable storage device. The storage device 42 may be, for example, a storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), or a memory device such as a RAM (Random Access Memory). The storage device 42 stores the setting data, the measurement data, the defect data, and the defect cause data input from the data input unit 41. Further, the storage device 42 stores various data generated by the alternative proposal device 40. The storage device 42 includes the following storage areas A11 to A14.

The storage area A11 is an area for storing a data set related to a work piece manufactured in the past, that is, a wafer. In the alternative proposal device 40, the data set includes setting data, measurement data, defect data, and defect cause data. The storage area A11 stores a large number of wafer data sets processed in the past by SPE2 in association with each wafer individually (for example, each identification information given to the wafer in advance).

The data stored in the storage area A11 may be input by the operator directly to the alternative proposal device 40 by using the keyboard of the data input unit 41. Alternatively, what the operator inputs with the input device of another information processing device may be received via the data communication interface device of the data input unit 41.

Defect data is data related to defects such as the position, size, and type of defects on the work piece that the work piece has. Even when no defect is included, the data indicating that there is no defect is referred to as defect data, and the defect data when there is no defect is particularly referred to as normal data. The defect data may be generated by the operator visually confirming the defect on the wafer. Alternatively, the defect data may be generated by identifying defects by a computer performing image recognition processing on an external image obtained by photographing the wafer.

The defect cause data is data indicating the cause of the defect, which is generated by a worker such as a skilled worker by judging each defect of the workpiece. An example of the cause of a defect is a shift in the position of the stage movement that occurs when the stage on which the wafer is placed is moved in the exposure apparatus. Further, as another example, in the ion implantation apparatus, there is a position shift when the ions are implanted into the wafer. In addition, as another example, there are cases where an electron microscope used in a wafer inspection device or the like is installed incorrectly, and the temperature and humidity of CR21, the factory 22, and the temperature and humidity outside the factory are inappropriate.

The storage area A12 is an area for storing a learned model for estimating the cause of defects (hereinafter, also referred to as a cause estimation model) constructed by the cause machine learning unit 43 described later.

The storage area A13 is an area for storing setting data and measurement data related to the wafer Wx to be estimated by the cause estimation processing unit 44, which will be described later.

The storage area A14 stores defect data related to the wafer Wx to be estimated. The defect data may be generated by the operator visually confirming the defect on the wafer. Alternatively, the defect data may be data generated by the computer performing image recognition processing on the external image of the wafer to detect the defect as described above.

The cause machine learning unit 43 and the cause estimation processing unit 44 are composed of an arithmetic processing unit and a program executed by the arithmetic processing unit. As the arithmetic processing unit of the cause machine learning unit 43 and the cause estimation processing unit 44, for example, a CPU (Central Processing Unit) or a GPGPU (General-purpose computing on graphics processing unit) can be considered.

The cause machine learning unit 43 reads setting data, measurement data, defect data, and defect cause data from the storage area A11 for each of a large number of wafers. Based on the read data for each wafer, the cause machine learning unit 43 performs machine learning to build a model. Cause The machine learning unit 43 stores the constructed model in the storage area A12.

The model represents a correspondence relationship between a combination of setting data, measurement data, and defect data related to one workpiece and a defect cause of each defect included in the defect data. The cause machine learning unit 43 constructs a cause estimation model by repeatedly constructing and updating a model based on setting data, measurement data, defect data, and defect cause data for a sufficient number of wafers.

The cause estimation processing unit 44 reads out the setting data and the measurement data for the wafer Wx to be estimated from the storage area A13. Further, the cause estimation processing unit 44 reads the defect data of the wafer Wx from the storage area A14. Further, the cause estimation processing unit 44 reads out the cause estimation model from the storage area A12. By applying or inputting these setting data, measurement data, and defect data to the cause estimation model, the cause estimation processing unit 44 calculates the estimated defect cause data, which is the estimation result of the cause of the defect existing in the wafer Wx. ..

The alternative proposal unit 50 will be described with reference to FIG. The alternative proposal unit 50 includes a storage device 51, an alternative machine learning unit 52, and an alternative proposal processing unit 53.

The storage device 51 is a readable and writable storage device. The storage device 51 may be, for example, a storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), or a memory device such as a RAM (Random Access Memory). The storage device 51 includes storage areas A15, A16, and A17. The storage device 51 may be integrated with the storage device 42. That is, instead of providing the storage device 51 separately from the storage device 42, the storage areas A15, A16, and A17 may be provided as a part of the storage device 42. In this case, the storage device 51 can be omitted.

The storage area A15 includes defect data, defect cause data, and alternative data as teacher data used when the alternative machine learning unit 52, which will be described later, constructs a trained model for alternative proposal (hereinafter, also referred to as an alternative proposal model). It is an area to store.

The defect data and defect cause data stored in the storage area A15 may be the same as those stored in the storage area A11, or may be different data. In the same case, the storage area A15 receives the defect data and the defect cause data from the storage area A11. In another case, the storage area A15 receives defect data and defect cause data from the data input unit 41.

The alternative data is data showing an alternative devised by a worker such as a skilled worker in order to reduce or eliminate defects. The storage area A15 stores defect data, defect cause data, and alternative data for one defect in association with each other. The storage area A15 receives alternative data from the data input unit 41.

Only a specific worker may devise the alternative data to be stored in the storage area A15. If the inventor of the alternative is limited to a specific skilled worker and the alternative proposal model is constructed, the alternative proposal device 40 can be expected to output the same alternative as the skilled worker. Alternatively, it may be stored in the storage area A15 regardless of the creator of the alternative data. In this case, the alternative output by the alternative proposal device 40 is the result of a kind of collective intelligence by the entire inventor.

For example, it is assumed that the cause of the defect indicated by the defect cause data is a shift in the position of the stage movement that occurs when the stage on which the wafer is placed is moved in the exposure apparatus. At this time, it is conceivable that a skilled worker or the like creates alternative data instructing the change of the setting data related to the stage movement and stores it in the storage area A15 via the data input unit 41.

The storage area A16 stores an alternative proposal model constructed by the alternative machine learning unit 52, which will be described later.

The storage area A17 stores the estimated defect cause data of the wafer Wx estimated by the cause estimation processing unit 44.

The alternative machine learning unit 52 and the alternative proposal processing unit 53 are composed of an arithmetic processing unit and a program executed by the arithmetic processing unit. As the arithmetic processing device of the alternative machine learning unit 52 and the alternative proposal processing unit 53, for example, a CPU or GPGPU can be considered.

The alternative machine learning unit 52 builds an alternative proposal model based on the defect cause data and the alternative data stored in the storage area A15. The alternative proposal model represents the correspondence between the position, type, and cause of defects in the workpiece processed in the past by SPE2 and the alternative devised by the inventor such as a skilled worker. The alternative machine learning unit 52 stores the constructed alternative proposal model in the storage area A16.

The alternative machine learning unit 52 divides the surface or the inside of the work piece into a region of a predetermined size in advance, and includes the position of the defect for each of the defects indicated by the defect data stored in the storage area A15. Identify the area of. The region referred to here is, for example, a rectangular region surrounded by four-sided grids when a virtual grid is drawn on the wafer surface, or a region surrounded by arcs formed by the three-sided grid and the edge of the wafer. Refer to defect data for each defect type. Refer to the defect cause data for the defect cause of each defect. Refer to the alternative data for alternatives to mitigate or eliminate each defect. The alternative machine learning unit 52 constructs an alternative proposal model based on the read out defect causes and alternatives for each region and each type.

The alternative proposal processing unit 53 reads the alternative proposal model from the storage area A16. Further, the alternative proposal processing unit 53 reads the estimated defect cause data of the wafer Wx from the storage area A17. Then, the alternative proposal processing unit 53 inputs the estimated defect cause data of the wafer Wx into the alternative proposal model, and calculates data indicating an alternative for reducing or eliminating defects existing in the wafer Wx, that is, alternative proposal data. And output.

(Construction of cause estimation model)
Among the operations of the alternative proposal device 40, the operation of constructing the cause estimation model will be described with reference to FIG.

First, teacher data is stored in advance in the storage area A11 of the storage device 42. The teacher data includes setting data for a wafer machined in the past using SPE2, measurement data measured when the wafer is machined with SPE2, defect data of the wafer, and defect cause data of the wafer. It is an association. The association is performed by, for example, a worker or the like assigning identification information to each wafer in advance and associating the identification information with the setting data, measurement data, defect data, and defect cause data of the wafer. .. A sufficient number of sets of such setting data, measurement data, defect data, and defect cause data are stored in the storage area A11 so that a practical model can be constructed.

As for the defect data, the human-generated defect data may be stored in the storage area A11. In this case, an operator such as an operator of SPE2 visually observes the wafer used for the teacher data to identify the presence / absence, position, type, size, etc. of defects, and the identification result is used as defect data in the data input unit 41. It is stored in the storage area A11 via.

Alternatively, the alternative proposal device 40 may include a computer that executes an image recognition process for generating the defect data as described above. In this case, the operator does not need to prepare the defect data. Instead, the measurement data includes the appearance image of the wafer, and the defect data is generated and stored in the storage area A11 by executing the image recognition process on the appearance image.

For the defect cause data, the result of observing each wafer processed in the past by a worker such as a skilled worker and identifying the defect cause is used. The defect cause data is stored in the storage area A11 via the data input unit 41.

The cause machine learning unit 43 reads out the setting data, measurement data, defect data (including normal data), and defect cause data of the storage area A11 as teacher data (step S21). Then, a model is constructed based on these teacher data (step S22).

After building the model based on a sufficient number of teacher data, the cause machine learning unit 43 stores the cause estimation model in the storage area A12 (step S23).

(Estimation of the cause of defects)
Next, among the operations of the alternative proposal device 40, the operation of the cause estimation processing unit 44 will be described. The cause estimation processing unit 44 estimates the cause of defects in the target wafer Wx, which is a specific wafer processed by SPE2. It is assumed that the setting data, the measurement data, and the defect data when the target wafer Wx is processed by the SPE2 are stored in the storage area A14 in advance.

With reference to FIG. 5, the cause estimation processing unit 44 reads the cause estimation model from the storage area A12 (step S31). Further, the cause estimation processing unit 44 reads out the setting data and the measurement data of the target wafer Wx from the storage area A13 (step S32). Further, the cause estimation processing unit 44 reads out the defect data from the storage area A14 (step S33).

Next, the cause estimation processing unit 44 applies the read setting data, measurement data, and defect data (including normal data) to the cause estimation model read in step S33 to obtain each defect of the target wafer Wx. The probability of each defect cause is calculated (step S34). For example, defect cause data includes a shift in the stage movement position in the exposure apparatus, a shift in the ion implantation position in the ion implantation apparatus, an electron microscope installation error, and an abnormality in temperature and humidity in the CR21.

Next, the cause estimation processing unit 44 generates and outputs estimated defect cause data based on the calculated probability of each defect cause (step S35). The estimated defect cause data consists of the defect cause generated as the defect cause data. For example, it includes one or more defects such as a shift in the stage movement position in the exposure apparatus, a shift in the ion implantation position in the ion implanter, an electron microscope installation error, and an inappropriate environmental temperature and humidity. The estimated defect cause data further includes the probability of each defect cause.

(Construction of alternative proposal model)
The operation when the alternative proposal unit 50 constructs the alternative proposal model will be described with reference to FIG.

In the storage area A15, defect data, defect cause data, and alternative data associated with each other are stored in advance as teacher data. The storage area A15 stores in advance a sufficient number of teacher data so that a practical alternative proposal model can be constructed.

The defect data is defect data about defects in a workpiece (for example, a wafer) processed in the past using SPE2.

The defect cause data is data indicating the cause of the defect, which is generated by a worker such as a skilled worker by judging each defect of the workpiece. In addition to or instead of the defect cause data determined and generated by the operator, the defect cause data output by the cause estimation processing unit 44 may be used. An example of the cause of a defect is a shift in the position of the stage movement that occurs when the stage on which the wafer is placed is moved in the exposure apparatus. Further, as another example, in the ion implantation apparatus, there is a position shift when the ions are implanted into the wafer. In addition, as another example, there are cases where an electron microscope used in a wafer inspection device or the like is installed incorrectly, and the temperature and humidity of CR21, the factory 22, and the temperature and humidity outside the factory are inappropriate.

Alternative data was devised and actually devised by skilled workers and other workers to reduce the degree of defects or eliminate the defects when there are certain types of defects on the surface or inside of the work piece. It is data showing an alternative that was able to reduce or eliminate defects. Alternative data includes, for example, changes in setting data and measurement data. More specifically, when the SPE2 is an exposure apparatus, the setting data regarding the stage movement is changed, and when the SPE2 is an ion implantation apparatus, the setting data regarding the ion implantation position on the wafer is changed.

The alternative machine learning unit 52 reads out the defect cause data and the alternative data of the storage area A15 as teacher data (step S41). Then, a model is constructed based on these teacher data (step S42). After building the model based on a sufficient number of teacher data, the alternative machine learning unit 52 stores the alternative proposal model in the storage area A16 (step S43).

(Proposal of alternative)
Next, the operation of the alternative proposal processing unit 53 will be described with reference to FIG. 7.

The alternative proposal processing unit 53 reads the target data set of the wafer Wx from the storage area A13 (step S51).

Next, the alternative proposal processing unit 53 reads out the estimated defect data of the wafer Wx from the storage area A14 (step S52).

Next, the alternative proposal processing unit 53 reads out the estimated defect cause data of the wafer Wx from the storage area A17 (step S53).

Next, the alternative proposal processing unit 53 reads the alternative proposal model from the storage area A16 (step S54).

Next, the alternative proposal processing unit 53 inputs the read target data set of the wafer Wx, estimated defect data, and estimated defect cause data into the alternative proposal model, and generates alternatives for each of the defects of the wafer Wx. , Output (step S55). When the cause estimation processing unit 44 outputs a plurality of estimated defect causes for one defect, one alternative is output for each estimated defect cause.

As described above, if a worker or the like performs the same processing using SPE2 according to the alternative output by the alternative proposal device 40, a processed product having defects reduced or removed can be obtained. If an alternative is generated by using the alternative proposal device 40, an alternative similar to that of a skilled worker can be obtained even if a skilled worker is absent.

(Transformation of alternative proposal unit 50 1: alternative proposal unit 60)
The alternative proposal unit 60, which is a modification of the alternative proposal unit 50, will be described with reference to FIG. The alternative proposal unit 60 includes a plurality of alternative proposal units 50A, 50B, and 50C, each of which corresponds to the alternative proposal unit 50.

Hereinafter, as the reference code when referring to each part of the alternative proposal unit 50A, 50B, 50C, the reference code of each part of the alternative proposal unit 50 with A, B or C added to the end is used. For example, the storage areas A15 of the alternative proposal unit 50A, 50B, and 50C are referred to as storage areas A15A, A15B, and A15C, respectively.

The alternative proposal units 50A, 50B, and 50C correspond to the above-mentioned alternative proposal unit 50, respectively. However, the teacher data stored in the storage areas A15A, A15B, and A15C are different from each other. Also, the alternative proposal models generated based on these different teacher data are different from each other.

In the storage area A15 of the alternative proposal unit 50 described above, alternative data by a specific inventor or alternative data regardless of the inventor is stored. On the other hand, the storage areas A15A, A15B, and A15C each store alternative data by a specific inventor. Here, the creators of the alternative data stored in the storage areas A15A, A15B, and A15C are referred to as the creators A, B, and C in order. The alternative proposal model stored in the storage area A16A mimics the judgment of the inventor A. Similarly, the alternative proposal model stored in the storage areas A16B and A16C mimics the judgments of the inventors B and C, respectively.

The alternative proposal units 50A, 50B, and 50C output alternative proposal data A, B, and C, respectively. As a result, the user of the alternative proposal device 40 can obtain three alternative proposal data A, B, and C that imitate the judgments of the three creators A, B, and C. The user may process three workpieces according to each of the three alternative proposal data A, B, and C, or the user compares the alternative proposal data A, B, and C, and one of the alternative proposal data. The work piece may be processed according to the above. Alternatively, the user may devise another alternative proposal data with reference to the alternative proposal data A, B, and C.

Here, the alternative proposal unit 60 has been described as having three alternative proposal units 50A, 50B, and 50C, but the number of alternative proposal units may be two or four or more. According to this modification, it is possible to propose an alternative based on an alternative proposal model, which is derived from an alternative in which the inventor is different, even if the inventor is absent.

(Transformation of alternative proposal unit 50: alternative proposal unit 70)
The alternative proposal unit 70, which is another modification of the alternative proposal unit 50, will be described with reference to FIG. The alternative proposal unit 70 includes an image composition unit 71 in addition to the configuration of the alternative proposal unit 60 described above.

Based on the alternative proposal data A generated by the alternative proposal unit 50A, the image composition unit 71 generates a computer image of the processed product that is expected when processing is performed using the alternative, and the composite image A is generated. Output as. Similarly, the composite images B and C are output based on the alternative proposal data B and C, respectively. When synthesizing images, the image synthesizing unit 71 shall read necessary setting data, measurement data, and the like from the storage devices 42 and 51. As the image compositing unit 71, for example, GAN (Generative Adversarial Network), VAE (Variational Auto Encoder), or the like can be considered.

According to this modification, the user of the alternative proposal device 40 can see the state of the workpiece after processing by adopting the alternative as a computer image, so that it is easier to compare the alternative proposal data A, B, and C. Can be done. Therefore, the user can intuitively select the alternative proposal data A, B, and C. In addition, it becomes easy to obtain awareness from alternative proposal data A, B, and C.

(Transformation of alternative proposal unit 50: alternative proposal unit 80)
The alternative proposal unit 80, which is another modification of the alternative proposal unit 50, will be described with reference to FIG. The alternative proposal unit 80 includes an alternative comparison unit 81 in addition to the configuration of the alternative proposal unit 60 described above.

The alternative comparison unit 81 receives alternative proposal data A, B, and C from the alternative proposal units 50A, 50B, and 50C, respectively. The alternative comparison unit 81 compares these alternative proposal data A, B, and C with each other, and outputs any one of them as alternative proposal data.

Various methods can be considered for comparing alternative proposal data A, B, and C. For example, alternative proposal data A, B, and C each contain probabilities, but it is conceivable to compare these probabilities with each other and select and output the one containing the highest probability. Further, any output may be selected based on the hyperparameters of the alternative proposal unit 50A, 50B, 50C. In addition, a majority vote may be taken with the alternative proposal data A, B, and C.

According to this modification, the user of the alternative proposal device 40 can obtain the alternative proposal created by the consensus of the inventors A, B, and C.

(Semiconductor Manufacturing System 60)
Although the alternative proposal device 40 has been described above, the semiconductor manufacturing system of the present invention is not limited to a system centered on a single semiconductor manufacturing device, such as the semiconductor manufacturing device 2 in the semiconductor manufacturing system 1.

The semiconductor manufacturing system 60 shown in FIG. 5 includes a semiconductor design device 2A, a mask / reticle manufacturing device 2B, an exposure device 2C, an etching device 2D, an ion implantation device 2E, a CMP device 2F, and a sputtering device 2G as semiconductor manufacturing devices. , The photoresist coating apparatus 2H is provided. Further, as the workpiece inspection device 4, a wafer inspection device 4A is provided. As described above, the semiconductor manufacturing system 60 includes eight types of semiconductor manufacturing devices, but the operation is the same as that of the semiconductor manufacturing system 1 described above.

FIG. 6 shows examples of the pre-processed workpiece, the post-machined workpiece, the input setting data, and the output measurement data of each of the semiconductor manufacturing apparatus 2A to 2H. In FIG. 6, a CAD device (row B) is given as an example of a device that is responsible for the circuit design / pattern design process (row A) among the pre-processes of semiconductor manufacturing. There are no objects input / output from the CAD device (column C, column D). There is an input operation performed on the CAD device by the operator of the CAD device as the data input to the CAD device, that is, the data corresponding to the manufacturing device setting data of the CAD device (column E). The data output by the CAD device, that is, the manufacturing device measurement data of the CAD device, includes CAD design data, circuit design data, pattern design data, and OPC (Optical Proximity correction) correction data (column F).

(Example 1)
A semiconductor manufacturing system 90, which is an embodiment of the present invention, will be described with reference to FIG. Compared with the semiconductor manufacturing system 1 described above, the semiconductor manufacturing system 90 differs in that the alternative proposal device is built in the workpiece inspection device.

The workpiece inspection device 91 of the semiconductor manufacturing system 90 will be described with reference to FIG. The work piece inspection device 91 includes an alternative proposal device 40, and further includes an image generation unit 92 and a control unit 93.

The image generation unit 92 takes an image of the work piece as a sample and generates an image of the work piece. Specifically, in this embodiment, it is assumed that the SEM type D2DB (Die to Database) inspection device is used as the image generation unit 92. The SEM type D2DB inspection device is an inspection device that inspects by comparing the design data (CAD) of a semiconductor circuit with the image obtained by imaging a workpiece using SEM. This type of inspection device is described, for example, in Japanese Patent No. 3524853. The image generation unit 92 outputs the work piece measurement data Vw to the data input unit 41.

In this example, SEM was used as the image generation unit 92. An SEM creates an image of the sample surface by shining electrons on the surface of the sample and detecting the electrons reflected or generated from it. Instead of such SEM, a TEM (Transmission Electron Microscope) is used as an image generation unit 92, and an electron beam transmitted through the sample is made to collide with a phosphor screen to obtain a magnified image of the sample. May be good. Further, an image generation unit 92 may be used as an image generation unit 92 that generates an digitized image by reading an optically magnified image using an optical microscope with a solid-state image sensor.

The control unit 93 is a computer that controls the operation of the entire workpiece inspection device 91 including the alternative proposal device 40. The control unit 93 includes an arithmetic unit for controlling the operation of the entire workpiece inspection device 91, and a storage device for storing a program for causing a computer to execute the control operation.

For convenience of explanation, the control unit 93 is described in a form independent of the alternative proposal device 40, but the cause machine learning unit 43, the cause estimation processing unit 44, the alternative proposal unit 50, and the control unit 93 are configured by the same computer. It may be obvious to those skilled in the art that it may be composed of a plurality of computers.

As described above, the alternative proposal device 40 outputs the alternative proposal data. The alternative indicated by the alternative proposal data depends on the type of data input from the data input unit 41. Here, a specific combination of data input from the data input unit 41 to the storage device 42 and an effect obtained by the combination will be illustrated.

As an example, difference data comparing an image obtained by scanning a work piece with design data of the work piece and data indicating an on / off state of another device existing around the work piece inspection device 91 (peripheral device). Combination with on / off data) is conceivable.

A method for acquiring difference data between an image obtained by scanning a work piece and design data of the work piece is described in, for example, Japanese Patent No. 3524853. According to this method, in the inspection target pattern image obtained by scanning the workpiece to be inspected, the deformation that may occur in the inspection target pattern is corrected, the design data is converted into a reference pattern and stored in the storage device. To do. In addition, the pattern image to be inspected is imaged, and the edge of the pattern image to be inspected is detected with subpixel accuracy. The difference between the edge of the inspection target pattern image detected in this way and the edge of the reference pattern stored in advance is obtained as difference data.

In addition to the above-mentioned difference data, the inclination of the image profile obtained at the time of acquiring the difference data may be used. Alternatively, statistics such as the mean value, standard deviation, maximum value, and minimum value of the difference data may be used. Alternatively, the ratio of the portion where the edge could not be detected due to insufficient contrast of the image may be used.

Peripheral device on / off data is any one or a combination of data indicating an on / off state of various devices inside the CR21, inside the Fab22, or outside the Fab22.

For example, the data input unit 41 acquires the on / off state of various devices in the CR21 including the SPE2 as a part of the manufacturing device setting data Vset. The data input unit 41 acquires the on / off state of various devices controlled by the CR internal environment control device 6 as a part of the CR internal environment setting data Vset_cr. The data input unit 41 acquires the on / off state of various devices controlled by the in-fab environment control device 8 as a part of the in-fab environment setting data Vset_fab. The data input unit 41 acquires the on / off state of various devices to be monitored by the Fab external environment measurement device 9 as a part of the Fab external environment measurement data Vex. Here, as a monitoring target by the Fab outside environment measuring device 9, for example, a device having a large power consumption installed on a site adjacent to the Fab 22 can be considered.

Consider a case where the alternative proposal device 40 generates alternative proposal data based on the difference data between the image obtained by scanning the workpiece and the design data and the peripheral device on / off data. In this case, the cause machine learning unit 43 compares the image obtained by scanning the workpiece with the design data of the workpiece, the difference data, the on / off state of the peripheral devices inside and outside the CR21 and the Fab22, and the workpiece. It is possible to build a trained model that reflects the relationship between the presence or absence of defects in. As a result, the cause estimation processing unit 44 is based on a learned model that reflects the difference data comparing the image obtained by scanning the workpiece with the design data of the workpiece and the on / off state of the peripheral device. , Alternative proposal data can be generated.

Although the present invention has been described above in accordance with the embodiments and modifications thereof, the present invention is not limited thereto. For example, in the above-described embodiment and modification, the environmental measurement data includes the atmosphere (temperature, humidity, pressure, noise / sound), electromagnetic environment (light intensity, radio field strength, magnetic field strength), clean room 21, vibration state of factory 22, and the like. , Although it has been described as including the measured value obtained by measuring the natural phenomenon, the measured value of the human activity may be included as the environmental measurement data in addition to or instead of the measured value of the natural phenomenon.

For example, data on workers involved in semiconductor manufacturing, such as the time of entering and exiting the clean room 21 and the factory 22, the names of the entering and exiting persons, and the like may be used as part of the environmental measurement data. Further, data on a person who is not directly involved in semiconductor manufacturing, for example, data indicating an operating state of a facility existing in the vicinity of the factory 22, may be used as a part of the environmental measurement data.

Further, the manufacturing apparatus setting data, the manufacturing measurement data, the workpiece measurement data, and the environment measurement data may include data representing the specifications of the apparatus, that is, the specification data. .. The specification data includes a semiconductor manufacturing device 2, an SPE measuring device 3, a workpiece inspection device 4, a CR internal environment measuring device 5, a CR internal environment control device 6, a Fab internal environment measuring device 7, a Fab internal environment control device 8, and the like. It is data which shows the size, weight, performance, etc. of the whole or a part of the Fab external environment measuring apparatus 9. For example, the specification data of the single crystal pulling device may include the crucible diameter.

The degree of vacuum may be included as part of the manufacturing measurement data and the manufacturing setting data. For example, in general, when the SPE2 includes a sputtering device, the inside of the vacuum chamber of the sputtering device is evacuated. Further, when the SPE2 is provided with a vacuum transfer system that transfers a substrate or the like in a vacuum, the inside of the vacuum chamber of the vacuum transfer system is evacuated. The degree of vacuum related to these vacuum states may be included as a part of the manufacturing apparatus setting data Vset, the manufacturing measurement data Vman, and the manufacturing measurement data Vo.

Further, when the workpiece inspection device 4 includes the SEM, the SEM generally includes a sample chamber for arranging the sample, and the inside of the sample chamber is evacuated. This degree of vacuum may be included as a part of the work piece measurement data Vw. Further, when the workpiece inspection device 4 is provided with the vacuum transfer system, the degree of vacuum in the vacuum chamber of the vacuum transfer system may be included as a part of the workpiece measurement data Vw.

The alternative proposal device 40 receives the specification data as a part of the setting data or the measurement data from the semiconductor manufacturing device 2, the various measuring devices 3 to 5, 7, 9, and the various environmental control devices 6 and 8, but instead proposes an alternative. Specification data may be input directly from the data input unit 41 of the device 40.

Further, in the above description, the machine learning method is described on the premise of machine learning with teacher data, but the present invention is not limited to a specific machine learning method. For example, semi-supervised learning and unsupervised learning may be used. Further, instead of rule-based machine learning, so-called deep learning in which neural networks are multi-layered may be used.

With such machine learning, or instead of machine learning, estimation may be performed using artificial intelligence. In particular, it is preferable to use artificial intelligence that applies brain science or artificial intelligence that applies developmental psychology.

Further, in the above description, the operation of constructing the trained model only once has been described, but the present invention is not limited to this. Another trained model may be constructed and used in place of the previous trained model by updating the dataset and performing the model construction again.

Further, in the above description, it has been described that various setting data and measurement data in each process can be obtained unconditionally. The present invention is not limited to the case where various setting data and measurement data can be obtained unconditionally, and can be carried out even when there are restrictions on the provision of setting data and measurement data from the process.

When manufacturing a semiconductor device, it may include a process performed not only at its own factory but also at another company's factory. Depending on the factory, setting data and measurement data are know-how and may be treated as non-public information.

In such a case, the factory of another company may give a label to the data type of the setting data and the measurement data, and provide the setting data and the measurement data in a hidden manner. For example, instead of providing the numerical values together with the labels indicating the measured physical quantities themselves such as room temperature, humidity, and vacuum degree, other companies' factories use the numerical label indicating the room temperature as the first parameter and the numerical label indicating the humidity as the first parameter. The numerical value is provided with two parameters and a numerical label indicating the degree of vacuum as the third parameter.

If a numerical value is used as it is, the original label of the numerical value may be estimated from the range of the numerical value. In order to avoid this, in addition to the above-mentioned separate labeling, the factories of other companies may substitute the actual numerical value into a predetermined function and provide the obtained value instead of the actual numerical value. For example, when the measurement data of room temperature is 25 degrees, the label "first parameter" is used instead of the label "room temperature". Instead of the number 25, use an appropriate linear function, for example the number 80 obtained by substituting y = 3x + 5. As a result, the third-party factory provides the label "first parameter", number = 80, instead of providing the label "room temperature", number = 25.

In the above description, the alternative proposal device 40 simply outputs the alternative proposal data. Instead of this, the alternative proposal data may be fed back in the semiconductor manufacturing system 1. Based on the alternative proposal data, various setting data, that is, manufacturing equipment setting data Vset, CR environment setting data Vset_cr, and Fab environment setting data Vset_fab are changed. In this way, by feeding back the alternative proposal data and updating various setting data, the semiconductor manufacturing system 1 can autonomously improve the quality of the workpiece processed by the SPE2.

1 Semiconductor manufacturing system 2 Semiconductor manufacturing equipment (SPE)
2A Semiconductor Design Equipment (SPE)
2B Mask / Reticle Manufacturing Equipment (SPE)
2C exposure equipment (SPE)
2D etching equipment (SPE)
2E Ion Implantor (SPE)
2F CMP equipment (SPE)
2G sputtering equipment (SPE)
2H photoresist coating device (SPE)
3 Semiconductor manufacturing equipment (SPE) measuring equipment 4 Work piece inspection equipment 4A Wafer inspection equipment 5 Clean room (CR) internal environment measuring equipment 6 Clean room (CR) internal environment control equipment 7 Factory (Fab) internal environment measuring equipment 8 Factory (Fab) Internal environment control device 9 Factory (Fab) External environment measurement device 10 Data communication network 21 Clean room (CR)
22 Factory (Fab)
40 Alternative proposal device 50, 50A, 50B, 50C, 60, 70, 80 Alternative proposal unit 51 Storage device 52 Alternative machine learning unit 53 Alternative proposal processing unit 71 Image synthesis unit 81 Alternative comparison unit 91 Work piece inspection device 92 Image generation unit 93 Control unit

Claims (8)

Teacher data that correlates defect cause data related to the cause of defects found in a workpiece processed by a semiconductor manufacturing apparatus with alternative data related to an alternative devised by the inventor in order to reduce or eliminate the defect. The first storage means to memorize,
A first machine learning that constructs an alternative proposal model showing a correspondence relationship between the defect cause data and the alternative data by executing machine learning based on the teacher data stored in the first storage means. Means and
It is provided with an alternative proposal means for outputting alternative data regarding the processing of the target work piece, which can reduce or eliminate the defect by inputting the data regarding the cause of the defect of the target work piece into the alternative proposal model. , Alternative proposal device. With a plurality of the first storage means,
The plurality of first storage means store a plurality of teacher data in which the inventor who devised the alternative data stores different teacher data.
The first machine learning means constructs a plurality of the alternative proposal models corresponding to the plurality of teacher data.
The alternative proposal means outputs a plurality of the alternative data by inputting data regarding the cause of the defect for each of the plurality of alternative proposal models.
The alternative proposal device according to claim 1. The alternative proposal device according to claim 1 or 2, further comprising an alternative comparison means for comparing the plurality of alternative data with each other and outputting one or a plurality of the alternative data selected based on the comparison result. The invention according to any one of claims 1 to 3, further comprising an image synthesizing means for synthesizing an image showing the state of the processed product processed according to the alternative indicated by the alternative data based on the alternative data. Alternative proposal device. Manufacture obtained by measuring at least one of the manufacturing device setting data set in the semiconductor manufacturing device, the state of the semiconductor manufacturing device, and the state of the work piece being processed as the processing target of the semiconductor manufacturing device. Measurement data, workpiece measurement data obtained by measuring the state of the workpiece after processing with the semiconductor manufacturing apparatus, and environment measurement data obtained by measuring the state of the environment surrounding the semiconductor manufacturing apparatus. A data input means that accepts input of defect data, which is data related to defects in a workpiece processed in the past by the semiconductor manufacturing apparatus, and defect cause data, which is data related to the cause of defects related to the defect data.
The manufacturing apparatus setting data, the manufacturing measurement data, the workpiece measurement data, the environment measurement data, the defect data, and the defect data relating to one of the workpieces processed in the past by the semiconductor manufacturing apparatus. A second storage means for storing a plurality of past data sets as one data set in which defect cause data are associated with each other, and
By executing machine learning based on the plurality of past data sets stored in the second storage means, the manufacturing apparatus setting data, the manufacturing measurement data, the work piece measurement data, the environment measurement data, and the like. A second machine learning means for constructing a model showing the correspondence between the combination of the defect data and the defect cause data.
It is provided with an estimation processing means for inputting a target data set, which is a data set of the target work piece, into the model, performing estimation on the cause of a defect of the target work piece, and outputting a cause estimation result. ,
The alternative proposal device according to any one of claims 1 to 4, wherein the alternative proposal means uses the cause estimation result as data regarding the cause of the defect. The alternative proposal device according to claim 5 and
With at least one semiconductor manufacturing device
A manufacturing measurement means for measuring the state of the semiconductor manufacturing apparatus or the state of a workpiece being processed by the semiconductor manufacturing apparatus and generating the manufacturing measurement data.
A work piece measuring means for measuring the state of the work piece after processing with the semiconductor manufacturing apparatus and generating the work piece measurement data,
An environmental measurement means for measuring the state of the environment surrounding the semiconductor manufacturing apparatus and generating the environmental measurement data,
A semiconductor manufacturing system. Teacher data that correlates defect cause data related to the cause of defects found in a workpiece processed by a semiconductor manufacturing device with alternative data related to an alternative devised by the inventor in order to reduce or eliminate the defect. The first storage means to memorize,
A first machine learning that constructs an alternative proposal model showing a correspondence relationship between the defect cause data and the alternative data by executing machine learning based on the teacher data stored in the first storage means. Means and
As an alternative proposal means for outputting alternative data regarding the processing of the target work piece, which can reduce or eliminate defects by inputting data regarding the cause of the defect of the target work piece into the alternative proposal model.
A program to make a computer work. Teacher data that correlates defect cause data related to the cause of defects found in a workpiece processed by a semiconductor manufacturing device with alternative data related to an alternative devised by the inventor in order to reduce or eliminate the defect. The first storage stage, which is stored in the storage device,
A first machine learning that constructs an alternative proposal model showing a correspondence relationship between the defect cause data and the alternative data by executing machine learning based on the teacher data stored in the first storage means. Stages and
Includes an alternative proposal stage in which data on the cause of defects in the target work piece is input to the alternative proposal model to output alternative data on the processing of the target work piece, which can reduce or eliminate defects. , Alternative proposal method.

Images