CN113390856A

CN113390856A - Analysis device, analysis method, and analysis program

Info

Publication number: CN113390856A
Application number: CN202110240073.8A
Authority: CN
Inventors: 田中康基; 永井龙
Original assignee: Tokyo Electron Ltd
Current assignee: Tokyo Electron Ltd
Priority date: 2020-03-13
Filing date: 2021-03-04
Publication date: 2021-09-14
Also published as: TW202138943A

Abstract

The invention provides an analysis device, an analysis method and an analysis program for quantitatively evaluating the environment of a processing space of a manufacturing process based on a time series data set. The analysis device includes: a learning unit that performs machine learning using a time-series data set measured in association with processing of a target object in a processing space, and calculates a value indicating a correlation between time-series data in corresponding time ranges between measurement items; and an evaluation unit that evaluates the unknown environment of the processing space based on the value indicating the correlation calculated by the machine learning by the learning unit, using a time-series data set measured in association with the processing of the target object in the known environment of the processing space.

Description

Analysis device, analysis method, and analysis program

Technical Field

The present invention relates to an analysis device, an analysis method, and an analysis program.

Background

In general, when the environment (condition) changes in a processing space of a manufacturing process, the quality of a product when an object is processed in the processing space is affected. Therefore, it is important to maintain the quality of the product to grasp the environment of the processing space in advance when processing the object.

Meanwhile, in the manufacturing process, a set of various data (a data set of a plurality of kinds of time-series data, hereinafter referred to as a time-series data group) is acquired along with the processing of the object. In addition, the acquired time-series data group also includes time-series data related to the environment of the processing space.

Documents of the prior art

Patent document

Patent document 1: japanese patent laid-open publication No. 2010-219263

Disclosure of Invention

Technical problem to be solved by the invention

The invention provides an analysis device, an analysis method and an analysis program for quantitatively evaluating the environment of a processing space in a manufacturing process based on a time series data set.

Technical solution for solving technical problem

An analysis device according to an aspect of the present invention has, for example, the following configuration. Namely, comprising:

a learning unit that performs machine learning using a time-series data set measured in association with processing of a target object in a processing space, and calculates a value indicating a correlation between time-series data in corresponding time ranges between measurement items; and

an evaluation unit that evaluates an unknown environment of a processing space based on a value indicating the correlation calculated by the learning unit through machine learning, using a time-series data set measured in association with processing of an object in the processing space under a known environment

Effects of the invention

According to the present invention, it is possible to provide an analysis device, an analysis method, and an analysis program that quantitatively evaluate the environment of a processing space in a manufacturing process based on a time-series data set.

Drawings

Fig. 1 is a diagram 1 showing an example of a system configuration of the environment adjustment system.

Fig. 2 is a diagram showing an example of a semiconductor manufacturing process.

Fig. 3 is a diagram showing an example of a hardware configuration of the analyzer.

Fig. 4 is a diagram showing an example of data for learning.

Fig. 5 is a diagram showing an example of a time-series data group.

Fig. 6 is a view 1 showing a specific example of the processing performed by the learning unit.

Fig. 7 is a view 2 showing a specific example of the processing performed by the learning unit.

Fig. 8 is a view 1 showing a specific example of the processing performed by the evaluation unit.

Fig. 9 is a view 2 showing a specific example of the processing performed by the evaluation unit.

Fig. 10A is a flow chart 1 showing the flow of the environment adjustment processing.

Fig. 10B is a flow chart 2 showing the flow of the environment adjustment processing.

Fig. 11 is a diagram showing an example of a system configuration of the environment adjustment system when OES data is used.

Fig. 12 is a diagram showing an example of OES data.

Fig. 13 is a diagram showing a specific example of the process performed by the evaluation unit when OES data is used.

Fig. 14 is a graph showing a relationship between the total value of the intensity values showing the correlation between the wavelengths of the OES data and the emission intensity of each wavelength.

Fig. 15 is a diagram showing a specific example of the environment adjustment method when the environment is evaluated using OES data.

Fig. 16 is a diagram showing an example of a system configuration of the environment adjustment system when the process data set is used.

Fig. 17 is a diagram showing an example of a process data set.

Fig. 18 is a diagram showing a specific example of processing performed by the evaluation unit when the process data is used.

Fig. 19 is a diagram showing a specific example of an environment adjustment method when an environment is evaluated using process data.

Fig. 20 is a view 2 showing an example of a system configuration of the environment adjustment system.

Fig. 21 is a diagram 3 showing a specific example of the processing performed by the learning unit.

Fig. 22 is a diagram 3 showing a specific example of the processing performed by the evaluation unit.

Fig. 23 is a flow chart of fig. 3 showing the flow of the environment adjustment processing.

Fig. 24 is a diagram showing an example of the system configuration of the end point detection system.

Fig. 25 is a diagram 4 showing a specific example of the processing performed by the learning unit.

Fig. 26 is a diagram showing a specific example of the processing performed by the end point detecting unit.

Fig. 27 is a flowchart showing the flow of the end point detection processing.

Description of the reference numerals

100. 100', 100 ": environmental conditioning system

110: wafer before processing

120: treatment space

130: processed wafer

140_1 to 140_ n: time series data acquisition device

160: analysis device

161: learning part

162: evaluation unit

170: control device

610: regression model generation unit

620: regression model

621-623: node point

801: regression model generation unit

802: similarity calculation unit

1140: emission spectrum analysis device

1500: environmental tuning parameter decision table

1640_ 1-1640 _ n: process data acquisition device

1900: environmental tuning parameter decision table

2000: environmental conditioning system

2010: analysis device

2011: learning part

2012: evaluation unit

2101: regression model generation unit

2210: regression model execution unit

2220: count value calculation unit

2400: endpoint detection system

2410: analysis device

2411: learning part

2412: end point detection unit

2420: control device

2501: regression model generation unit

2610: regression model execution unit

2620: a count value calculation unit.

Detailed Description

Next, each embodiment will be described with reference to the drawings. In the present specification and the drawings, the same reference numerals are given to components having substantially the same functional configuration, and redundant description is omitted.

[ embodiment 1 ]

< System configuration of Environment adjustment System >

First, a system configuration of the environment adjustment system will be described. Fig. 1 is a diagram 1 showing an example of a system configuration of the environment adjustment system. As shown in FIG. 1, the environmental adjustment system 100 includes a semiconductor manufacturing process as an example of the manufacturing process, time series data acquisition devices 140_1 to 140_ n, an analysis device 160, and a control device 170.

In a semiconductor manufacturing process, an object (pre-process wafer 110) is processed in a predetermined processing space 120 to produce a product (post-process wafer 130). The pre-process wafer 110 is a wafer (substrate) before being processed in the processing space 120, and the post-process wafer 130 is a wafer (substrate) after being processed in the processing space 120.

The timing data acquisition devices 140_1 to 140_ n measure timing data associated with the processing of the pre-processed wafer 110 in the processing space 120, respectively. The time-series data acquisition devices 140_1 to 140_ n are devices for measuring different types of measurement items. In addition, the number of measurement items measured by each of the time-series data acquisition devices 140_1 to 140_ n may be 1 or more.

The time-series data sets measured by the time-series data acquisition devices 140_1 to 140_ n are stored as learning data in the learning data storage unit 163 of the analysis device 160.

The analysis device 160 is provided with an analysis program, and by executing the program, the analysis device 160 functions as a learning unit 161 and an evaluation unit 162.

The learning unit 161 uses the data measured by the time series data acquisition devices 140_1 to 140_ n,

The processing space 120 is a normal environment, and

use of a Standard protocol (predetermined specific protocol)

The time series data set (1 st learning data) measured when the wafer 110 before processing is processed is machine-learned. Thus, the learning unit 161 generates "1 st evaluation data" for quantitatively evaluating the environment of the processing space 120.

The learning unit 161 performs machine learning using each time-series data set when the pre-process wafer 110 is processed in the processing space 120 under a plurality of known environments (all of which are normal environments), and generates the 1 st evaluation data. The learning unit 161 stores each of the generated 1 st evaluation data in the evaluation data storage unit 164 as information indicating the corresponding environment.

The evaluation unit 162 uses the values measured by the time-series data acquisition devices 140_1 to 140_ n,

The processing space 120 is an unknown environment, and

use of standard protocols

The time series data set (data for 2 nd learning) measured when the wafer 110 before processing is processed is machine-learned to generate "data for 2 nd evaluation".

The evaluation unit 162 compares the 2 nd evaluation data with each 1 st evaluation data stored in the evaluation data storage unit 164, and determines which 1 st evaluation data the 2 nd evaluation data is similar to. Thus, the evaluation unit 162 evaluates the unknown environment of the processing space 120. Then, the evaluation unit 162 notifies the control device 170 of the evaluated environment.

The control device 170 adjusts the environment of the processing space 120 based on the environment evaluated by the evaluation unit 162 of the analysis device 160.

< processing space in semiconductor manufacturing Process >

Next, a predetermined processing space 120 of the semiconductor manufacturing process will be described. Fig. 2 is a diagram showing an example of a semiconductor manufacturing process. As shown in fig. 2, the semiconductor manufacturing process 200 includes a plurality of chambers as one example of the processing space. In the example of fig. 2, the semiconductor manufacturing process 200 has chambers of reference numerals 121 (name "chamber a") to 123 (name "chamber C"), and wafers before processing are processed in the respective chambers.

In the semiconductor manufacturing process 200, each chamber has the above-mentioned timing data acquisition devices 140_1 to 140_ n, and the timing data sets are measured in the respective chambers. Therefore, for example, the environment of the chamber B can be evaluated by comparing the 1 st evaluation data generated using the time-series data set measured in the chamber a with the 2 nd evaluation data generated using the time-series data set measured in the chamber B.

However, hereinafter, for simplification of description, a case of evaluating the environment using the 1 st and 2 nd evaluation data generated for the same chamber will be described. Further, hereinafter, a target chamber of the evaluation environment will be explained as the chamber a.

< hardware configuration of analyzing apparatus >

Next, the hardware configuration of the analyzer 160 will be described. Fig. 3 is a diagram showing an example of a hardware configuration of the analyzer. As shown in FIG. 3, the analyzer 160 includes a CPU (Central Processing Unit) 301, a ROM (Read Only Memory) 302, and a RAM (Random Access Memory) 303. The analysis device 160 also has a GPU (Graphics Processing Unit) 304. Further, the processor (Processing Circuit ) such as the CPU301 or the GPU304 and the memory such as the ROM302 or the RAM303 form a so-called computer.

The analysis device 160 further includes an auxiliary storage device 305, a display device 306, an operation device 307, an I/F (Interface) device 308, and a drive device 309. The hardware of the analysis device 160 is connected to each other via a bus 310.

The CPU301 is an arithmetic device that executes various programs (for example, analysis programs and the like) installed in the auxiliary storage device 305.

The ROM302 is a nonvolatile memory and functions as a main storage device. The ROM302 stores various programs, data, and the like necessary for the CPU301 to execute various programs installed in the auxiliary storage device 305. Specifically, the ROM302 stores boot programs (boot program) such as BIOS (Basic Input/Output System), EFI (Extensible Firmware Interface), and the like.

The RAM303 is a volatile Memory such as a DRAM (Dynamic Random Access Memory) or an SRAM (Static Random Access Memory), and functions as a main Memory device. The RAM303 provides a work area that is expanded when the CPU301 executes various programs installed in the auxiliary storage device 305.

The GPU304 is an arithmetic device for image processing, and in the present embodiment, when the CPU301 executes the analysis program, the GPU304 performs high-speed arithmetic operation by parallel processing on the time-series data group. The GPU304 is equipped with an internal memory (GPU memory) and temporarily holds information necessary for parallel processing of various time-series data sets.

The auxiliary storage device 305 stores various programs, various data used when the CPU301 executes the various programs, and the like. For example, the learning data storage unit 163 and the evaluation data storage unit 164 are implemented in the auxiliary storage device 305.

The display device 306 is a display device that displays the internal state of the analysis device 160. The operation device 307 is an input device used when the administrator of the analysis device 160 inputs various instructions to the analysis device 160. The I/F device 308 is a connection device for performing communication connected to a network not shown.

The drive device 309 is a device for setting the storage medium 320. The storage medium 320 referred to herein includes a medium for optically, electrically or magnetically storing information, such as a CD-ROM, a floppy disk, a magneto-optical disk, and the like. The storage medium 320 may include a semiconductor memory or the like for electrically storing information, such as a ROM, a flash memory, or the like.

The various programs installed in the auxiliary storage device 305 are installed by, for example, providing the storage medium 320 in which the programs are installed in the drive device 309, and reading the various programs stored in the storage medium 320 by the drive device 309. Alternatively, various programs installed in the auxiliary storage device 305 may be installed by downloading them via a network not shown.

< specific example of data for learning >

Next, the learning data read by the learning data storage unit 163 when the machine learning is performed by the learning unit 161 or the evaluation unit 162 will be described. Fig. 4 is a diagram showing an example of data for learning.

As shown in fig. 4, the data for learning includes "device", "lot number", "type of recipe", and "time series data group" as items of information. The "apparatus" stores the chamber name, and the "lot number" stores the lot number of each wafer before processing.

In addition, the name used for determining the recipe is saved in the "category of recipe". As described above, the learning data is a time-series data set when processed using the standard recipe, and therefore the "standard recipe" is stored in the "recipe type". In addition, the measured time-series data set is stored in the "time-series data set".

Fig. 4 (a) shows an example of the 1 st learning data. As shown in fig. 4 (a), the 1 st learning data 410_1, 410_2, and 410_3 … … respectively include time series data sets measured when the environment of the chamber a is environment 1, environment 2, and environment 3 … ….

On the other hand, fig. 4 (b) shows the 2 nd learning data. As shown in fig. 4 (b), the 2 nd learning data 420 includes a time series data set measured for evaluating the unknown environment of the chamber a.

< specific example of time series data set >

Next, a specific example of the time-series data set measured by the time-series data acquisition devices 140_1 to 140_ n will be described. Fig. 5 is a diagram showing an example of a time-series data group. In the example of fig. 5, for the sake of simplicity of explanation, the time-series data acquisition devices 140_1 to 140_ n each measure one-dimensional data, but two-dimensional data (a data set of plural kinds of one-dimensional data) may be measured by 1 time-series data acquisition device.

FIG. 5 (a) shows a time series data group consisting of time series data measured in the same time range by the time series data obtaining means 140_1 to 140_ n.

On the other hand, fig. 5 (b) shows a time series data group consisting of time series data measured in a corresponding time range by the time series data acquisition means 140_1 to 140_ n. As shown in fig. 5 (b), the learning data used for machine learning may include not only time-series data groups made up of time-series data measured in the same time range but also time-series data groups made up of time-series data measured in corresponding time ranges.

< specific example of processing by the learning section >

Next, a specific example of the processing performed by the learning unit 161 of the analysis device 160 will be described. Fig. 6 is a view 1 showing a specific example of the processing performed by the learning unit. As shown in fig. 6, the learning unit 161 includes a regression model generation unit 610.

The regression model is a machine learning model that extracts correlations between a plurality of time-series data comprehensively and at high speed from the plurality of time-series data, and is a model that represents correlations between a plurality of time-series data by a linear regression expression or a nonlinear regression expression. As an example of the regression model, a cross-correlation model can be given. In the case of the cross-correlation model, a time delay term in which the time difference between the plurality of time-series data is considered may be included.

In general, a regression model applied to a manufacturing process monitors the time and position of a change in correlation of time series data, and detects an abnormality occurring in the manufacturing process.

In contrast, in the analysis device 160 according to the present embodiment, a regression model is used to quantitatively evaluate the environment of the processing space.

Specifically, the regression model generation unit 610 performs machine learning on the regression model using the time series data group included in the 1 st learning data stored in the learning data storage unit 163. Thus, the regression model generation unit 610 calculates a value (an example of a value indicating the correlation) indicating the strength of the correlation between the time-series data of the measurement items measured by the time-series data acquisition devices 140_1 to 140 — n.

In the example of fig. 6, for the sake of simplicity of explanation, a value indicating the strength of the correlation of time series data in the same time range among 3 measurement items is calculated. Specifically, the regression model generation unit 610 is shown to be inputted with the input

Time-series data 1 measured by the time-series data acquisition device 140_1,

Time-series data 2 measured by the time-series data acquisition device 140_2,

Time-series data 3 measured by the time-series data acquisition device 140_3

In contrast, the regression model 620 is machine-learned to calculate a value representing the strength of the correlation of the time-series data.

In the regression model 620, the node 621 corresponds to the time-series data acquisition device 140_1, and the node 622 corresponds to the time-series data acquisition device 140_ 2. Further, the node 623 corresponds to the timing data acquisition means 140_ 3.

According to the example of fig. 6, the value indicating the strength of the correlation between the time series data 1 and the time series data 2 is "F12", and the value indicating the strength of the correlation between the time series data 1 and the time series data 3 is "F13". Further, a value indicating the strength of the association between the time-series data 2 and the time-series data 3 is "F23".

In addition, in the regression model generation unit 610, for example, in the chamber a, the same machine learning is performed for each regression model for each environment using the 1 st learning data stored for each environment.

Fig. 7 is a 2 nd view showing a specific example of the processing performed by the learning unit, and shows a case where the 1 st evaluation data is generated by the regression model generation unit 610. As shown in fig. 7, the 1 st evaluation data includes "1 st node" (1 st measurement item), "2 nd node" (2 nd measurement item), and "strength of correlation" as items of information.

The "1 st node" and the "2 nd node" store time series data for calculating a value indicating the strength of the correlation, from among the time series data groups included in the 1 st learning data.

The "strength of association" stores a value indicating the strength of association between the time-series data stored in the "1 st node" and the time-series data stored in the "2 nd node".

Further, as shown in fig. 7, the 1 st evaluation data is generated for each environment. In the example of fig. 7, the 1 st evaluation data 710_1 is evaluation data represented as environment 1, and the 1 st evaluation data 710_2 is evaluation data represented as environment 2. The 1 st evaluation data 710_3 is evaluation data represented as environment 3.

The 1 st evaluation data 710_1, 710_2, and 710_3 … … generated by the regression model generation unit 610 are stored in the evaluation data storage unit 164 as information indicating different environments.

< specific example 1 of the treatment by the evaluation section >

Next, specific example 1 of the processing performed by the evaluation unit 162 of the analyzer 160 will be described. Fig. 8 is a view 1 showing a specific example of the processing performed by the evaluation unit. As shown in fig. 8, the evaluation unit 162 includes a regression model generation unit 801 and a similarity calculation unit 802.

The regression model generation unit 801 performs machine learning on the regression model using the time series data group included in the 2 nd learning data stored in the learning data storage unit 163. Thus, the regression model generation unit 801 generates a regression model, and calculates a value indicating the strength of the correlation between the measurement items between the plurality of time-series data measured by the time-series data acquisition devices 140_1 to 140 — n.

As a result, the regression model generation unit 801 generates the 2 nd evaluation data 820. As shown in fig. 8, the 2 nd evaluation data 820 has items of the same information as the 1 st evaluation data 710_1, 710_2, and 710_3 … ….

The similarity calculation unit 802 calculates the similarity between the 2 nd evaluation data 820 generated by the regression model generation unit 801 and the 1 st evaluation data 710_1, 710_2, and 710_3 … … stored in the evaluation data storage unit 164.

Specifically, the similarity calculation unit 802 compares the values of the strength of the correlation between the 1 st evaluation data and the 2 nd evaluation data, which indicates that all the measurement items of the time series data of the "1 st node" and the "2 nd node" are identical, and calculates the similarity.

For example, the similarity calculation unit 802 compares:

a value "f 12" indicating the strength of association when the measurement item of the time-series data of the "1 st node" of the 2 nd evaluation data 820 is "time-series data 1" and the measurement item of the time-series data of the "2 nd node" is "time-series data 2", and

a value "F12" indicating the strength of the correlation when the measurement item of the time-series data of the "1 st node" of the 1 st evaluation data 710_1 is "time-series data 1" and the measurement item of the time-series data of the "2 nd node" is "time-series data 2".

Similarly, the similarity calculation unit 802 compares:

a value "f 13" indicating the strength of association when the measurement item of the time-series data of the "1 st node" in the 2 nd evaluation data 820 is "time-series data 1" and the measurement item of the time-series data of the "2 nd node" is "time-series data 3", and

a value "F13" indicating the strength of the correlation when the measurement item of the time-series data of the "1 st node" of the 1 st evaluation data 710_1 is "time-series data 1" and the measurement item of the time-series data of the "2 nd node" is "time-series data 3".

The similarity calculation unit 802 compares the values of the strength of the correlation for all combinations in the 2 nd evaluation data 820 to calculate the similarities with the 1 st evaluation data 710_1, 710_2, and 710_3 … …, respectively.

The similarity calculation unit 802 evaluates the 1 st evaluation data determined to have the highest calculated similarity as the environment of the chamber a when the time series data group included in the 2 nd learning data 420 is measured.

For example, when the similarity with the 1 st evaluation data 710_1 is the greatest, the environment of the chamber a when the time series data group included in the 2 nd learning data 420 is measured is evaluated as "environment 1" in the similarity calculation unit 802.

As described above, in the analysis device 160 according to the present embodiment, instead of individually analyzing the characteristics of time-series data, a value indicating the strength of the correlation of the time-series data is calculated to evaluate the environment.

Thus, according to the analysis device 160 of the present embodiment, it is possible to appropriately capture a fine change in time series data accompanying a change in the environment of the chamber. As a result, the analysis device 160 according to the present embodiment can accurately evaluate the environment of the chamber based on the time-series data set.

< specific example 2 of the treatment by the evaluation section >

Next, specific example 2 of the processing performed by the evaluation unit 162 of the analyzer 160 will be described. Fig. 9 is a view 2 showing a specific example of the processing performed by the evaluation unit. In the case of fig. 9, the similarity calculation unit 802 calculates the similarity by summing up the 1 st evaluation data and the 2 nd evaluation data for each measurement item of the 1 st node time series data, and evaluates the environment.

In fig. 9, a graph 910_1 is a graph obtained by summing 1 st evaluation data 710_1 for each measurement item of the 1 st node, the horizontal axis represents the measurement item of the time series data of the 1 st node, and the vertical axis represents the total value of the values representing the strength of the correlation.

For example, the total value of the values representing the strength of the correlation corresponding to "time series data 1" of the graph 910_1 will be

A value "F12" indicating the strength of correlation when the measurement item of the time series data of the "1 st node" of the 1 st evaluation data 710_1 is "time series data 1" and the measurement item of the time series data of the "2 nd node" is "time series data 2", and

a value "F13" … … indicating the strength of correlation when the measurement item of the time series data of the "1 st node" in the 1 st evaluation data 710_1 is "time series data 1" and the measurement item of the time series data of the "2 nd node" is "time series data 3

And the like are added up.

Similarly, in fig. 9, a graph 920 is a graph obtained by summing up 2 nd evaluation data for each measurement item of the 1 st node time-series data, the horizontal axis represents the measurement item of the 1 st node time-series data, and the vertical axis represents the total value of the values representing the strength of the correlation (another example of the values representing the correlation).

For example, the total value of the values indicating the strength of the correlation corresponding to "time series data 1" of graph 920 is to be

A value "f 12" indicating the strength of the correlation when the measurement item of the time-series data of the "1 st node" of the 2 nd evaluation data 820 is "time-series data 1" and the measurement item of the time-series data of the "2 nd node" is "time-series data 2",

a value "f 13" … … indicating the strength of the correlation when the measurement item of the time series data of the "1 st node" of the 2 nd evaluation data 820 is "time series data 1" and the measurement item of the time series data of the "2 nd node" is "time series data 3

And the like are added up.

In the case of fig. 9, the similarity calculation unit 802 compares the graph 920 with the graphs 910_1, 910_2, and 910_3 … … to calculate the similarity. In the case of fig. 9, the similarity calculation unit 802 evaluates the environment corresponding to the graph with the highest calculated similarity as the environment of the chamber a when the time series data group included in the 2 nd learning data 420 is measured.

For example, when the similarity to the graph 910_1 is the maximum, the similarity calculation unit 802 evaluates the environment of the chamber a when the time series data group included in the 2 nd learning data 420 is measured as "environment 1".

As described above, in the analysis device 160 according to the present embodiment, instead of analyzing the characteristics of the time-series data individually, the total value of the values indicating the strength of the correlation of the time-series data is calculated to evaluate the environment.

< flow of Environment adjustment processing >

Next, the flow of the entire environment adjustment process performed by the environment adjustment system 100 will be described. In the atmosphere adjustment process performed by the atmosphere adjustment system 100, any of the following 2 adjustment methods is included as an adjustment method when the controller 170 adjusts the atmosphere of the chamber.

A method of adjusting the recipe in real time based on the environment evaluated in real time during the processing of the wafer before processing.

A method of evaluating the environment after the treatment of the wafer before the treatment is completed and adjusting the environment to a constant environment by cleaning or the like.

Therefore, the flow of the environment adjustment process including each adjustment method will be described below.

(1) Environment adjustment process including method of adjusting a scenario in real time

Fig. 10A is a flow chart 1 showing a flow of the environment adjustment processing, and is a flow chart showing a flow of the environment adjustment processing including a method of adjusting a scenario in real time based on an evaluated environment.

In step S1001A, the time-series data acquisition devices 140_1 to 140 — n measure time-series data sets accompanying the processing of the wafers before processing in the chamber a, and store the time-series data sets in the learning data storage unit 163. In addition, the time series data acquisition devices 140_1 to 140_ n store time series data sets measured in association with the processing of the chamber a in a plurality of known environments as 1 st learning data 410_1, 410_2, 410_3 … ….

In step S1002A, the learning unit 161 of the analysis device 160 performs machine learning on each regression model using the 1 st learning data 410_1, 410_2, and 410_3 … … stored in the learning data storage unit 163. The learning unit 161 of the analysis device 160 generates the 1 st evaluation data 710_1, 710_2, and 710_3 … … using the values indicating the strength of the correlation calculated when each regression model is machine-learned.

In step S1003A, the time-series data acquisition devices 140_1 to 140 — n measure time-series data sets (in a predetermined time range) associated with the processing of the wafers before processing in the chamber a, and store the time-series data sets in the learning data storage unit 163. In addition, the time series data acquisition devices 140_1 to 140_ n store time series data sets (within a predetermined time range) measured in association with the processing of the chamber a in the unknown environment as the 2 nd learning data 420.

In step S1004A, the time-series data acquisition devices 140_1 to 140 — n determine whether a predetermined adjustment period (for example, 1 second) has elapsed. If it is determined in step S1004A that the predetermined adjustment period has not elapsed (No in step S1004A), the system waits until the predetermined adjustment period has elapsed. On the other hand, if it is determined in step S1004A that the predetermined adjustment period has elapsed (Yes in step S1004A), the process proceeds to step S1005A.

In step S1005A, the evaluation unit 162 of the analysis device 160 performs machine learning on the regression model using the 2 nd learning data 420 (data in a predetermined time range before the time point at which the predetermined adjustment cycle has elapsed) stored in the learning data storage unit 163. The evaluation unit 162 of the analysis device 160 generates the 2 nd-evaluation data 820 (in a predetermined time range) using the value indicating the strength of the correlation calculated when the regression model is machine-learned.

In step S1006A, the evaluation unit 162 of the analysis device 160 determines the 1 st evaluation data having the greatest similarity to the 2 nd evaluation data 820 among the 1 st evaluation data 710_1, 710_2, and 710_3 … …. Alternatively, the evaluation unit 162 of the analysis device 160 determines the graph with the highest similarity to the graph 920 calculated based on the 2 nd evaluation data 820 among the graphs 910_1, 910_2, and 910_3 … … calculated based on the plurality of 1 st evaluation data. Thus, the evaluation unit 162 of the analyzer 160 evaluates the unknown environment of the chamber a.

In step S1007A, the controller 170 performs the pre-process wafer processing using the recipe corresponding to the evaluated environment.

In step S1008A, it is determined whether or not the processing of the wafer before processing has ended, and if it is determined that the processing has not ended (no in step S1008A), the process returns to step S1003A. On the other hand, if it is determined in step S1008A that the environment adjustment process is ended (yes in step S1008A), the environment adjustment process is ended.

(2) Environment adjustment process including method of adjusting to certain environment

Fig. 10B is a flow chart 2 showing a flow of the environment adjustment processing, and is a flow chart of the environment adjustment processing including a method of adjusting the evaluated environment to a constant environment by cleaning or the like.

In step S1001B, the time-series data acquisition devices 140_1 to 140 — n measure time-series data sets accompanying the processing of the wafers before processing in the chamber a, and store the time-series data sets in the learning data storage unit 163. In addition, the time series data acquisition devices 140_1 to 140_ n store time series data sets measured in association with the processing of the chamber a in a plurality of known environments as 1 st learning data 410_1, 410_2, 410_3 … ….

In step S1002B, the learning unit 161 of the analysis device 160 performs machine learning on each regression model using the 1 st learning data 410_1, 410_2, and 410_3 … … stored in the learning data storage unit 163. The learning unit 161 of the analysis device 160 generates the 1 st evaluation data 710_1, 710_2, and 710_3 … … using the values indicating the strength of the correlation calculated when each regression model is machine-learned.

In step S1003B, the time-series data acquiring devices 140_1 to 140 — n measure time-series data sets along with the processing of the wafers before processing in the chamber a, and store the time-series data sets in the learning data storage unit 163. In addition, the time-series data acquisition devices 140_1 to 140_ n store time-series data sets (time-series data sets up to the end of processing) measured in association with processing in an unknown environment of the chamber a as the 2 nd learning data 420.

In step S1004B, the evaluation unit 162 of the analysis device 160 performs machine learning on the regression model using the 2 nd learning data 420 stored in the learning data storage unit 163. The evaluation unit 162 of the analysis device 160 generates the 2 nd evaluation data 820 using the value indicating the strength of the correlation calculated when the regression model is machine-learned.

In step S1005B, the evaluation unit 162 of the analysis device 160 determines the 1 st evaluation data having the greatest similarity to the 2 nd evaluation data among the 1 st evaluation data 710_1, 710_2, and 710_3 … …. Alternatively, the evaluation unit 162 of the analysis device 160 determines the graph with the highest similarity to the graph 920 calculated based on the 2 nd evaluation data among the graphs 910_1, 910_2, and 910_3 … … calculated based on the plurality of 1 st evaluation data. Thus, the evaluation unit 162 of the analyzer 160 evaluates the unknown environment of the chamber a.

In step S1006B, control device 170 adjusts the environment to a constant level by cleaning or the like.

< summary >

From the above description, it is clear that, in the analysis device according to embodiment 1,

the machine learning is performed using the time series data group measured along with the processing of the wafer before processing in the chamber, and a value indicating the strength of the correlation of the time series data in the corresponding time range between the measurement items is calculated.

Using each time-series data set measured along with the processing of the pre-processing wafer in the plurality of known environments of the chamber, the unknown environment of the chamber is evaluated based on the value representing the strength of the correlation calculated by performing the machine learning.

Thus, the analysis device according to embodiment 1 can quantitatively evaluate the environment of the chamber in the semiconductor manufacturing process based on the time-series data group.

[ 2 nd embodiment ]

In the above-described embodiment 1, a specific example of the time-series data acquisition device and the time-series data group is not mentioned. In embodiment 2, the case where the time-series data acquisition device is an Emission spectrum analysis device and the time-series data set is OES (Optical Emission Spectroscopy) data will be described. The OES data is a data set including time series data of emission intensities corresponding to the types of wavelengths.

Here, OES data is known to be related to the kind of deposit attached in the chamber, and the amount of deposit. Therefore, by using OES data as a time series data set, the environment of the chamber can be evaluated from the viewpoint of the kind of deposit and the amount of deposit adhering to the inside of the chamber. Hereinafter, embodiment 2 will be described mainly focusing on differences from embodiment 1.

< System configuration of Environment adjustment System >

First, a system configuration of the environment adjustment system will be described. Fig. 11 is a diagram showing an example of a system configuration of the environment adjustment system when OES data is used. The difference from fig. 1 is that: an emission spectrum analysis device 1140 is provided as the time-series data acquisition device; OES data is stored as a time series data set in the learning data storage unit 163; the cleaning protocol is used to adjust to a certain environment.

The emission spectroscopy apparatus 1140 measures OES data with the processing of the pre-process wafer 110 in chamber a by an emission spectroscopy technique. The OES data is, for example, time series data indicating emission intensity at each time for each wavelength included in a wavelength range of visible light.

The cleaning recipe is a recipe used for cleaning the inside of the chamber a, and is a recipe for adjusting the environment of the chamber a to a constant environment from the viewpoint of the kind of deposits adhering to the inside of the chamber a and the amount of deposits.

< specific examples of OES data >

Next, a specific example of OES data measured by the emission spectrum analyzer 1140 will be described. FIG. 12 is a graph showing an example of OES data, and shows an emission intensity data set at each wavelength included in a wavelength range (400[ nm ] to 800[ nm ]) of visible light measured on a 1[ nm ] scale. In fig. 12, the horizontal axis represents time, and the vertical axis represents emission intensity of each wavelength.

In the case of fig. 12, for example, the graph of the uppermost layer shows emission intensity data at each time with a wavelength of 400[ nm ], and the graph of the second layer shows emission intensity data at each time with a wavelength of 401[ nm ]. The graph of the third layer in fig. 12 shows emission intensity data at each time having a wavelength of 402[ nm ].

< specific example of treatment by evaluation section >

Next, a specific example of the processing performed by the evaluation unit 162 when OES data is used will be described. Fig. 13 is a diagram showing a specific example of the process performed by the evaluation unit when OES data is used.

As shown in fig. 13, when OES data is used as the time series data group, the emission intensity data of each wavelength is stored in the time series data of the "1 st node" and the "2 nd node" of the 1 st evaluation data 710_1 ', 710_2 ', 710_3 ' … ….

Further, the 1 st evaluation data 710_ 1' shows:

a value indicating the correlation between the emission intensity data at each time having a wavelength of 400[ nm ] and the emission intensity data at each time having a wavelength of 401[ nm ] is "F12",

a value indicating the correlation between the emission intensity data at each time with the wavelength of 400[ nm ] and the emission intensity data at each time with the wavelength of 402[ nm ] is "F13",

a value indicating the correlation between the emission intensity data at each time with the wavelength of 401[ nm ] and the emission intensity data at each time with the wavelength of 402[ nm ] is "F23".

As shown in fig. 13, when OES data is used as the time series data group, the horizontal axes of the graphs 910_1 ', 910_2 ', and 910_3 ' … … represent wavelengths included in the visible light wavelength range (400 nm to 800 nm).

Similarly, when OES data is used as the time series data set, the emission intensity data of each wavelength is stored in the time series data of the "1 st node" and the "2 nd node" of the 2 nd evaluation data 820'.

The data 820' for evaluation 2 shows:

a value indicating the correlation between the emission intensity data at each time having a wavelength of 400[ nm ] and the emission intensity data at each time having a wavelength of 401[ nm ] is "f 12",

a value indicating the correlation between the emission intensity data at each time having the wavelength of 400[ nm ] and the emission intensity data at each time having the wavelength of 402[ nm ] is "f 13",

the value indicating the correlation between the emission intensity data at each time with the wavelength of 401[ nm ] and the emission intensity data at each time with the wavelength of 402[ nm ] is "f 23".

In addition, when OES data is used as the time series data group, the horizontal axis of the graph 920' represents the wavelengths included in the visible light wavelength range (400 nm to 800 nm).

The similarity calculation unit 802 calculates the similarity by the same calculation method as in embodiment 1 described above, and evaluates the environment by the same evaluation method.

However, although not mentioned in embodiment 1, the similarity between the 2 nd evaluation data 820 'and the 1 st evaluation data 710_ 1', 710_2 ', 710_ 3' … … may be low. Alternatively, it may happen that the graphs 920 'are all less similar to the graphs 910_ 1', 910_2 ', 910_ 3' … ….

In such a case, the evaluation unit 162 of the analyzer 160 determines that the environment of the chamber a is abnormal from the viewpoint of the type of deposit and the amount of deposit adhering to the inside of the chamber a. That is, the evaluation unit 162 of the analyzer 160 can determine whether or not the environment of the chamber a corresponds to any one of the predetermined environments, and can also determine whether or not the environment is normal.

< relationship between the total value of the intensity values representing the correlation between the wavelengths of the OES data and the emission intensity of each wavelength >

Next, a relationship between a total value of intensity values indicating a correlation between wavelengths of OES data and emission intensity of each wavelength will be described. Fig. 14 is a graph showing a relationship between the total value of the intensity values indicating the correlation between the wavelengths of the OES data and the emission intensity of each wavelength.

In fig. 14, a graph 1410 shows a total value of values representing the intensity of the correlation between the wavelengths. On the other hand, graph 1420 shows the maximum light emission intensity at each wavelength.

As is clear from a comparison between graph 1410 and graph 1420, the total value of the intensity values indicating the correlation between the wavelengths is low at the wavelength at which the emission intensity becomes the peak.

Here, the learning unit 161 uses emission intensity data of a wavelength having a large total value of the intensity values indicating the correlation between the wavelengths when evaluating the environment of the chamber a. In other words, the learning unit 161 evaluates the environment using emission intensity data of a wavelength whose emission intensity is not a peak. This is a general evaluation method for evaluating the environment of the chamber using OES data, and is largely different from an evaluation method using emission intensity data of a wavelength at which the emission intensity is a peak.

That is, the analyzer 160 of the present embodiment can evaluate the environment of the chamber a by an evaluation method different from the conventional method when evaluating the environment of the chamber a using OES data.

< specific example of Environment adjustment method >

Next, a specific example of the environment adjustment method will be described. Fig. 15 is a diagram showing a specific example of the environment adjustment method when evaluating the environment using OES data. As described above, when the environment is evaluated using the OES data, the cleaning recipe is adjusted to a certain environment by the control device 170.

At this time, the control device 170 refers to the environment adjustment parameter determination table 1500 shown in fig. 15. As shown in fig. 15, the environmental adjustment parameter determination table 1500 includes "current environment", "recipe corresponding to environment", and "cleaning recipe" as items of information.

The "current environment" stores information indicating the current environment of the chamber a, which is output from the evaluation unit 162 of the analyzer 160.

The "recipe corresponding to the environment" stores a recipe corresponding to the current environment of the chamber a, which is used when the environment of the chamber a is adjusted.

The "cleaning recipe" is a prescribed cleaning recipe used for cleaning the inside of the chamber a.

The environment of the chamber a is adjusted by the control device 170 using a recipe corresponding to the current environment that has been evaluated, and then the interior of the chamber a is cleaned using a predetermined cleaning recipe, thereby adjusting the interior of the chamber a to a constant environment.

However, the method of adjusting to a certain environment using the cleaning protocol is not limited thereto. For example, the process of cleaning the inside of the chamber a using a predetermined cleaning recipe may be a process of adjusting the environment of the chamber a using a recipe corresponding to the environment.

Specifically, instead of performing the process of adjusting the environment of the chamber a using the recipe corresponding to the environment, the process time of the process of cleaning the inside of the chamber a may be adjusted using a predetermined cleaning recipe to adjust the environment to a constant environment.

< summary >

From the above description, it is clear that, in the analysis device according to embodiment 2,

the OES data measured in association with the processing of the wafer before processing in the chamber is used to perform machine learning, and a value of the intensity indicating the correlation of the emission intensity data in the corresponding time range between the wavelengths is calculated.

Using the OES data measured in association with the processing of the pre-processed wafer in the plurality of known environments of the chamber, the unknown environment of the chamber is evaluated based on the value representing the strength of the correlation calculated by performing the machine learning.

Thus, according to the analysis device of embodiment 2, the environment of the chamber in the semiconductor manufacturing process can be quantitatively evaluated from the viewpoint of the kind of the deposit attached to the chamber and the amount of the deposit based on the OES data.

In embodiment 2, the case where the time-series data acquisition device is an emission spectrum analysis device and the time-series data group is OES data has been described, but the time-series data acquisition device may be a mass spectrometry device (for example, a quadrupole mass spectrometry device). In this case, the time-series data group is a data set of the time-series data (mass spectrometry data) of the measured intensities by the number corresponding to the number of kinds of the values (m/z values) regarding the mass.

[ embodiment 3 ]

In the above-described embodiment 2, a case where the time series data group is OES data is described. However, the time series data set is not limited to OES data, and may be, for example, process data sets (RF power data, pressure data, temperature data … …, etc.) measured by various process sensors.

Here, it is known that process data has a correlation with the degree of consumption (or deterioration) of each component in the chamber. Therefore, by using the process data set as the time-series data set, the environment of the chamber can be evaluated from the viewpoint of the degree of consumption (or the degree of deterioration) of each component in the chamber. Hereinafter, embodiment 3 will be described mainly focusing on differences from embodiment 1 or embodiment 2.

< System configuration of Environment adjustment System >

First, a system configuration of the environment adjustment system will be described. Fig. 16 is a diagram showing an example of a system configuration of the environment adjustment system when the process data set is used. The difference from fig. 1 is that process data acquiring devices 1640_1 and 1640_2 … … 1640_ n are arranged as time series data acquiring devices. Further, the process data set is stored in the learning data storage unit 163 as a time series data set, and is adjusted to a constant environment using the position data of the focus ring and the like.

The process data obtainers 1640_1, 1640_2 … … 1640_ n measure process data sets accompanying the processing of the pre-process wafer 110 in the corresponding chamber a. The process data group includes, for example, RF power data, pressure data, gas flow rate data, current data, GAP length data, temperature data, and the like at each time.

The position data of the focus ring and the like are changed position data when the position of the focus ring in the height direction is changed based on the degree of wear of the focus ring, which is an example of a member in the chamber a. The position data of the focus ring and the like are data for adjusting the environment in the chamber a to a constant environment.

< specific example of Process data set >

Next, a specific example of the process data set measured by the process data acquiring apparatuses 1640_1 and 1640_2 … … 1640_ n will be described. Fig. 17 is a diagram showing an example of a process data set. The example of fig. 17 shows a case where the process data acquiring device 1640_1 measures RF power supply data as process data 1 and the process data acquiring device 1640_2 measures pressure data as process data 2. In addition, the example of fig. 17 shows a case where the process data acquirer 1640_3 measures gas flow rate data as the process data 3.

Similarly, the example of FIG. 17 shows a case where the process data acquiring device 1640_ n-2 measures current data as the process data n-2 and the process data acquiring device 1640_ n-1 measures GAP length data as the process data n-2. In addition, the example of fig. 17 shows a case where the process data acquiring device 1640_ n measures temperature data as the process data n.

< specific example of treatment by evaluation section >

Next, a specific example of the processing performed by the evaluation unit 162 when the process data set is used will be described. Fig. 18 is a diagram showing a specific example of processing performed by the evaluation unit when the process data set is used.

As shown in fig. 18, when the process data set is used as the time-series data set, the process data of each measurement item is stored in the time-series data of the "1 st node" and the "2 nd node" of the 1 st evaluation data 710_1 ", 710_ 2", 710_3 "… ….

In addition, the 1 st evaluation data 710_1 ″ shows:

a value representing the strength of the correlation of the process data 1 with the process data 2 is "F12",

a value representing the strength of the correlation of the process data 1 with the process data 3 is "F13",

the value indicating the strength of the correlation between the process data 2 and the process data 3 is "F23".

As shown in fig. 18, when the process data set is used as the time-series data set, the horizontal axes of the graphs 910_1 ", 910_ 2", and 910_3 "… … represent the measurement items.

Similarly, when the process data group is used as the time-series data group, the process data of each measurement item is stored in the time-series data of the "1 st node" and the "2 nd node" of the 2 nd evaluation data 820 ″.

The 2 nd evaluation data 820 ″ shows:

the value representing the strength of the correlation of the process data 1 with the process data 2 is "f 12",

the value representing the strength of the correlation of the process data 1 with the process data 3 is "f 13",

the value indicating the strength of the correlation between the process data 2 and the process data 3 is "f 23".

As shown in fig. 18, when the process data set is used as the time-series data set, the horizontal axis of the graph 920 ″ represents each measurement item.

However, although not mentioned in embodiment 1, the similarity between the 2 nd evaluation data 820 "and the 1 st evaluation data 710_ 1", 710_2 ", and 710_ 3" … … may be low. Alternatively, a situation may occur where the graphs 920 "are all less similar to the graphs 910_ 1", 910_2 ", 910_ 3" … ….

In such a case, the evaluation unit 162 of the analyzer 160 determines that the environment of the chamber a is abnormal from the viewpoint of the degree of consumption of the components in the chamber a. In other words, the evaluation unit 162 of the analyzer 160 can determine whether the environment of the chamber a corresponds to any one of the predetermined environments, and can also determine whether the environment is normal.

< specific example of Environment adjustment method >

Next, a specific example of the environment adjustment method will be described. Fig. 19 is a diagram showing a specific example of an environment adjustment method when evaluating an environment using a process data set. As described above, when the environment is evaluated using the process data set, the controller 170 adjusts the environment to a constant environment using the position data of the focus ring and the like.

At this time, the controller 170 refers to the environment adjustment parameter determination table 1900 shown in fig. 19. As shown in fig. 19, the environment adjustment parameter determination table 1900 includes "current environment", "focus ring position", and "applied voltage" as items of information.

The "focus ring position" stores changed position data when the position of the focus ring in the height direction is changed according to the evaluated environment (the degree of wear of each member).

The "applied voltage" holds applied voltage data in the case where a voltage is applied instead of changing the position of the focus ring.

When the information indicating the current environment is notified from the analyzer 160, the controller 170 refers to the environment adjustment parameter determination table 1900, and determines the position data of the focus ring when the position of the focus ring can be changed. In addition, when the position of the focus ring cannot be changed, the control device 170 determines the applied voltage data. Further, the control device 170 adjusts the chamber a to a constant environment using the determined position data or voltage data.

In the example of fig. 19, for example, when the current environment is "environment 1", the control device 170 determines that the focus ring position is "position 1" or the applied voltage is "DC 1".

< summary >

From the above description, it is clear that, in the analysis apparatus according to embodiment 3,

the machine learning is performed using a process data set measured in association with the processing of the wafer before processing in the chamber, and a value indicating the strength of the correlation of the process data in the corresponding time range between the measurement items is calculated.

The unknown environment of the chamber is evaluated based on the value representing the strength of the correlation calculated by performing the machine learning, using each process data set measured along with the processing of the pre-processed wafer in the plurality of known environments of the chamber.

Thus, according to the analysis apparatus of embodiment 3, the environment of the chamber in the semiconductor manufacturing process can be quantitatively evaluated from the viewpoint of the degree of consumption of each component in the chamber based on the process data set.

[ 4 th embodiment ]

In the above-described embodiments 2 and 3, the description has been given of the case where the similarity of the 2 nd evaluation data with respect to the plurality of 1 st evaluation data is low, and it is determined that the environment of the chamber is abnormal. That is, in the above-described embodiments 2 and 3, the description has been made as to whether or not the time-series data deviates from the normal environment is determined based on the value indicating the strength of the correlation of the time-series data.

However, the value used for determining whether or not the environment of the chamber is normal is not limited to the value representing the strength of the correlation of the time series data. For example, the count value may be a predetermined count value that can be calculated by executing a regression model. Hereinafter, the 4 th embodiment will be described mainly focusing on differences from the 1 st to 3 rd embodiments.

< System configuration of Environment adjustment System >

First, a system configuration of the environment adjustment system will be described. Fig. 20 is a view 2 showing an example of a system configuration of the environment adjustment system. The difference from fig. 1 is that: in the environment adjustment system 2000, the function of the learning unit 2011 of the analysis device 2010 is different from the function of the learning unit 161; the function of the evaluation unit 2012 is different from that of the evaluation unit 162. In the environment adjustment system 2000, the analysis device 2010 does not have an evaluation data storage unit.

The learning unit 2011 performs machine learning on the regression model using the data for learning.

The evaluation unit 2012 calculates a predetermined count value by inputting a time series data set (estimation data) measured in an unknown environment to a regression model generated by machine learning by the learning unit 161 using the learning data.

The evaluation unit 2012 counts the number of combinations of the 1 st node and the 2 nd node, the combinations of the 1 st node and the 2 nd node having a predetermined relevance. The predetermined count value is the number of combinations in which the predetermined association is broken (the value indicating the strength of the association is equal to or less than a predetermined threshold) among combinations in which the 1 st node and the 2 nd node have the predetermined association.

When calculating a predetermined count value, the evaluation unit 2012 first acquires a regression model generated by the learning unit 2011 through machine learning using a time-series data set (learning data) measured in a normal environment. Next, the evaluation unit 2012 inputs a time series data set (estimation data) measured in an unknown environment to the obtained regression model, and calculates a predetermined count value. Thus, according to the evaluation unit 2012, it is possible to determine whether or not the environment of the chamber a is normal (or the degree of abnormality of the chamber a).

Further, based on the information indicating the environment (whether or not the chamber a is normal (or the degree of abnormality of the chamber a)) output from the evaluation unit 2012, it is determined that, for example:

whether maintenance of chamber A is required, whether maintenance of parts affecting chamber A is required, or

The timing of maintenance of the chamber a, the timing of maintenance of components that affect the chamber a, and the like.

< specific example of processing by the learning section >

Next, a specific example of the processing performed by the learning unit 2011 of the analysis device 2010 is described. Fig. 21 is a diagram 3 showing a specific example of the processing performed by the learning unit. As shown in fig. 21, the learning unit 2011 includes a regression model generation unit 2101.

The regression model generation unit 2101 performs machine learning on the regression model using the time series data group included in the learning data stored in the learning data storage unit 163. Thus, the regression model generation unit 2101 defines the relationship of the time-series data between the measurement items measured by the time-series data acquisition devices 140_1 to 140 — n by using the mathematical expression shown by reference numeral 2110.

Specifically, the regression model generation unit 2101 calculates each parameter of the equation shown by reference numeral 2110 so that the 1 st node time-series data is input to the equation shown by reference numeral 2110 to derive the 2 nd node time-series data.

In the mathematical expression shown by reference numeral 2110, it is represented by

T: the time of day is,

m: auto-correlation (a parameter indicating whether periodicity is present),

n: cross-correlation (a parameter indicating whether or not they are correlated),

k: the time delay is set to a time when the time delay,

β, α, and C represent predetermined coefficients.

In fig. 21, a learning result 2120 shows each parameter of the mathematical expression shown by reference numeral 2110 calculated by machine learning on the regression model. Specifically, the learning result 2120 includes "1 st node" and "2 nd node" as items of information, and "autocorrelation", "cross-correlation", and "time lag" as other examples of the value indicating the relevance.

In the learning result 2120, time series data for deriving the mathematical expression shown by reference numeral 2110 among time series data groups included in the data for learning are stored in the "1 st node" and the "2 nd node", respectively.

In the learning result 2120, the parameters m, n, and k calculated to derive the time series data of the 2 nd node by inputting the time series data of the 2 nd node to the mathematical expression shown by reference numeral 2110 are stored in "autocorrelation", "cross correlation", and "time delay".

As shown in fig. 21, only 1 learning result 2120 is generated for the learning data including the time series data group measured in the normal environment.

< specific example of treatment by evaluation section >

Next, a specific example of the processing performed by the evaluation unit 2012 of the analysis device 2010 will be described. Fig. 22 is a diagram 3 showing a specific example of the processing performed by the evaluation unit. As shown in fig. 22, the evaluation unit 2012 includes a regression model execution unit 2210 and a count value calculation unit 2220.

The regression model executor 2210 extracts time series data (actually measured value 2211) of the 1 st node in a time series data set (estimation data) measured in an unknown environment of the chamber a. The regression model execution unit 2210 also estimates the time-series data of the 2 nd node (estimated value 2212) by inputting the extracted time-series data of the 1 st node to the mathematical expression shown by reference numeral 2110.

At this time, the regression model execution unit 2210 reads out the parameters m, n, and k corresponding to the time series data input to the equation 2110 from the learning result 2120, sets the equation 2110, and estimates the time series data of the 2 nd node.

In fig. 22, an actual measurement value 2211 indicates time series data of the 1 st node inputted to the mathematical expression shown by reference numeral 2110 in a time series data set (estimation data) measured in an unknown environment of the chamber a. The estimated value 2212 indicates time series data of the 2 nd node estimated by inputting the measured value 2211.

On the other hand, the count value calculation unit 2220 includes a difference calculation unit 2221 and a count unit 2222.

The difference calculation unit 2221 extracts time series data (actual measurement value 2223) of the 2 nd node in a time series data set (estimation data) measured in an unknown environment of the chamber a. Further, the difference calculation unit 2221 obtains the estimated value 2212 from the regression model execution unit 2210. Further, the difference calculation unit 2221 calculates the difference between the actual measurement value 2223 and the estimated value 2212.

The counting unit 2222 counts the number of 1 st nodes (that is, a predetermined count value) for which the difference calculated by the difference calculation unit 2221 is equal to or greater than a predetermined threshold value. The counting unit 2222 outputs a predetermined count value obtained by counting as information indicating the environment of the chamber a (information indicating whether or not the chamber a is normal (or abnormal degree)).

In fig. 22, the horizontal axis of the graph 2230 represents time, and the vertical axis represents a predetermined count value output by the counting unit 2222. As shown in the graph 2230, when the predetermined count value output from the counting unit 2222 is less than the level (abnormality determination level) indicated by the broken line, the predetermined count value can be said to be information indicating that the chamber a is in a normal environment.

On the other hand, when the count value output from the counting unit 2222 reaches the abnormality determination level, the predetermined count value can be said to be information indicating that the chamber a is not in a normal environment.

The predetermined count value output by the counting unit 2222 can also be understood as information indicating the degree of abnormality in the environment of the chamber a by comparison with the abnormality determination level. Alternatively, the predetermined count value output by the counting unit 2222 may be understood as information for predicting the time when the environment of the chamber a becomes abnormal by predicting the time when the abnormality determination level is reached.

< flow of Environment adjustment processing >

Next, the flow of the entire environment adjustment processing performed by the environment adjustment system 2000 will be described. Fig. 23 is a flow chart of fig. 3 showing the flow of the environment adjustment processing.

In step S2301, the time-series data acquisition devices 140_1 to 140 — n measure time-series data sets associated with the processing of the pre-processed wafer in the chamber a, and store the measured time-series data sets in the learning data storage 163. In addition, the time series data acquisition devices 140_1 to 140_ n store time series data groups measured in association with the processing of the chamber a in the normal environment as learning data.

In step S2302, the learning unit 2011 of the analysis device 2010 performs machine learning on the regression model using the learning data stored in the learning data storage unit 163.

In step S2303, the timing data acquisition devices 140_ 1-140 _ n measure timing data sets associated with processing of the pre-processed wafer in the chamber A. In addition, the timing data acquisition devices 140_1 to 140_ n measure a timing data set (data for estimation) accompanying the processing of the chamber A in the unknown environment.

In step S2304, the evaluation unit 2012 of the analysis device 2010 inputs the time-series data set (estimation data) measured in step S2303 into the regression model, and calculates a predetermined count value.

In step S2305, the evaluation unit 2012 of the analyzer 2010 outputs the calculated predetermined count value to the controller 170 as information indicating the environment of the chamber a.

In step S2306, the control device 170 determines whether maintenance is necessary or not, or the time of maintenance, or the like, based on the information indicating the environment.

< summary >

From the above description, it is clear that, in the analysis apparatus according to embodiment 4,

the regression model is machine-learned using a time series data set measured in association with the processing of the wafer before processing in the chamber, and the autocorrelation, the cross correlation, the time delay, and the like, which are other examples of values indicating the correlation of the time series data in the corresponding time range between the measurement items, are calculated.

A predetermined count value is calculated by inputting a time series data set (estimation data) measured in association with the processing of the wafer before processing in the non-processing environment of the chamber into the regression model, and is output as information indicating the environment of the chamber.

Thus, the analysis apparatus according to embodiment 4 can quantitatively evaluate whether or not the environment of the chamber in the semiconductor manufacturing process is normal (or abnormal degree) based on the time-series data set.

[ 5 th embodiment ]

In the above-described embodiment 4, a case has been described in which a time series data set measured in association with processing of a wafer before processing is input to a regression model to calculate a predetermined count value, and information indicating the environment of a chamber is output.

In contrast, in embodiment 5, a predetermined count value is calculated by inputting a time series data set into a regression model, and a change in the predetermined count value is monitored. Thus, in embodiment 5, the change in the environment of the chamber can be grasped, and the end point detection of the etching process and the cleaning process can be performed.

< System configuration of end Point detection System >

First, a system configuration of the end point detection system will be described. Fig. 24 is a diagram showing an example of the system configuration of the end point detection system. The difference from the environment adjustment system 100 shown in fig. 1 is that the function of the analyzer 2410 differs from the function of the controller 2420 in the end point detection system 2400.

As shown in fig. 24, the analyzer 2410 functions as a learning unit 2411 and an end point detection unit 2412. The learning unit 2411 and the end point detection unit 2412 have two functions, and perform end point detection by either function.

(i) Description of function No. 1 of learning Unit and end Point detecting Unit

The learning part 2411 uses the time series data measured by the time series data acquisition devices 140_1 to 140_ n

The time point at which the etching process is completed in the processing space 120, or

The point in time at which the cleaning process ends in the process space 120

The measured time series data set (learning data) is machine-learned for the regression model.

The end point detection unit 2412 inputs a regression model generated by the learning unit 2411 performing machine learning using the learning data

A time series data set (data for detection) measured during the etching process in the processing space 120, or

A time-series data set (data for detection) measured in the cleaning process in the processing space 120,

thereby, the "prescribed count value" is calculated.

Then, the end point detecting unit 2412 detects a time point at which the predetermined count value becomes equal to or less than a predetermined threshold value as the end point of the etching process or the end point of the cleaning process. The end point detecting unit 2412 transmits end point information such as the detected end point of the etching process or the detected end point of the cleaning process to the control device 2420.

The end point detection unit 2412 counts the number of combinations having a predetermined relevance between the 1 st node and the 2 nd node among the combinations of the 1 st node and the 2 nd node. The "predetermined count value" is the number of combinations in which the predetermined relevance is broken (the value indicating the strength of the relevance is equal to or less than a predetermined threshold value) among combinations in which the 1 st node and the 2 nd node have the predetermined relevance.

(ii) Description of function No. 2 of learning Unit and end Point detecting Unit

The time point at which the etching process is started in the processing space 120, or

The point in time at which the cleaning process is started in the processing space 120

Time series data set (data for detection) measured in the cleaning process in the processing space 120

The "prescribed count value" is calculated.

Then, the end point detecting unit 2412 detects a time point at which the predetermined count value changes to or above a predetermined threshold value as the end point of the etching process or the end point of the cleaning process. The end point detecting unit 2412 transmits end point information such as the detected end point of the etching process or the detected end point of the cleaning process to the control device 2420.

(iii) Description of the function of the control device

The controller 2420 adjusts, for example, etching time, etching recipe, or cleaning time, cleaning recipe, etc., based on the end point information output from the end point detecting unit 2412 of the analyzer 2410.

< specific example of processing by the learning section >

Next, a specific example of the processing performed by the learning unit 2411 of the analysis device 2410 will be described. Fig. 25 is a diagram 4 showing a specific example of the processing performed by the learning unit. As shown in fig. 25, the learning unit 2411 includes a regression model generation unit 2501.

The regression model generation unit 2501 performs machine learning on the regression model using the time series data group stored in the data storage unit 2413. Fig. 25 shows a case where machine learning is performed using a time-series data set measured at the time point when the etching process ends or the cleaning process ends. Although not shown in fig. 25, machine learning may be performed using a time-series data set measured at the time point when the etching process is started or the time point when the cleaning process is started.

Thus, the regression model generation unit 2501 defines the relationship of time series data between the measurement items measured by the time series data acquisition devices 140_1 to 140_ n by using the mathematical expression shown by reference numeral 2110.

Note that a method of specifying the relationship between time series data of the measurement items by using the mathematical expression shown by reference numeral 2110 has already been described in embodiment 4 above with reference to fig. 21, and therefore, the description thereof is omitted here.

In fig. 25, the learning result 2520 shows each parameter of the mathematical expression shown by reference numeral 2110 calculated by machine learning on the regression model. Note that, the details of the learning result 2520 have been described in embodiment 4 above using fig. 21, and therefore the description thereof is omitted here.

< specific example of processing by the end-point detecting section >

Next, a specific example of the processing performed by the end point detecting unit of the analyzer 2410 will be described. Fig. 26 is a diagram showing a specific example of the processing performed by the end point detecting unit. As shown in fig. 26, the end point detecting unit 2412 includes a regression model executing unit 2610 and a count value calculating unit 2620.

The regression model execution unit 2610 extracts time series data (actually measured value 2211) of the 1 st node in the time series data set (detection data) measured in the etching process or the cleaning process. The regression model execution unit 2610 also estimates the time-series data of the 2 nd node (estimated value 2212) by inputting the extracted time-series data of the 1 st node to the mathematical expression shown by reference numeral 2110.

At this time, the regression model execution unit 2610 reads out the parameters m, n, and k corresponding to the time series data of the equation inputted thereto shown by reference numeral 2110 from the learning result 2520, sets the equation shown by reference numeral 2110, and estimates the time series data of the 2 nd node.

In fig. 26, an actual measurement value 2211 indicates time series data input to the 1 st node of the mathematical expression shown by reference numeral 2110 in a time series data set (data for detection) measured in the etching process or the cleaning process. The estimated value 2212 indicates time series data of the 2 nd node estimated by inputting the measured value 2211.

On the other hand, the count value calculation section 2620 includes a difference calculation section 2621, a count section 2622, and a determination section 2623.

The difference calculation unit 2621 extracts the time-series data (actually measured value 2223) of the 2 nd node in the time-series data set (detection data) measured during the etching process or the cleaning process. The difference calculation unit 2621 acquires the estimated value 2212 by the regression model execution unit 2610. The difference calculation unit 2621 also calculates the difference between the measured value 2223 and the estimated value 2212.

The counting unit 2622 counts the number of 1 st nodes (i.e., a predetermined count value) whose difference calculated by the difference calculation unit 2621 is equal to or greater than a predetermined threshold.

When the determination unit 2623 performs machine learning using a time-series data set at the time point when the etching process ends or the time point when the cleaning process ends,

the time point at which the predetermined count value counted by the counting unit 2622 becomes equal to or less than the preset threshold is determined as the end point of the etching process or the end point of the cleaning process, and the end point information is output.

In addition, when the determination unit 2623 performs machine learning using a time series data set at the time point when the etching process is started or the time point when the cleaning process is started,

a time point at which the predetermined count value counted by the counting unit 2622 becomes equal to or more than a predetermined threshold value is determined as an end point of the etching process or an end point of the cleaning process, and end point information is output.

< flow of end-point detection processing >

Next, the flow of the entire end point detection process performed by the end point detection system 2400 will be described. Fig. 27 is a flowchart showing the flow of the end point detection processing.

In step S2701, the time series data acquisition devices 140_1 to 140_ n save

At the point of time when the etching process (or cleaning process) ends, or

A time series data set (learning data) measured at a time point when the etching process (or the cleaning process) starts.

In step S2702, the learning unit 2411 of the analysis device 2410 performs machine learning on the regression model using the time-series data set (learning data) stored in the data storage unit 2413.

In step S2703, the time-series data acquisition devices 140_1 to 140_ n measure the time-series data sets in the etching process or the time-series data sets (data for detection) in the cleaning process.

In step S2704, the end point detecting unit 2412 of the analysis device 2410 inputs the time series data set (data for detection) measured in step S2703 to the regression model, and calculates a predetermined count value.

In step S2705, the end point detecting unit 2412 of the analysis device 2410 determines whether or not the calculated predetermined count value satisfies a predetermined condition. The preset conditions here are:

when machine learning is performed using a time-series data set measured at the time point when the etching process (or cleaning process) ends, the time-series data set is a case where the time-series data set is equal to or less than a preset threshold value;

when machine learning is performed using a time-series data set measured at the time point when the etching process (or cleaning process) starts, the time-series data set is equal to or more than a predetermined threshold value.

If it is determined in step S2705 that the preset condition is not satisfied (if no in step S2705), the process returns to step S2703.

On the other hand, if it is determined in step S2705 that the predetermined condition is satisfied (if yes in step S2705), the process proceeds to step S2706.

In step S2706, the end point detecting unit 2412 of the analyzer 2410 determines that the end point of the etching process or the end point of the cleaning process is detected, and outputs end point information.

< summary >

From the above description, it is clear that, in the analysis device according to embodiment 5,

machine learning against a regression model using a time series data set measured at the point in time of the end of the etching process or at the point in time of the cleaning process, or

Machine learning is performed on the regression model using a time series data set measured at the time point when the etching process starts or the time point when the cleaning process starts.

The number of combinations of time series data having a predetermined correlation between the measurement items is counted based on an estimated value obtained when a time series data set (data for detection) measured during the etching process or the cleaning process is input to the regression model subjected to the machine learning.

When the count value satisfies a predetermined condition, it is determined that the end point of the etching process or the end point of the cleaning process is detected, and end point information is output.

Thus, according to the analysis device of embodiment 5, the end point of the etching process or the cleaning process can be determined with high accuracy.

[ 6 th embodiment ]

In the above-described embodiment 5, a specific example of the time-series data acquisition device and the time-series data group is not mentioned, but the time-series data acquisition device may be, for example, the same as in the above-described embodiments 2 and 3

An emission spectrum analysis device;

a quadrupole mass spectrometry device;

various process sensors.

In addition, the time series data set may be, for example,

OES data;

mass spectrometry data;

process data set.

In the above-described embodiment 5, in order to detect the end point of the etching process or the end point of the cleaning process, machine learning is performed on the regression model using a time series data set measured at the time point when the etching process ends or the time point when the cleaning process ends. However, in order to detect a specific state of the etching process or a specific state of the cleaning process, machine learning may be performed on the regression model using a time series data set measured at a specific time point of the etching process or a specific time point of the cleaning process.

[ other embodiments ]

In the above embodiments, the learning unit is described as a device for performing machine learning on a regression model. However, the model for machine learning by the learning unit is not limited to the regression model, and may be another model as long as it can calculate the correlation (correlation) of the time series data.

In the above-described embodiment 2, the contents of the 1 st learning data and the 2 nd learning data generated for the emission intensity data of each wavelength included in the visible light wavelength range are described. However, the emission intensity data used for the generation of the 1 st learning data and the 2 nd learning data may be emission intensity data of a specific wavelength. The data may be emission intensity data of wavelengths outside the wavelength range of visible light.

In the above-described embodiment 4, the description has been given of the content of the counting unit 2222 calculating a predetermined count value by counting the number of 1 st nodes whose difference calculated by the difference calculation unit 2221 is equal to or greater than a predetermined threshold value. However, the method of counting the predetermined count value is not limited to this. For example, the predetermined count value may be calculated by counting the number of 1 st nodes set in advance among the 1 st nodes whose difference calculated by the difference calculation unit 2221 is equal to or greater than a predetermined threshold value.

In the above-described embodiment 2, the OES data (or the mass spectrometry data) is exemplified as a specific example of the time series data set, and in the above-described embodiment 3, the process data set is exemplified as a specific example of the time series data set, but the time series data set is not limited thereto. For example, the time series data set may be a time series data set indicating a physical quantity of plasma measured by the plasma device.

In the above embodiments, the analyzing device and the control device are separately configured, but the analyzing device and the control device may be integrally configured.

The present invention is not limited to the configurations given herein, such as the configurations exemplified in the above embodiments, or combinations with other elements. These aspects can be changed without departing from the spirit of the present invention, and can be determined as appropriate according to the application form thereof.

Claims

1. An analysis apparatus, comprising:

and an evaluation unit that evaluates an unknown environment of the processing space based on the value indicating the correlation calculated by the machine learning by the learning unit, using a time-series data set measured in association with processing of the object in the known environment of the processing space.

2. The analysis device of claim 1, wherein:

further comprising a storage unit that stores, as information indicating a corresponding environment, a value indicating the correlation calculated by machine learning by the learning unit using a plurality of time-series data sets measured in association with processing of the object in a plurality of known environments using a processing space,

the evaluation unit evaluates the unknown environment of the processing space by determining which of the values representing the relevance calculated by the learning unit through machine learning using the time-series data set measured in association with the processing of the object in the unknown environment in the processing space is similar to the value representing the relevance stored in the storage unit.

3. The analysis device of claim 2, wherein:

the storage unit stores, as information indicating a corresponding environment, a value indicating the correlation calculated by the learning unit performing machine learning on a plurality of time-series data sets measured by processing on the object based on a specific recipe under a plurality of normal known environments using a processing space.

4. The analysis device of claim 3, wherein:

the storage unit stores, as information indicating a corresponding environment, a value indicating the correlation calculated by machine learning by the learning unit using a plurality of time-series data sets measured with respect to processing of the object under a plurality of known environments in which the kind of deposit or the amount of deposit adhering to the processing space is different.

5. The analysis device of claim 3, wherein:

the storage unit stores, as information indicating a corresponding environment, a value indicating the correlation calculated by machine learning by the learning unit using a plurality of time-series data sets measured in association with processing of the object in a plurality of known environments in which the consumption degrees of the respective components in the processing space are different.

6. The analysis device of claim 4, wherein:

the plurality of time series data sets are OES data measured by an emission spectroscopy apparatus or mass spectrometry data measured by a mass spectrometry apparatus.

7. The analysis device of claim 5, wherein:

the plurality of time-series data sets are process data sets measured by a process data acquisition device.

8. The analysis device of claim 5, wherein:

the plurality of time-series data sets are time-series data sets of plasma physical quantities measured by the plasma device.

9. The analysis device according to any one of claims 1 to 8, wherein:

the value indicating the correlation is a value indicating the strength of the correlation of the time series data in the corresponding time range between the measurement items.

10. The assay device of any one of claims 1 to 8, wherein:

the value indicating the correlation is a value obtained by summing up values indicating the strength of the correlation of time series data in the corresponding time range between the measurement items for each measurement item.

11. The assay device of any one of claims 1 to 10, wherein:

an environment adjustment parameter for changing the environment of the processing space is determined based on the environment evaluated by the evaluation unit.

12. The analysis device of claim 1, wherein:

the value indicating the correlation includes a parameter indicating autocorrelation, cross-correlation, or time lag, which is calculated by inputting the time series data in the 1 st measurement item into a predetermined expression to derive the time series data of the 2 nd measurement item.

13. The analysis device of claim 1, wherein:

determining, based on the environment evaluated by the evaluation unit:

whether maintenance of the processing space, maintenance of components of the processing space, and maintenance of components that affect the processing space need to be performed; or

The time of maintenance of the processing space, the time of maintenance of components of the processing space, and the time of maintenance of components that affect the processing space.

14. An analysis apparatus, comprising:

a learning unit that performs machine learning on a regression model using a time series data set measured at a specific time point of an etching process or a cleaning process in a processing space; and

and an end point detection unit that counts the number of combinations of time series data having a predetermined correlation between the measurement items based on an estimated value obtained when a time series data set measured during the etching process or the cleaning process in the processing space is input to the regression model obtained by the machine learning by the learning unit, and determines an end point of the etching process or the cleaning process when the counted value satisfies a predetermined condition.

15. The analysis device of claim 14, wherein:

when the learning unit performs machine learning on the regression model using a time series data set at a time point when the etching process or the cleaning process ends, the end point detection unit determines that the etching process or the cleaning process is an end point when the count value is equal to or less than a predetermined threshold value.

16. The analysis device of claim 14, wherein:

when the learning unit performs machine learning on the regression model using a time series data set at a time point when the etching process or the cleaning process starts, the end point detection unit determines that the etching process or the cleaning process is an end point when the count value is equal to or greater than a predetermined threshold value.

17. An analysis method, comprising:

a learning step of performing machine learning using a time series data group measured in association with processing of a target object in a processing space, and calculating a value indicating a correlation of time series data in a corresponding time range between measurement items; and

and an evaluation step of evaluating an unknown environment of the processing space based on the value representing the correlation calculated by the machine learning in the learning step, using a time-series data set measured in association with the processing of the object in the known environment of the processing space.

18. An analysis program characterized by:

the analysis program is a program for causing a computer to execute the steps of: