Endpoint Detection Device For Realizing Real-Time Control Of Plasma Reactor, Plasma Reactor With Endpoint Detection Device, And Endpoint Detection Method

• US 20090029489A1
• Filed: 02/25/2008
• Published: 01/29/2009
• Est. Priority Date: 07/24/2007
• Status: Active Grant

• Current Assignee: DMS Co., Ltd.
• Original Assignee: DMS Co., Ltd.
• Inventors: Kim, Kee Hyun, Han, Kwang Hoon, Park, Kun Joo, Lee, Weon Mook, Chae, Heeyeop, Han, Kyounghoon
• Granted Patent: US 8,049,872 B2
• Time in Patent Office: Days
• US Class Current: 438/9
• CPC Class Codes: G01N 21/71 thermally excited
• CPC Class Codes: H01J 37/32935 Monitoring and controlling tubes by information coming from the object and/or discharge
• CPC Class Codes: H01J 37/32963 End-point detection
• CPC Class Codes: H01L 21/67069 for drying etching
• CPC Class Codes: H01L 21/67253 Process monitoring, e.g. flow or thickness monitoring

Abstract

An endpoint detection device, a plasma reactor with the endpoint detection device, and an endpoint detection method are provided. The endpoint detection device includes an OES data operation unit, a data selector, a product generator, an SVM, and an endpoint determiner. The OES data operation unit processes reference OES data by normalization and PCA. The data selector selects part of the linear reference loading vectors and selects part of the selected linear reference loading vectors. The product generator outputs at least one reference product value. The SVM performs regression and outputs a prediction product value. The endpoint determiner detects a process wafer etch or deposition endpoint and outputs a detection signal.

Claims

1. An endpoint detection device comprising:

an Optical Emission Spectrometer (OES) data operation unit for processing reference OES data by normalization and Principal Component Analysis (PCA), outputting linear reference loading vectors and rate values of reference principal components, and outputting reference ranking values on the basis of the linear reference loading vectors;

a data selector for first selecting part of the linear reference loading vectors on the basis of the rate values of the reference principal components, selecting part of the reference OES data or selecting part of process OES data on the basis of the reference ranking values, and second selecting part of the first selected linear reference loading vectors on the basis of the reference ranking values;

a product generator for outputting at least one reference product value on the basis of the first selected linear reference loading vectors and the reference OES data;

a Support Vector Machine (SVM) for performing regression on the basis of the selected reference OES data and the at least one reference product value, generating nonlinear reference loading vectors, and periodically outputting a prediction product value on the basis of the second selected linear reference loading vectors, the nonlinear reference loading vectors, and the selected process OES data; and

an endpoint determiner for detecting a process wafer etch or deposition endpoint on the basis of the periodically received prediction product value and outputting a detection signal,

wherein first lights of a whole wavelength emitted from the inside of a plasma reaction chamber are converted into the reference OES data by a spectrometer during a reference wafer etch or deposition process and second lights of a whole wavelength emitted from the inside of the plasma reaction chamber are converted into the process OES data by the spectrometer during a process wafer etch or deposition process executed after the reference wafer etch or deposition process,

wherein the reference ranking values represent a ranking for the intensity of the first lights of the whole wavelength, and

wherein the linear reference loading vectors each correspond to one-dimensional function values expressed by the reference OES data and the nonlinear reference loading vectors each correspond to S-dimensional (S;

integer larger than

     1) function values expressed by the selected process OES data.

2. The device of claim 1, further comprising:

an abnormality diagnosis unit for diagnosing normality or abnormality of a currently etched or deposited process wafer depending on whether a mean deviation value is comprised within a stable range determined by an average value and controlling a display unit to visually display the normality or abnormality of the process wafer,

wherein the OES data operation unit processes the process OES data by normalization and PCA, further outputs linear process loading vectors and rate values of process principal components, and further outputs process ranking values representing a ranking for the intensity of the second lights of the whole wavelength on the basis of the linear process loading vectors,

wherein the data selector selects part of the linear process loading vectors on the basis of the rate values of the process principal components,

wherein the product generator further outputs an actual product value on the basis of the selected linear process loading vectors and the process OES data,

wherein whenever the process wafer etch or deposition process is executed, the SVM accumulates each of the actual product values received from the product generator and further calculates the average value and the mean deviation value for the accumulated actual product values, and

wherein the linear process loading vectors each correspond to one-dimensional function values expressed by the process OES data.

3. The device of claim 2, wherein when the number of times of execution of the process wafer etch or deposition process reaches the preset number of times, the data selector outputs an update request signal and further selects part of the process OES data and part of the selected linear process loading vectors on the basis of the process ranking values,

the SVM performs regression on the basis of the actual product value generated by the product generator and the selected process OES data during a current process wafer etch or deposition process in response to the update request signal, generates nonlinear process loading vectors, and uses the further selected linear process loading vectors and the nonlinear process loading vectors to generate prediction product values from a time of a subsequent process wafer etch or deposition process up to a time of a process wafer etch or deposition process before execution of a subsequent regression operation, and

the nonlinear reference loading vectors each correspond to S-dimensional (S:

integer larger than

     1) function values expressed by the selected process OES data.

4. The device of claim 2, wherein the OES data operation unit comprises:

a normalization processor for normalizing the reference OES data and outputting reference average scaling data or normalizing the process OES data and outputting process average scaling data;

a PCA processor for processing the reference average scaling data by PCA and outputting the linear reference loading vectors and the rate values of the reference principal components or processing the process average scaling data by PCA and outputting the linear process loading vectors and the rate values of the process principal components; and

a ranking determiner for outputting the reference ranking values on the basis of the linear reference loading vectors or outputting the process ranking values on the basis of the linear process loading vectors.

5. The device of claim 4, wherein the OES data operation unit further comprises:

a storage unit for storing the linear reference loading vectors and the rate values of the reference principal components or the linear process loading vectors and the rate values of the process principal components received from the PCA processor and storing the reference ranking values or the process ranking values received from the ranking determiner.

6. The device of claim 2, wherein an upper limit value of the stable range is determined by a sum of the average value and a deviation value, a lower limit value of the stable range is determined by a difference between the average value and the deviation value, and the deviation value is preset to the abnormality diagnosis unit.

7. A plasma reactor comprising:

a plasma reaction chamber which a reference wafer or a process wafer is mounted inside;

a spectrometer for converting first lights of a whole wavelength emitted from the inside of the plasma reaction chamber into reference OES data during a reference wafer etch or deposition process or converting second lights of a whole wavelength emitted from the inside of the plasma reaction chamber into process OES data during a process wafer etch or deposition process executed after the reference wafer etch or deposition process;

an optical fiber cable for collecting the first or second lights of the whole wavelength emitted from the inside of the plasma reaction chamber through a window provided at an outer wall of the plasma reaction chamber and forwarding the collected lights to the spectrometer during the reference wafer or process wafer etch or deposition process;

an endpoint detection device for detecting a process wafer etch or deposition endpoint and outputting a detection signal on the basis of the reference OES data and the process OES data; and

a plasma reaction controller for controlling an etch or deposition condition within the plasma reaction chamber in response to the detection signal,

the endpoint detection device comprising;

an OES data operation unit for processing the reference OES data by normalization and PCA, outputting linear reference loading vectors and rate values of reference principal components, and outputting reference ranking values on the basis of the linear reference loading vectors;

a data selector for first selecting part of the linear reference loading vectors on the basis of the rate values of the reference principal components, selecting part of the reference OES data or selecting part of the process OES data on the basis of the reference ranking values, and second selecting part of the first selected linear reference loading vectors on the basis of the reference ranking values;

a product generator for outputting at least one reference product value on the basis of the first selected linear reference loading vectors and the reference OES data;

an SVM for performing regression on the basis of the selected reference OES data and the at least one reference product value, generating nonlinear reference loading vectors, and periodically outputting a prediction product value on the basis of the second selected linear reference loading vectors, the nonlinear reference loading vectors, and the selected process OES data; and

an endpoint determiner for detecting a process wafer etch or deposition endpoint on the basis of the periodically received prediction product value and outputting a detection signal,

wherein the reference ranking values represent a ranking for the intensity of the first lights of the whole wavelength, and

wherein the linear reference loading vectors each correspond to one-dimensional function values expressed by the reference OES data and the nonlinear reference loading vectors each correspond to S-dimensional (S;

integer larger than

     1) function values expressed by the selected process OES data.

8. The plasma reactor of claim 7, further comprising:

an abnormality diagnosis unit for diagnosing normality or abnormality of a currently etched or deposited process wafer depending on whether a mean deviation value is comprised within a stable range determined by an average value and controlling a display unit to visually display the normality or abnormality of the process wafer,

wherein the OES data operation unit processes the process OES data by normalization and PCA, further outputs linear process loading vectors and rate values of process principal components, and further outputs process ranking values representing a ranking for the intensity of the second lights of the whole wavelength on the basis of the linear process loading vectors,

wherein the data selector selects part of the linear process loading vectors on the basis of the rate values of the process principal components,

wherein the product generator further outputs an actual product value on the basis of the selected linear process loading vectors and the process OES data,

wherein whenever the process wafer etch or deposition process is executed, the SVM accumulates each of the actual product values received from the product generator and further calculates the average value and the mean deviation value for the accumulated actual product values, and

wherein the linear process loading vectors each correspond to one-dimensional function values expressed by the process OES data.

9. The plasma reactor of claim 8, wherein when the number of times of execution of the process wafer etch or deposition process reaches the preset number of times, the data selector outputs an update request signal and further selects part of the process OES data and part of the selected linear process loading vectors on the basis of the process ranking values,

the SVM performs regression on the basis of the actual product value generated by the product generator and the selected process OES data during a current process wafer etch or deposition process in response to the update request signal, generates nonlinear process loading vectors, and uses the further selected linear process loading vectors and the nonlinear process loading vectors to generate prediction product values from a time of a subsequent process wafer etch or deposition process up to a time of a process wafer etch or deposition process before execution of a subsequent regression operation, and

the nonlinear reference loading vectors each correspond to S-dimensional (S:

integer larger than

     1) function values expressed by the selected process OES data.

10. The plasma reactor of claim 8, wherein the OES data operation unit comprises:

a normalization processor for normalizing the reference OES data and outputting reference average scaling data or normalizing the process OES data and outputting process average scaling data;

a PCA processor for processing the reference average scaling data by PCA and outputting the linear reference loading vectors and the rate values of the reference principal components or processing the process average scaling data by PCA and outputting the linear process loading vectors and the rate values of the process principal components; and

a ranking determiner for outputting the reference ranking values on the basis of the linear reference loading vectors or outputting the process ranking values on the basis of the linear process loading vectors.

11. The plasmas reactor of claim 10, wherein the OES data operation unit further comprises:

a storage unit for storing the linear reference loading vectors and the rate values of the reference principal components or the linear process loading vectors and the rate values of the process principal components received from the PCA processor and storing the reference ranking values or the process ranking values received from the ranking determiner.

12. An endpoint detection device comprising:

a data selector for setting a data selection range and outputting an operation control signal in response to a data selection signal, selecting part of reference OES data on the basis of the data selection range during a reference wafer etch or deposition process, and selecting part of process OES data on the basis of the data selection range during a process wafer etch or deposition process;

an OES data operation unit for processing the selected reference OES data by normalization and PCA and outputting linear reference loading vectors in response to the operation control signal;

a product generator for periodically outputting a prediction product value on the basis of the linear reference loading vectors and the selected process OES data; and

an endpoint determiner for detecting a process wafer etch or deposition endpoint and outputting a detection signal on the basis of the periodically received prediction product value,

wherein first lights of a whole wavelength emitted from the inside of a plasma reaction chamber are converted into the reference OES data by a spectrometer during a reference wafer etch or deposition process and second lights of a whole wavelength emitted from the inside of the plasma reaction chamber are converted into the process OES data by the spectrometer during a process wafer etch or deposition process executed after the reference wafer etch or deposition process, and

wherein the linear reference loading vectors each correspond to one-dimensional function values expressed by the reference OES data.

13. The device of claim 12, wherein the data selector changes and sets the data selection range and outputs the operation control signal whenever further receiving the data selection signal,

wherein the OES data operation unit processes, by normalization and PCA, process OES data selected on the basis of the changed data selection range by the data selector and outputs linear process loading vectors during a current process wafer etch or deposition process in response to the operation control signal, and

wherein the product generator uses the linear process loading vectors to generate prediction product values from a time of a subsequent process wafer etch or deposition process up to a time of a process wafer etch or deposition process before the OES data operation unit executes a subsequent normalization and PCA process operation.

14. The device of claim 13, wherein the OES data operation unit comprises:

a normalization processor for normalizing the reference OES data and outputting reference average scaling data or normalizing the process OES data and outputting process average scaling data, in response to the operation control signal; and

a PCA processor for processing the reference average scaling data by PCA and outputting the linear reference loading vectors or processing the process average scaling data by PCA and outputting the linear process loading vectors.

15. The device of claim 14, wherein the OES data operation unit further comprises:

a storage unit for storing the linear reference loading vectors or the linear process loading vectors received from the PCA processor in response to the operation control signal.

16. The device of claim 13, wherein the data selection range is any one or more of at least one set wavelength band range, at least one set time range, and a wavelength band range of lights emitted by at least one set element.

17. A plasma reactor comprising:

a plasma reaction chamber which a reference wafer or a process wafer is mounted inside;

a spectrometer for converting first lights of a whole wavelength emitted from the inside of the plasma reaction chamber into reference OES data during a reference wafer etch or deposition process or converting second lights of a whole wavelength emitted from the inside of the plasma reaction chamber into process OES data during a process wafer etch or deposition process executed after the reference wafer etch or deposition process;

an optical fiber cable for collecting the first or second lights of the whole wavelength emitted from the inside of the plasma reaction chamber through a window provided at an outer wall of the plasma reaction chamber and forwarding the collected lights to the spectrometer during the reference wafer or process wafer etch or deposition process;

an endpoint detection device for detecting a process wafer etch or deposition endpoint and outputting a detection signal on the basis of the reference OES data and the process OES data; and

a plasma reaction controller for controlling an etch or deposition condition within the plasma reaction chamber in response to the detection signal,

the endpoint detection device comprising;

a data selector for setting a data selection range and outputting an operation control signal in response to a data selection signal, selecting part of reference OES data on the basis of the data selection range during the reference wafer etch or deposition process, and selecting part of process OES data on the basis of the data selection range during the process wafer etch or deposition process;

an OES data operation unit for processing the selected reference OES data by normalization and PCA and outputting linear reference loading vectors in response to the operation control signal;

a product generator for periodically outputting a prediction product value on the basis of the linear reference loading vectors and the selected process OES data; and

an endpoint determiner for detecting a process wafer etch or deposition endpoint and outputting a detection signal on the basis of the periodically received prediction product value,

wherein the linear reference loading vectors each correspond to one-dimensional function values expressed by the reference OES data.

18. The plasma reactor of claim 17, wherein the data selector changes and sets the data selection range and outputs the operation control signal whenever further receiving the data selection signal,

wherein the OES data operation unit processes, by normalization and PCA, process OES data selected on the basis of the changed data selection range by the data selector and outputs linear process loading vectors during a current process wafer etch or deposition process in response to the operation control signal, and

wherein the product generator uses the linear process loading vectors to generate prediction product values from a time of a subsequent process wafer etch or deposition process up to a time of a process wafer etch or deposition process before the OES data operation unit executes a subsequent normalization and PCA process operation.

19. The plasma reactor of claim 17, further comprising:

a display unit for displaying the reference OES data or the process OES data in a three-dimensional graphic picture; and

an input unit comprising a plurality of keys and outputting the data selection signal in response to pressing the plurality of keys.

20. An endpoint detection method comprising:

processing, by an OES data operation unit, reference OES data by normalization and PCA and generating linear reference loading vectors, rate values of reference principal components, and reference ranking values;

making an SVM learn on the basis of the linear reference loading vectors, the rate values of the reference principal components, the reference ranking values, and the reference OES data;

selecting, by a data selector, part of process OES data on the basis of the reference ranking values;

periodically generating a prediction product value using the learning SVM and the selected process OES data; and

detecting, by an endpoint determiner, a process wafer etch or deposition endpoint on the basis of the prediction product value and outputting a detection signal,

wherein the reference OES data are data obtained by converting, by a spectrometer, first lights of a whole wavelength emitted from the inside of a plasma reaction chamber during a reference wafer etch or deposition process, and the process OES data are data obtained by converting, by the spectrometer, second lights of a whole wavelength emitted from the inside of the plasma reaction chamber during a process wafer etch or deposition process executed after the reference wafer etch or deposition process,

wherein the linear reference loading vectors each correspond to one-dimensional function values expressed by the reference OES data, and

wherein the reference ranking values represent a ranking for the intensity of the first lights of the whole wavelength.

21. The detection method of claim 20, wherein making the SVM learn comprises:

first selecting, by the data selector, part of the linear reference loading vectors on the basis of the rate values of the reference principal components;

generating, by a product generator, at least one reference product value on the basis of the first selected linear reference loading vectors and the reference OES data;

selecting, by the data selector, part of the reference OES data on the basis of the reference ranking values; and

performing, by the SVM, regression on the basis of the selected reference OES data and the at least one reference product value and generating nonlinear reference loading vectors,

wherein the nonlinear reference loading vectors each correspond to S-dimensional (S;

integer larger than

     1) function values expressed by the selected process OES data.

22. The detection method of claim 21, wherein the reference ranking values are generated on the basis of the linear reference loading vectors, and

wherein the generating of the prediction product value comprises:

second selecting, by the data selector, part of the first selected linear reference loading vectors on the basis of the reference ranking values; and

periodically outputting, by the SVM, the prediction product value on the basis of the second selected linear reference loading vectors, the nonlinear reference loading vectors, and the selected process OES data.

23. The detection method of claim 20, further comprising:

whenever an additional process wafer etch or deposition process is executed, repeating

selecting part of the process OES data;

periodically generating the prediction product value; and

outputting the detection signal.

24. The detection method of claim 23, further comprising:

when the number of times of execution of the process wafer etch or deposition process reaches the preset number of times, normalizing and PCA processing, by the OES data operation unit, process OES data generated by the spectrometer during a current process wafer etch or deposition process and generating linear process loading vectors, rate values of process principal components, and process ranking values; and

additionally making the SVM learn on the basis of the linear process loading vectors, the rate values of the process principal components, and the process ranking values,

wherein the data selector selects part of the process OES data on the basis of the process ranking values from a process wafer etch or deposition process after the SVM learns up to a process wafer etch or deposition process before the SVM learns subsequently, and

wherein the process ranking values represent a ranking for the intensity of the second lights of the whole wavelength.

25. The detection method of claim 23, further comprising:

whenever the additional process wafer etch or deposition process is executed,

normalizing and PCA processing, by the OES data operation unit, the process OES data and generating linear process loading vectors and rate values of process principal components;

selecting, by the data selector, part of the linear process loading vectors on the basis of the rate values of the process principal components;

generating, by the product generator, an actual product value on the basis of the selected linear process loading vectors and the process OES data;

accumulating, by the SVM, the actual product value and calculating an average value and a mean deviation value for the accumulated actual product value; and

determining, by an abnormality diagnosis unit, normality or abnormality of a currently etched or deposited process wafer depending on whether the mean deviation value is comprised within a stable range determined by the average value.