Method of Feature Exaction from Time-series of Spectra to Control Endpoint of Process

• US 20180182632A1
• Filed: 12/23/2016
• Published: 06/28/2018
• Est. Priority Date: 12/23/2016
• Status: Active Grant

• Current Assignee: Lam Research Corporation
• Original Assignee: Lam Research Corporation
• Inventors: Feng, Ye, Kumar, Prashanth, Bailey, III, Andrew D.
• Granted Patent: US 10,262,910 B2
• Time in Patent Office: Days
• CPC Class Codes: H01J 37/32926 Software, data control or modelling
• CPC Class Codes: H01J 37/32963 End-point detection
• CPC Class Codes: H01J 37/32972 Spectral analysis
• CPC Class Codes: H01L 21/67259 Position monitoring, e.g. misposition detection or presence detection
• CPC Class Codes: H01L 22/00 Testing or measuring during manufacture or treatment; Reliability measurements, i.e. testing of parts without further processing to modify the parts as such; Structural arrangements therefor
• CPC Class Codes: H01L 22/12 for structural parameters, e.g. thickness, line width, refractive index, temperature, warp, bond strength, defects, optical inspection, electrical measurement of structural dimensions, metallurgic measurement of diffusions electrical measurement of diffusions H01L22/14
• CPC Class Codes: H01L 22/24 Optical enhancement of defects or not directly visible states, e.g. selective electrolytic deposition, bubbles in liquids, light emission, colour change voltage contrast G01R31/311
• CPC Class Codes: H01L 22/26 Acting in response to an ongoing measurement without interruption of processing, e.g. endpoint detection, in-situ thickness measurement endpoint detection arrangements in CMP apparatus B24B37/013, in discharge apparatus H01J37/32

Abstract

Methods and systems for using a time-series of spectra to identify endpoint of an etch process. One method includes accessing a virtual carpet that is formed from a time-series of spectra for the etch process collected during a training operation. And, running a fabrication etch process on a fabrication wafer, such that while the fabrication etch process is performed portions of a carpet defined from a time-series of spectral is generated for the fabrication etch process. Then, comparing the portions of the carpet of the fabrication etch process to the virtual carpet. End pointing is processed for the fabrication etch process when said comparing indicates that a desired metric has been reached for the fabrication wafer. In one example, said portions of the carpet include a current frame of captured spectra and at least one previous frame of captured spectra. The portions of the carpet of the fabrication etch process are fitted to the virtual carpet to identify a virtual frame number and associated floating parameters that are used in a correlation to predicted a value for the metric. Further, each of the carpets produced during the training operation and the virtual carpet are defined by a polynomial. The coefficients of the carpets produced during the training operation are a subset of the coefficients of the polynomial of the virtual carpet.

Claims

1. A method for using a time-series of spectra to identify endpoint of an etch process, comprising,

accessing a virtual carpet that is generated from a time-series of spectra for the etch process collected during a training operation;

running a fabrication etch process on a fabrication wafer, such that while the fabrication etch process is performed portions of a carpet defined from a time-series of spectra is generated for the fabrication etch process;

comparing the portions of the carpet of the fabrication etch process to the virtual carpet; and

processing end pointing of the fabrication etch process when said comparing indicates that a desired metric has been reached for the fabrication wafer.

2. The method of claim 1, wherein the desired metric is associated with at least one of a predefined etch depth, a predefined critical dimension value, a predefined wafer bow value, a predetermined line width value, a predefined feature pitch value, a predefined features spacing value, or predetermined measurable value.

3. The method of claim 1, wherein said portions of the carpet include a current frame of captured spectra and at least one previous frame of captured spectra, wherein each portion of said carpet includes at least two frames and each portion is a patch of the carpet.

4. The method of claim 1, wherein the portions of the carpet of the fabrication etch process are fitted to the virtual carpet to identify a virtual frame number that is correlated to a predicted value for said metric.

5. The method of claim 1, wherein the training operation includes performing the etch process on a plurality of wafers, and for each wafer capturing a time series of spectra at individually sampled frame times, the captured time series of spectra at the individually sampled frame times defining a respective carpet, and each respective carpet being characterized by a polynomial with respective coefficients, wherein said virtual carpet has a corresponding polynomial with coefficients that is derived from the polynomials and respective coefficients of the carpets produced during said training operation.

6. The method of claim 5, wherein the coefficients of the respective polynomials of each carpet produced during training is processed by a dimensionality reduction algorithm to produce the virtual carpet, said virtual carpet is defined by a standardized polynomial with respective coefficients that are a superset of all coefficients obtained from carpets produced during said training operation.

7. The method of claim 6, wherein comparing the portions of the carpet of the fabrication etch process to the virtual carpet includes fitting the portions of the carpet to the virtual carpet in order to identify a virtual frame number from the virtual carpet, the virtual frame number being mapped to a predicted value for the metric, and said predicted value for the metric when substantially matched to a desired metric is indicative of etch endpoint.

8. The method of claim 1, wherein said training operation includes,

generating a plurality of carpets from sampled spectral data generated during etching of substrates, wherein each carpet is produced from a substrate being etched;

measuring or receiving data regarding a metric obtained for each of the etched substrates;

generating the virtual carpet from each of the plurality of carpets, the virtual carpet defined by a polynomial with coefficients produced by fitting a plurality of polynomials with respective coefficients of each of the plurality of carpets;

wherein at least some of the coefficients of the virtual carpet are floating parameters and others are fixed or coupled to the floating parameters, such that all of the coefficients of the polynomials of each of the plurality of carpets produced during said training operation are a subset of the coefficients of the virtual carpet.

9. The method of claim 8, wherein said floating parameters in addition to the virtual frame number are used to map to a predicted value for the metric, and said predicted value for the metric when substantially matched to a desired metric is indicative of etch endpoint.

10. The method of claim 1, wherein the time-series of spectra is intensity spectral data associated with broadband in-situ reflectometry, or is intensity spectral data associated with Optical Emission Spectroscopy (OES), or is ellipsometric spectral data associated with broadband in-situ ellipsometry, wherein said spectral data is collected from a chamber used for etching while a feature is being etch on a wafer.

11. A method for generating training data from a time-series of spectra generated during an etching process in a chamber, comprising,

etching a plurality of substrates in one or more chambers, wherein while the etching is processed, capturing a plurality of samples of frames of spectra, each captured frame of spectra identifying an intensity of the spectra as a function of wavelength;

measuring a metric associated with each substrate after said etching is complete by associating the measured metric to a last frame of spectra of the corresponding substrate;

generating a plurality of carpets, one for each of the substrates etched, wherein each of the carpets is defined by the plurality of frames of spectra and a polynomial with corresponding coefficients define characteristics of said each;

generating a virtual carpet by processing the plurality of carpets using a polynomial fitting algorithm, the virtual carpet is a superset of the plurality of carpets, such that every one of the plurality of carpets can be projected onto the virtual carpet to determine a virtual frame number in the virtual carpet;

correlating the virtual frame numbers of the virtual carpet to a predicted value of the metric;

wherein the virtual carpet is accessed by a controller during real-time processing of a substrate in order to determine when a current predicted value of the metric obtained from the virtual carpet corresponds to a predefined value for the metric that is indicative of endpoint of the etching process in the real-time processing.

12. The method of claim 11, wherein the metric associated with the etching is one of an etch depth, a critical dimension value, wafer bow value, a line width value, a feature pitch value, a features spacing value, or a measurable value.

13. The method of claim 12, wherein the controller is configured to connect to a database or a recipe file, to access the virtual carpet produced during said generating of training data for said real-time processing of the substrate.

14. The method of claim 12,

wherein during said real-time processing the controller is configured to be generating a carpet for the real-time processing;

wherein as the carpet is being generated, at least a portion or a patch of the carpet is used to compare to said virtual carpet to determine if the predefined value for the metric has been reached.

15. The method of claim 14, wherein said comparing includes performing a polynomial fitting of said portion or said patch of the carpet against hyper parameters of the virtual carpet, to enable correspondence to the virtual carpet and said carpet being produced during said real-time processing, wherein said virtual carpet is used to determine the virtual frame number, wherein the metric is determined using recipe parameters implemented during generation of the training data.

16. The method of claim 15, wherein said portion of the carpet includes a current frame of captured spectra and at least one previous frame of captured spectra, wherein using said at least one previous frame of captured spectra during said comparing to said virtual carpet enables accurate identification of a current state of said real-time processing of said substrate.

17. The method of claim 15, wherein the metric is a depth of etch, and the portion of the carpet is fitted to the virtual carpet to identify a virtual frame number that is correlated to a predicted depth of etch for the etching process.

22. The computer readable media of claim 17, wherein said training operation includes,

program instructions for generating a plurality of carpets from sampled spectral data generated during etching of substrates, wherein each carpet is produced from a substrate being etched;

program instructions for measuring or receiving data regarding a metric obtained for each of the etched substrates; and

program instructions for generating the virtual carpet from each of the plurality of carpets, the virtual carpet defined by a polynomial with coefficients produced by fitting a plurality of polynomials with respective coefficients of each of the plurality of carpets;

wherein at least some of the coefficients of the virtual carpet are floating parameters and others are fixed or coupled to the floating parameters, such that all of the coefficients of the polynomials of each of the plurality of carpets produced during said training operation are a subset of the coefficients of the virtual carpet.

18. Computer readable media having program instructions for using a time-series of spectra to identify endpoint of an etch process, comprising,

program instructions for accessing a virtual carpet that is generated from a time-series of spectra for the etch process collected during a training operation;

program instructions for running a fabrication etch process on a fabrication wafer, such that while the fabrication etch process is performed portions of a carpet defined from a time-series of spectral is generated for the fabrication etch process;

program instructions for comparing the portions of the carpet of the fabrication etch process to the virtual carpet; and

program instructions for processing end pointing of the fabrication etch process when said comparing indicates that a desired metric has been reached for the fabrication wafer.

19. The computer readable media of claim 18, wherein said portions of the carpet include a current frame of captured spectra and at least one previous frame of captured spectra, wherein each portion of said carpet includes at least two frames and each portion is a patch of the carpet.

20. The computer readable media of claim 18, wherein the portions of the carpet of the fabrication etch process are fitted to the virtual carpet to identify a virtual frame number that is correlated to a predicted value for said metric.

21. The computer readable media of claim 19, wherein the training operation includes performing the etch process on a plurality of wafers, and for each wafer capturing a time series of spectra at individually sampled frame times, the captured time series of spectra at the individually sampled frame times defining a respective carpet, and each respective carpet being characterized by a polynomial with respective coefficients, wherein said virtual carpet has a corresponding polynomial with coefficients that is derived from the polynomials and respective coefficients of the carpets produced during said training operation.

23. The computer readable media of claim 18, wherein the desired metric is associated with at least one of a predefined etch depth, a predefined critical dimension value, a predefined wafer bow value, a predetermined line width value, a predefined feature pitch value, a predefined features spacing value, or predetermined measurable value.