TOOL PERFORMANCE BY LINKING SPECTROSCOPIC INFORMATION WITH TOOL OPERATIONAL PARAMETERS AND MATERIAL MEASUREMENT INFORMATION

• US 20120185813A1
• Filed: 01/19/2011
• Published: 07/19/2012
• Est. Priority Date: 01/19/2011
• Status: Active Grant

• Current Assignee: Tokyo Electron Limited
• Original Assignee: Tokyo Electron Limited
• Inventors: Kaushal, Sanjeev, Patel, Sukesh Janubhai, Sugishima, Kenji
• Granted Patent: US 8,954,184 B2
• Time in Patent Office: Days
• US Class Current: 716/112
• CPC Class Codes: G05B 19/41875 characterised by quality surveillance of production
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: H01J 37/32926 Software, data control or modelling
• 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 22/20 Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps
• 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
• CPC Class Codes: H01L 2924/00 Indexing scheme for arrangements or methods for connecting or disconnecting semiconductor or solid-state bodies as covered by H01L24/00
• CPC Class Codes: H01L 2924/0002 Not covered by any one of groups H01L24/00, H01L24/00 and H01L2224/00
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

System(s) and method(s) are provided for adjustment and analysis of performance of a tool through integration of tool operational data and spectroscopic data related to the tool. Such integration results in consolidated data that enable, in part, learning at least one relationship amongst selected portions of the consolidated data. Learning is performed autonomously without human intervention. Adjustment of performance of the tool relies at least in part on a learned relationship and includes generation of process recipe parameter(s) that can adjust a manufacturing process in order to produce a satisfactory tool performance in response to implementation of the manufacturing process. A process recipe parameter can be generated by solving an inverse problem based on the learned relationship. Analysis of performance of the tool can include assessment of synthetic performance scenarios, identification of spectroscopic condition(s) that affect performance, and extraction of endpoints based at least on time dependence spectroscopic data.

Claims

1. A method facilitated by at least one processor of a computing system, comprising:

integrating disparate information related to a semiconductor production tool into a data structure, wherein the disparate information includes at least spectroscopic intensity information, sensor information, and measurement information;

learning at least one relationship that characterizes a value of a first variable of the data structure in terms of values of one or more disparate variables in the data structure; and

optimizing performance of the semiconductor production tool with the at least one relationship.

2. The method of claim 1, wherein the sensor information includes measurements of physical quantities and measurements of maintenance information,

where the measurements of physical quantities includes measurements of at least one of pressure, gas flow, or power, and

where the measurements of maintenance information includes measurements of elapsed time since last wet-clean and age of tool parts.

3. The method of claim 1, wherein learning the at least one relationship further comprising:

generating an expression for an intensity of a spectral line as a function of a set of gas flows corresponding to respective gases; and

identifying mass flow control (MFC) drift and shift errors when a predicted value of spectral intensity does not match a measured value of spectral intensity

4. The method of claim 1, wherein learning the at least one relationship further comprising generating an expression for the first variable as a function of the one or more disparate variables,

wherein the first variable is a wafer output characteristic and the one or more disparate variables relate to one or more of tool sensor measurements, spectral intensity measurements, or wafer input characteristics.

5. The method of claim 1, further comprising selecting the first variable of the data structure.

6. The method of claim 1, wherein learning the at least one relationship further comprises employing a genetic algorithm to generate the at least one relationship.

7. The method of claim 1, wherein optimizing performance of the semiconductor production tool further comprises:

receiving a set of controllable parameter expressions, wherein a controllable parameter expression, in the set of controllable parameter expressions, specifies a minimum value of a controllable parameter, a maximum value of the controllable parameter, and a minimal value by which the controllable parameter can change;

receiving a set of target values, wherein a target value in the set of target values specifies a value for a wafer output characteristic; and

identifying parameters for the semiconductor production tool based upon the at least one relationship, the set of controllable parameter expressions, and the set of target values.

8. The method of claim 7, further comprising:

creating a recipe that includes the identified parameters,

wherein identifying the parameters further comprises employing an inverse problem solving technique.

9. The method of claim 1, further comprising identifying one or more influential variables of the data structure which contribute to favorable wafer yields.

10. The method of claim 9, further comprising:

receiving status information, wherein the status information specifies a state of each chamber of the semiconductor production tool as one of a good chamber or a poor chamber; and

integrating the status information into the data structure.

11. The method of claim 10, further comprising autonomously learning a Boolean-valued function that predicts the state of a chamber of the semiconductor production tool based upon at least one of the spectroscopic intensity information in the data structure or the sensor information in the data structure.

12. The method of claim 11, wherein autonomously learning the Boolean-valued function further comprises restricting the Boolean-valued function to a fixed size, in terms of a number of parameters.

13. The method of claim 11, further comprising:

for a parameter of the Boolean-valued function;

ordering chambers of the semiconductor production tool by corresponding values of the parameter;

ordering the chambers of the semiconductor production tool by yield performance; and

identifying the parameter as an influential variable when the order of the chambers by values of the parameter matches the order of the chambers by yield performance.

14. The method of claim 1, further comprising identifying a pair of wavelengths to employ in automated endpoint detection.

15. The method of claim 14, further comprising:

identifying a first set of wavelengths whose corresponding intensities in the spectroscopic intensity information increase during a etch step performed by the semiconductor production tool;

identifying a second set of wavelengths who corresponding intensities in the spectroscopic intensity information decrease during the etch step; and

selecting a first wavelength from the first set of wavelengths and a second wavelength form the second set of wavelengths, wherein the first wavelength and the second wavelength are employed for automated endpoint detection.

16. The method of claim 15, further comprising:

augmenting intensity values, from the spectroscopic intensity information and associated with a plurality of wafers, corresponding to the first wavelength and the second wavelength with state information that specifies a state of each of the plurality of wafers; and

autonomously learning a Boolean-valued function that predicts the state of a wafer based upon intensity values associated with the wafer and corresponding to the first wavelength and the second wavelength.

17. The method of claim 16, further comprising halting the etch step of the semiconductor production tool when the Boolean-valued function predicts a change in the state of the wafer from a poor state to a good state.

18. A system for a semiconductor production tool, comprising:

a learning engine configured to receive sensor information, spectral intensity information, and device measurement information related to the semiconductor production tool, to integrate the received information into a unified training matrix, and to utilize the unified training matrix to generate an expression that predicts a value for an output characteristic of a wafer processed by the semiconductor production tool based upon values of one or more disparate values included in the sensor information, spectral intensity information, or device measurement information.

19. The system of claim 18, further comprising a tuning engine configured to:

obtain a set of controllable parameter expressions, wherein a controllable parameter expression, in the set of controllable parameter expressions, specifies a minimum value of a controllable parameter, a maximum value of the controllable parameter, and a minimal value by which the controllable parameter can change;

obtain a set of target values, wherein a target value in the set of target values specifies a value for the wafer output characteristic; and

select values for the controllable parameters in accordance with the controllable parameter expressions, wherein the selected values, according to the expression generated by the learning engine, achieve the target values in the set of target values.

20. A computer-readable storage medium storing computer-executable instructions that, when executed by at least one processor, cause a computer to:

consolidate disparate information related to a semiconductor production tool into a data structure, wherein the disparate information includes at least spectroscopic intensity information, sensor information, and measurement information;

generate at least one relationship that characterizes a value of a first variable of the data structure in terms of values of one or more disparate variables in the data structure; and

optimize performance of the semiconductor production tool with the at least one relationship.