Method and process of implementing machine learning in complex multivariate wafer processing equipment

Patent metadata

Field	Value
Publication number	US-9972478-B2
Application number	US-201615268472-A
Country	US
Kind code	B2
Filing date	Sep 16, 2016
Priority date	Sep 16, 2016
Publication date	May 15, 2018
Grant date	May 15, 2018

Abstract

Methods and systems for controlling processing state of a plasma reactor to initiate processing of production substrates and/or to determine a ready state of a reactor after the reactor has been cleaned and needs to be seasoned for subsequent production wafer processing are provided. The method initiate processing of a substrate in the plasma reactor using settings for tuning knobs of the plasma reactor that are approximated to achieve desired processing state values. A plurality of data streams are received from the plasma reactor during the processing of the substrate. The plurality of data streams are used to identify current processing state values. The method includes generating a compensation vector that identifies differences between the current processing state values and the desired processing state values. The generation of the compensation vector uses machine learning to improve and refile the identification and amount of compensation needed, as identified in the compensation vector. The method further includes transforming the compensation vector into adjustments to the settings for the tuning knobs and then applying the adjustment to the tuning knobs of the plasma reactor.

First claim

The invention claimed is:

1. A method for controlling processing state of a plasma reactor, comprising:
initiate processing of a substrate in the plasma reactor using settings for tuning knobs of the plasma reactor that are approximated to achieve desired processing state values;
receiving a plurality of data streams from the plasma reactor during the processing of the substrate, the plurality of data streams used to identify current processing state values;
generating a compensation vector that identifies differences between the current processing state values and the desired processing state values;
transforming the compensation vector into adjustments to the settings for the tuning knobs;
applying the adjustment to the tuning knobs of the plasma reactor;
accessing a model database for a type of the plasma reactor and a type of a process to be used for the processing of the substrate in the plasma reactor; and
identifying a model from the model data, the model includes the settings for tuning knobs of the plasma reactor that are approximated to achieve desired processing state values.

2. The method of claim 1 , wherein the plurality of data streams are received from sensors that are interfaced with the plasma reactor or interfaced with a controller that is in communication with the plasma reactor, wherein the sensors include one or more of an optical emission spectrometry (OES) sensor, interferometry, a pressure sensor, a voltage sensor, a current sensor, a temperature sensor, a flow rate sensor, frequency sensor, a power sensor, a metrology sensor, and combinations of two or more thereof.

3. The method of claim 1 , wherein the tuning knobs relate to controls interfaced with the plasma reactor, the controls include one or more of controls to adjust gas flow meters, controls to adjust power settings, controls to adjust temperature settings, controls to adjust physical gap separations between top and bottom electrodes of the plasma reactor, controls for adjusting an electrostatic chuck (ESC) temperature or operation, controls for adjusting bias power settings, controls for setting chamber pressures, controls for setting frequencies of one or more radio frequency generators, controls for setting run time of specific recipe operations, controls for setting pumping rates of vacuum, controls for setting duration of gas flows, controls for setting monitoring algorithms, controls triggering borescope inspections, controls for setting or determining intervals between clean operations, or combinations of two or more thereof.

4. The method of claim 1 , wherein the desired processing state and the current processing state are defined in a virtual space that is descriptive of a physical state of plasma conditions sensed within a processing volume of the plasma reactor, the plasma conditions include a set of ion, electron and neutral fluxes at a plane of the substrate for a specific reactor wall surface condition.

5. The method of claim 4 , wherein the compensation vector identifies the differences between the current processing state values and the desired processing state values in the virtual space, and the transformation of the compensation vector identifies the adjustments to the settings for the knobs as a set of identified physical knobs having an identified physical adjustment, wherein a controller of the plasma reactor is configured to process program instructions that cause the adjustments of the settings to the knobs.

6. The method of claim 1 , further comprising,
continuing to receive the plurality of data streams from the plasma reactor during the processing of the substrate to produce the adjustments to the settings of the tuning knobs to assist in moving the current processing state values toward the desired processing state values.

7. The method of claim 6 , wherein a multivariate process is configured to identify the differences between the current processing state values and the desired processing state values, and further comprising,
processing machine learning to make adjustments to the desired processing state values to produce adjusted desired processing state values based, at least in part, on verification feedback received from one or both of etch rate performance or monitor wafer performance of the processing of the substrate.

8. The method of claim 7 , wherein the machine learning uses as input sensitivity data for the sensors of the plasma reactor, such that the produced compensation vector includes adjustments that are moderated based on sensitivity data.

9. The method of claim 1 , wherein the processing of the substrate is identified for a specific plasma reactor and a specific process recipe;
wherein each specific process recipe and each specific plasma reactor has an associated model that includes settings for tuning knobs and desired processing state values, the model is accessed from a model database.

10. The method of claim 9 , wherein as a model from the model database is used, a machine learning process makes adjustments to the settings for the tuning knobs of the model to improve settings to the specific plasma reactor to achieve the desired processing state values.

11. The method of claim 10 , further comprising,
updating the models in the model databased based on the adjustments made by the machine learning process.

12. A method for controlling processing state of a plasma reactor, comprising:
initiate processing of a seasoning substrate in the plasma reactor using settings for tuning knobs of the plasma reactor that are approximated to achieve desired processing state values;
receiving a plurality of data streams from the plasma reactor during the processing of the seasoning substrate, the plurality of data streams used to identify current processing state values;
continuing to process the seasoning substrate or additional seasoning substrates in the plasma reactor until the current processing state values are within a predefined threshold of the desired processing state values;
generating a compensation vector that identifies differences between the current processing state values and the desired processing state values;
transforming the compensation vector into adjustments to the settings for the tuning knobs;
applying the adjustment to the tuning knobs of the plasma reactor;
identifying the plasma reactor as ready for processing of production substrates when the adjustment to the tuning knobs have placed the plasma reactor in a state that substantially produces the desired processing state values;
initiate processing of a production substrate in the plasma reactor using settings for tuning knobs of the plasma reactor as set during the seasoning of the substrate that are approximated to achieve desired processing state values;
receiving a plurality of data streams from the plasma reactor during the processing of the production substrate, the plurality of data streams used to identify current processing state values;
generating a compensation vector that identifies differences between the current processing state values and the desired processing state values;
transforming the compensation vector into adjustments to the settings for the tuning knobs; and
applying the adjustment to the tuning knobs of the plasma reactor to compensate for drift in processing by the plasma reactor over time.

13. The method of claim 12 , wherein the desired processing state and the current processing state are defined in a virtual space that is descriptive of a physical state of plasma conditions sensed within a processing volume of the plasma reactor, the plasma conditions include a set of ion, electron and neutral fluxes at a p

Assignees

• Lam Res Corp

Inventors

• Guha Joydeep
• Daugherty John
• Vahedi Vahid
• Gottscho Richard Alan

Classifications

• H10P72/0604

  Process monitoring, e.g. flow or thickness monitoring · CPC title

• H10P72/0602

  Temperature monitoring · CPC title

• H10P72/0421

  for drying etching · CPC title

• H10P72/06

  Apparatus for monitoring, sorting, marking, testing or measuring · CPC title

• H10P50/242

  of Group IV materials · CPC title