Method to predict yield of a device manufacturing process

Patent metadata

Field	Value
Publication number	US-2022011728-A1
Application number	US-201917293373-A
Country	US
Kind code	A1
Filing date	Oct 30, 2019
Priority date	Dec 3, 2018
Publication date	Jan 13, 2022
Grant date	—

Abstract

A method for predicting yield relating to a process of manufacturing semiconductor devices on a substrate, the method including: obtaining a trained first model which translates modeled parameters into a yield parameter, the modeled parameters including: a) a geometrical parameter associated with one or more selected from: a geometric characteristic, dimension or position of a device element manufactured by the process and b) a trained free parameter; obtaining process parameter data including data regarding a process parameter characterizing the process; converting the process parameter data into values of the geometrical parameter; and predicting the yield parameter using the trained first model and the values of the geometrical parameter.

First claim

1 . A method comprising:
obtaining a trained first model which translates modeled parameters into a yield parameter relating to a process of manufacturing semiconductor devices on a substrate, the modeled parameters comprising: a) one or more geometrical parameters associated with one or more selected from: a geometric characteristic, dimension or position of a device element manufactured by the process and b) one or more trained free parameters; obtaining process parameter data comprising data of one or more process parameters characterizing the semiconductor manufacturing process; converting the process parameter data into values of the one or more geometrical parameters; and predicting the yield parameter using the trained first model and the values of the one or more geometrical parameters.

2 . The method according to claim 1 , wherein the yield parameter is a measure of probability that a device, or element thereof, manufactured by the process will function.

3 . The method according to claim 1 , wherein the one or more geometrical parameters comprise one or more selected from: Critical Dimension (CD), overlay, sidewall angle or any other parameter associated with a geometric profile of the device element.

4 . The method according to claim 1 , wherein the one or more process parameters relate to one or more selected from: overlay, alignment, leveling, focus or dose.

5 . The method according to claim 1 , wherein the process parameter data relates to variation in an alignment and/or overlay tree, such that the yield parameter is dependent on the overlay tree and/or alignment tree.

6 . The A method according to claim 1 , wherein the trained first model relates to one or more device elements of a device manufactured by the process, which are identified as being sensitive to process variations.

7 . The method according to claim 1 , further comprising constructing the first model prior to it being trained to obtain the trained first model, the constructing comprising determining a yield probability model for each of one or more device elements of a device manufactured by the process, each yield probability model comprising a geometric formulation which translates the modeled parameters into a yield probability for its corresponding device element.

8 . The method according to claim 7 , wherein each yield probability model comprises one or more free parameters relating to one or more statistical parameters of a probability function.

9 . The method according to claim 1 , further comprising using the first model to generate second training data for training a second model, the second training data comprising one or more predicted yield parameters based on previous process parameter data.

10 . The method according to claim 9 , wherein the second model comprises a neural network.

11 . The method according to claim 1 , further comprising performing a control action for control of the semiconductor manufacturing process based on the predicted yield parameter.

12 . A method for constructing a first model for predicting a yield parameter based on process parameter data, the method comprising:
determining a yield probability model for each device element of a plurality of device elements of a device manufactured by a manufacturing process, each yield probability model comprising a geometric formulation which translates modeled parameters into a yield probability for its corresponding device element, the modeled parameters comprising one or more geometrical parameters and one or more trained free parameters; and constructing the first model as a combination of the yield probability models for each of the device elements.

13 . The method according to claim 12 , wherein each yield probability model comprises one or more free parameters relating to one or more statistical parameters of a probability function.

14 .- 15 . (canceled)

16 . A computer program product comprising a non-transitory computer-readable medium having instructions therein, the instructions, upon execution by a computing system, configured to cause the computing system to at least:
obtain a trained first model which translates modeled parameters into a yield parameter relating to a process of manufacturing semiconductor devices on a substrate, the modeled parameters comprising: a) one or more geometrical parameters associated with one or more selected from: a geometric characteristic, dimension or position of a device element manufactured by a semiconductor manufacturing process and b) one or more trained free parameters; obtain process parameter data comprising data of one or more process parameters characterizing the semiconductor manufacturing process; convert the process parameter data into values of the one or more geometrical parameters; and predict the yield parameter using the trained first model and the values of the one or more geometrical parameters.

17 . The computer program product according to claim 16 , wherein the instructions are further configured to cause the computing system to construct the first model prior to it being trained to obtain the trained first model, the instructions to construct the first model are further configured to cause the computing system to determine a yield probability model for each of one or more device elements of a device manufactured by the semiconductor manufacturing process, each yield probability model comprising a geometric formulation which translates the modeled parameters into a yield probability for its corresponding device element.

18 . The computer program product according to claim 16 , wherein the instructions are further configured to cause the computing system to use the first model to generate second training data for training a second model, the second training data comprising one or more predicted yield parameters based on previous process parameter data.

19 . The computer program product according to claim 18 , wherein the second model comprises a neural network.

20 . The computer program product according to claim 16 , wherein the instructions are further configured to cause the computing system to cause performance of a control action for control of the semiconductor manufacturing process based on the predicted yield parameter.

21 . A computer program product comprising a non-transitory computer-readable medium having instructions therein, the instructions, upon execution by a computing system, configured to cause the computing system to at least:
determine a yield probability model for each device element of a plurality of device elements of a device manufactured by a manufacturing process, each yield probability model comprising a geometric formulation which translates modeled parameters into a yield probability for its corresponding device element, the modeled parameters comprising one or more geometrical parameters and one or more trained free parameters; and construct a first model for predicting a yield parameter based on process parameter data, as a combination of the yield probability models for each of the device elements.

22 . The computer program product according to claim 21 , wherein each yield probability model comprises one or more free parameters relating to one or more statistical parameters of a probability function.

Assignees

• Asml Netherlands Bv

Inventors

• Zhang Youping
• Menchtchikov Boris
• Tabery Cyrus Emil
• Zou Yi
• Lin Chenxi
• Cheng Yana
• Hastings Simon Philip Spencer
• Genin Maxime Philippe Frederic

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06F30/27

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G05B13/042

  in which a parameter or coefficient is automatically adjusted to optimise the performance · CPC title