Semiconductor substrate yield prediction based on spectral data from multiple substrate dies

By generating spectral data arrays and using a CNN to predict yield and identify defects, the method addresses the challenge of inefficient substrate yield prediction and defect detection in semiconductor manufacturing, enhancing accuracy and reducing waste.

JP2024543118AActive Publication Date: 2024-11-19ONTO INNOVATION INC
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2024530045
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2021-11-22
Filing Date
2022-11-21
Publication Date
2024-11-19
Estimated Expiration
2042-11-21

AI Technical Summary

Technical Problem

Existing semiconductor substrate manufacturing processes face challenges in accurately predicting yield and identifying defects during and after multiple fabrication steps, leading to inefficiencies and increased waste due to imperfect inspection methods.

Method used

The method involves generating spectral data for a subset of dies on a semiconductor substrate using inspection tools like reflectometers, combining this data into arrays or matrices, and using a convolutional neural network (CNN) to predict yield and identify defect-causing tools or processes by analyzing spectral data at various manufacturing stages.

Benefits of technology

This approach enhances the accuracy of yield prediction and defect identification, allowing for timely corrections and reducing waste by optimizing the inspection process, thereby improving manufacturing efficiency and reducing errors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024543118000001_ABST
    Figure 2024543118000001_ABST
Patent Text Reader

Abstract

A system and method for improving the fabrication of a substrate is provided. Subsets of the dies of a substrate can be inspected at various points in the fabrication process to generate spectral data. The spectral data can be used to generate data that is input into a machine learning model to predict the yield of the substrate.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] (Priority Claim) This application was filed as a PCT international patent application on November 21, 2022, and claims the benefit of and priority to U.S. Non-provisional Patent Application No. 17 / 532,700, filed on November 22, 2021, the entire disclosure of which is incorporated herein by reference.

[0002] FIELD OF THEINVENTION The present disclosure relates to semiconductor component manufacturing and metrology. [Background technology]

[0003] As part of the formation of a semiconductor chip or other type of integrated circuit (IC), a semiconductor substrate is manufactured or fabricated. The final IC components may be built into the substrate through a series of fabrication steps. The fabrication steps may include deposition steps in which thin film layers are added onto the substrate. The substrate may then be coated with photoresist and a circuit pattern from a reticle may be projected onto the substrate using lithography techniques. An etching process may then be performed. At each fabrication step, the tools performing the fabrication step may introduce defects or imperfections into the substrate.

[0004] In some applications, the semiconductor substrate consists of dies. A die is a block of semiconductor material (e.g., silicon) on which a given functional circuit is fabricated. For example, the functional circuit may take the form of a central processing unit. The functional circuit may be batch fabricated on a single substrate. The substrate is then cut into multiple dies, each of which contains one copy of the circuit. Each die may be rectangular. Thin, non-functional spaces may be provided between the dies, allowing the individual dies to be cut (e.g., with a saw) from the substrate without damaging the circuit. Summary of the Invention

[0005] TECHNICAL FIELD The present disclosure relates generally to inspection of semiconductor substrates during manufacturing of the substrates. One example of a semiconductor substrate is a semiconductor wafer.

[0006] In accordance with some aspects, the present disclosure relates to inspecting a subset of dies of a semiconductor substrate during manufacturing of the substrate.

[0007] In accordance with some other aspects, the present disclosure relates to inspecting a subset of dies of a semiconductor substrate after multiple manufacturing steps during manufacture of the substrate.

[0008] In accordance with some other aspects, the present disclosure relates to inspecting a subset of dies of a semiconductor substrate by generating spectral data for each of the dies of the subset.

[0009] According to some other aspects, the present disclosure relates to generating combined data including spectral data of inspected dies of a semiconductor substrate. In some examples, the combined data can include an array of distinct data subsets, each subset representing spectral data from one of the inspected dies. In some examples, the combined data can be represented as a matrix.

[0010] According to some other aspects, the present disclosure relates to generating combined data including spectral data for inspected dies of a semiconductor substrate inspected after each of a plurality of manufacturing steps of the substrate. As used herein, terms such as manufacturing and fabrication or fabricating are used interchangeably.

[0011] In accordance with some other aspects, the present disclosure relates to predicting yield of a semiconductor substrate based on spectral data of inspected dies of the substrate.

[0012] According to some other aspects, the present disclosure relates to predicting yield of a semiconductor substrate by inputting spectral data of inspected dies of the substrate into a trained machine learning model.

[0013] According to some other aspects, the present disclosure relates to predicting yield of a semiconductor substrate by processing combined data including spectral data of inspected dies of the substrate using a convolutional neural network.

[0014] According to some other aspects, the present disclosure relates to predicting a yield of a semiconductor substrate by processing combined data including spectral data of inspected dies of the substrate inspected after each of multiple manufacturing steps of the substrate using a convolutional neural network.

[0015] In accordance with some other aspects, the present disclosure relates to identifying tools or processes that cause defects in a semiconductor substrate.

[0016] According to some other aspects, the present disclosure relates to any combination of the aforementioned features.

[0017] In accordance with some particular aspects, the present disclosure relates to a method for predicting a yield of a substrate, the method including combining first spectral data for each of a plurality of dies on the substrate and second spectral data for each of the plurality of dies of the substrate into combined data, where the first data is generated from a first inspection of the plurality of dies at a first time point and the second data is generated from a second inspection of the plurality of dies at a second time point, and providing the combined data as an input to a trained machine learning model to generate a predicted yield for the substrate.

[0018] In accordance with certain other aspects, the present disclosure relates to a non-transitory computer-readable medium including instructions to cause a computing device to combine first spectral data for each of a plurality of dies on a substrate and second spectral data for each of the plurality of dies of the substrate into combined data, where the first data is generated from a first inspection of the plurality of dies at a first time and the second data is generated from a second inspection of the plurality of dies at a second time, and generate a predicted yield for the substrate including providing the combined data as an input to a trained machine learning model.

[0019] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Additional aspects, features and / or advantages of examples will be set forth in part in the description that follows, and in part will be obvious from the description, or may be learned by practice of the disclosure.

[0020] Non-limiting and non-exhaustive examples are described with reference to the following figures: [Brief description of the drawings]

[0021] [Figure 1] 1 illustrates an exemplary semiconductor substrate map representing a semiconductor substrate, the substrate including dies, a subset of the dies being selected for inspection according to the present disclosure. [Diagram 2] 2 illustrates an example plot of spectral data for one of the dies under test of FIG. 1. [Diagram 3] 4 illustrates a portion of an example matrix of spectral data that can be used to generate a substrate yield prediction according to the present disclosure. [Figure 4] 5A-5C show some further example matrices of spectral data that can be used to generate a substrate yield prediction according to the present disclosure. [Diagram 5]1 illustrates an exemplary three-dimensional structure of spectral data that can be used to generate a substrate yield prediction according to the present disclosure. [Figure 6] 1 illustrates a schematic diagram of an exemplary system for predicting yield of a semiconductor substrate according to the present disclosure. [Figure 7] 7 illustrates an example of using a machine learning model to predict substrate yield according to an example of the present disclosure using the system of FIG. [Figure 8] A method for predicting the yield of semiconductor substrates using the system of FIG. 6 will be described. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Examples of the present disclosure describe systems and methods for improving substrate fabrication. One example of such a substrate is a semiconductor wafer. After the substrate is processed by at least one fabrication tool, an inspection device inspects a subset of the substrate's dies to identify defects in the substrate. Based on inspection data generated from the inspection process, a yield of the substrate is predicted. The yield of the substrate may refer to the percentage of the substrate's dies that meet one or more defined operational, quality, or other acceptability criteria.

[0023] In some examples, the predicted yield may be used to identify tools or processes used to manufacture the substrate that may be causing manufacturing defects in the substrate.

[0024] In some examples, the predicted yield can be used to improve the production of semiconductor substrates based on the predicted yield (e.g., by adjusting or replacing tools that cause the defects) to compensate for the defects.

[0025] One exemplary inspection method for inspecting a substrate, and in particular for inspecting dies of a substrate, includes generating spectral data for each of a subset of the dies of the substrate. The spectral data may be generated using an inspection device or tool, such as a reflectometer or spectrometer, that detects and measures the wavelength spectrum of light or other radiation reflected by that die. Thus, in some examples, the inspection tool is configured to generate a multispectral image of the substrate.

[0026] For example, a substrate may have multiple dies, and those dies may be inspected at different stages of the manufacturing process, such as when a new layer is added to the substrate. It is not necessary for each die on the substrate to be inspected as part of this process. An exemplary substrate map 10 is shown generally in FIG. The substrate map 10 represents a semiconductor substrate during the substrate's fabrication. The exemplary substrate map 10 shows the locations of a subset of the substrate's dies 12 that may be inspected for purposes of predicting the substrate's yield in accordance with the present disclosure. Dies on the substrate map other than die 12, such as die 14, are not inspected in this example.

[0027] The number and locations of dies inspected to predict the yield of a substrate can be selected to optimize one or more factors. For example, inspecting fewer dies may be more cost- and / or time-efficient, but provide a larger margin of error in the yield prediction. Similarly, inspecting dies that are relatively densely packed on a substrate may be more cost- and / or time-efficient than inspecting dies that are scattered across the substrate, but again provide a larger margin of error in the yield prediction for the entire substrate. Thus, the number and locations of dies inspected can be selected to minimize cost and time and maximize accuracy and precision of the yield prediction using a convolutional neural network (CNN) or other type of machine learning model. In some examples, the number of dies selected can be a predetermined percentage of the total number of dies on the substrate, such as within a range of about 1 / 50,000 to about 1 / 1,000. A percentage of the total outside this range can also be selected.

[0028] A CNN or other machine learning model can be trained (e.g., using supervised training) using spectral data from dies of substrates having known yields to predict an unknown yield of a target substrate based on the spectral data from inspection of the dies of the target substrate.

[0029] In some examples, the number of dies inspected on a target substrate for purposes of predicting the yield of the target substrate is equal to the number of dies on each substrate used to train the CNN or other machine learning model.

[0030] In some examples, the locations of the die inspected on a target substrate for purposes of predicting the yield of the target substrate are the same as the locations of the die on each substrate used to train the CNN or other machine learning model.

[0031] In some examples, the number and / or location of the dies inspected on the target substrate may differ from the number and / or location of the dies on the training substrate.

[0032] 1, spectral data is generated for each of the die 12 under test. The spectral data for each of the die 12 under test can be one-dimensional serial data from different wavelengths of electromagnetic radiation (e.g., different wavelengths of light). For example, for each wavelength, the data represents a signal strength.

[0033] Other metrology tools, such as acoustic or ultrasonic metrology tools, can be used to inspect the dies of the training substrate and generate spectral data for the dies 12 of the target substrate.

[0034] In some examples, the spectral data may be obtained as a function of another spectral attribute, such as frequency rather than wavelength.

[0035] FIG. 2 illustrates an exemplary plot 18 of visible light spectrum data for a single die 12 of a substrate represented by map 10 of FIG. 1, with the vertical axis representing signal intensity of a signal reflected by die 12 and the horizontal axis representing wavelength. Plot 18 may include multiple measurements made by an inspection device (e.g., a reflectometer or spectrometer) at each measurement wavelength. Each such signal measurement at a given wavelength and at a given time for a given die after a given fabrication process (i.e., at a given inspection time point) may be considered a channel. Plot 18 may include reflected signal intensity for each of multiple channels at each of multiple reflected wavelengths (e.g., 1000 or more wavelengths). Multiple channels include signal measurements made at different times at the same wavelength for the same die after the same substrate fabrication process. Thus, although "inspection time point" is used herein to refer to inspection of a die at a single inspection time point after a given fabrication process, multiple channels of data may be acquired to improve reliability of the signal measurements, with each channel corresponding to a different measurement time point but the same inspection time point. That is, at each test time point, the spectral data can be measured multiple times, each time corresponding to a separate channel.

[0036] In some examples, one or more statistical operations may be applied to the signal intensities detected across multiple channels at a given wavelength to arrive at a single signal intensity value per wavelength, such as the median or average of the signal intensities per wavelength per die per inspection time point across all channels. In some examples, if the variance of the signal intensities across multiple channels at a given wavelength is determined to exceed a maximum threshold variance, the spectral data or a portion thereof is not used and / or the inspection is performed again.

[0037] Spectral data is generated for each die 12 inspected on the substrate. For example, if spectral data from 12 dies 12 of a substrate are to be used by a machine learning model to predict the yield of the substrate, then a set of 12 sets of spectral data is generated, one for each die 12 inspected.

[0038] The number of 12 dies is used merely as an example number, and other numbers of dies may be used. For example, the number of dies selected for inspection can be based on optimizing different factors. For example, inspecting more dies may improve the substrate yield prediction, but at the same time, it may reduce the efficiency of the inspection process because it requires more time to detect the spectrum and more computational power to train and apply the machine learning model that provides the prediction. Thus, for example, the number of dies selected for inspection to generate a predicted substrate yield can be based on optimizing the error margin of the prediction as a function of the number or percentage of dies to be inspected. For example, the number of dies selected for inspection of a given substrate can be the minimum number of dies that produces a substrate yield prediction with an error margin equal to or less than a predetermined threshold error margin.

[0039] The locations on the substrate of the die selected for inspection may be selected based on any number of factors. For example, the locations may be selected to provide a minimum number of die required for inspection such that the substrate yield projection falls within a threshold margin of error. In some examples, the die locations selected for inspection are selected randomly, for example, using a random number generator or other randomization tool.

[0040] The spectral data of the inspected dies are combined into combined data that can be provided as input to a machine learning model, such as a CNN, to predict the yield of the substrate. An example of such combined data is a two-dimensional matrix. The yield of the substrate may be the percentage of dies on the substrate that pass final testing (i.e., the percentage of dies on the substrate that have an acceptable predetermined operational or quality criterion). More specifically, the spectral data is correlated with the presence or absence of die defects that correlate with the substrate yield. Non-limiting examples of such defects may include uneven thickness of the deposition layer, cracks, improper dimensions, contamination, etc.

[0041] A substrate can be fabricated using multiple processes and tools. Fabrication tools can include, for example, oxidation systems, epitaxial reactors, diffusion systems, ion implanters, physical vapor deposition systems, chemical vapor deposition systems, photolithography tools, and etching tools, among other types of tools. The number of manufacturing steps employed to fabricate a substrate can vary, in some examples, from a small number (e.g., less than 10) to several hundred. Typically, a substrate can be fabricated by progressing through a fabrication line, with each fabrication tool in the fabrication line performing a process step on the substrate. The same tool can be used multiple times in the fabrication line. For example, multiple deposition, lithography, and / or etching steps can be performed to fabricate a substrate.

[0042] In some examples, the spectral data may also be correlated with the type of die defect that correlates with the substrate yield. For example, some defects may have more or less impact on the acceptability of the substrate than other types of defects. If the type of defect is known, the type of defect may be weighted according to its impact on the acceptability of the substrate when using the machine learning model to predict the substrate yield. In some examples, the type of defect may be determined based on the fabrication process performed immediately prior to the substrate inspection. For example, if the spectral data is acquired after a deposition process, to the extent that the spectral data indicates a defect in the substrate die, the defect may be a deposition type defect and / or a defect associated with the deposition tool. An appropriate weighting factor may then be applied by the machine learning model to the detected deposition type defect when using the spectral data to predict the substrate yield.

[0043] A machine learning model, such as a CNN, can be trained using the correlation to predict the yield of a target substrate based on the spectral data of the substrate's dies. For example, the machine learning model may be trained using a training substrate with known yield and predict the yield of the target substrate by comparing the spectral data of the target substrate's dies (such as the plot shown in FIG. 2) with the corresponding (e.g., of the corresponding dies and / or after the same fabrication process) spectral data of the training substrate with known yield. In this case, the machine learning model can classify the yield of the target substrate using the compared spectral data of the training substrate with known yield to provide an output yield prediction for the target substrate.

[0044] The spectral data can be combined into any suitable number of combined arrays, and the arrays can be provided as input to a machine learning model, such as a CNN. For example, the combined data can include a data subset for each die that represents the spectral data corresponding to that die. The data subsets in the combined data array can be organized such that each data subset is distinct from every other data subset in the array. Each data value in each data subset is associated with a wavelength of the corresponding spectral data for the corresponding die and represents a signal intensity at the associated wavelength of the corresponding spectral data for the corresponding die.

[0045] Non-limiting examples of combined data arrays organized in such data subsets include matrices. Figure 3 illustrates a portion of an exemplary matrix 30 of spectral data that may be input into a machine learning model to generate a target substrate yield prediction according to the present disclosure. Matrix 30 is an exemplary arrangement or structure of distinct combined data subsets corresponding to the spectral data.

[0046] Matrix 30 contains spectral data for the dies of a target substrate after fabrication steps of the target substrate. All of the spectral data in matrix 30 is taken at time 1 or inspection time 1, meaning that all of the data in matrix 30 is taken after a given fabrication step and before the next fabrication step or the next other step that physically modifies the substrate.

[0047] The completed matrix 30 (or other array of combined spectral data) may be provided as an input to a machine learning model to predict substrate yield. In some examples, the matrix 30 (or other array of combined spectral data) is assigned one or more attributes that the machine learning model can use to more accurately predict substrate yield. For example, the matrix 30 (or other array of combined spectral data) may be assigned a fabrication process attribute that identifies the type of fabrication process that was performed immediately prior to obtaining the spectral data included in the matrix 30 (or other array of combined spectral data).

[0048] Each horizontal row of matrix 30 represents a different inspected die of the target substrate after the fabrication process. For example, if 12 dies are to be inspected, matrix 30 has 12 rows and M is 12. Thus, in one example, a first row 32 of matrix 30 may represent all of the spectral data for a first inspected die 12 (FIG. 1), a second row 34 of matrix 30 may represent all of the spectral data for a second inspected die 12 (FIG. 1), and so on. Each inspected die has a location on the substrate, i.e., a die position, that is unique to that die.

[0049] Each vertical column of matrix 30 represents a different spectral wavelength. For example, a first column 36 can represent a first wavelength, a second column 38 can represent a second wavelength different from the first wavelength, and so on. In other examples, each column represents a different spectral frequency.

[0050] Each cell of the matrix 30 contains the reflected signal intensity of the corresponding die measured at the corresponding wavelength at the corresponding inspection time. Thus, for example, a value x in a cell 40 corresponding to a row 32 and a column 36 of the matrix 30 corresponds to a reflected signal intensity of a first inspected die 12 (FIG. 1) measured at a first wavelength at inspection time 1. A value y in a cell 42 corresponding to a row 32 and a column 38 of the matrix 30 corresponds to a reflected signal intensity of a first inspected die 12 measured at a second wavelength at time 1. A value z in a cell 44 corresponding to a row 34 and a column 36 of the matrix 30 corresponds to a reflected signal intensity of a second inspected die measured at a first wavelength at time 1 for each inspected die 12 (FIG. 1) of the substrate and each inspected wavelength of the spectrum, and so on.

[0051] In some examples, matrices generated according to the present disclosure can be assembled into a three-dimensional structure with a third dimension corresponding to an inspection channel, such that each cell of each matrix contains a value corresponding to a spectral signal intensity at a given wavelength in a given channel for a given die at a given inspection time point. In some examples, the three-dimensional structure is converted into a two-dimensional matrix by performing one or more statistical operations on the data across multiple channels for a given die, a given wavelength, and a given inspection time point, for example, as described above.

[0052] The number of wavelengths (or, in some examples, frequencies) to be inspected can be, for example, about 10, about 100, or about 1,000 or more. In some examples, there are about 1,000 wavelengths to be inspected for each spectrum (N is about 1,000), each die at each inspection time point is associated with about 1,000 spectral data points, and matrix 30 includes about 1,000 columns, with each data point being the signal intensity measured by a reflectometer or spectrometer at a given wavelength.

[0053] The completed (e.g., each cell contains signal intensity data) matrix 30, or other array of the combined spectral data, may be provided as an input to a machine learning model to generate a prediction of the yield of the substrate. In some examples, the machine learning model may be trained via a supervised training method based on previous matrices (or other arrays of the combined spectral data) from substrates with known yields. In some examples, the machine learning model is a convolutional neural network. In some examples, the training matrix is ​​the same size as matrix 30.

[0054] The matrix (or other array of combined spectral data) input to the machine learning model may be augmented with different fabrication steps. That is, another set or sets of spectral data may be generated by inspecting the same die of the substrate at a second inspection time point and optionally at one or more other subsequent inspection time points after one or more other fabrication steps. For example, inspection of 12 dies may be performed after a first layer of the substrate is fabricated. The spectral data generated from inspection at time point 1 may then be used to generate the 12-row matrix 30 (FIG. 3) as described above.

[0055] A second inspection of the 12 dies may be performed at inspection time 2 (or time 2) after another layer of the substrate is fabricated or another fabrication step (e.g., etching) is performed and before the next fabrication step or other step that physically alters the substrate. The matrix 30 (FIG. 3) generated from the first inspection may be extended with the spectral data from the second inspection to generate extended matrix 60 (FIG. 4). For example, another 12 rows may be added to matrix 30 to generate matrix 60, one additional row for each die to be inspected. Thus, the first row 62 of matrix 60 represents the spectral data of the first die after the first fabrication step (i.e., at inspection time 1), and the thirteenth row 64 of matrix 60 represents the spectral data of the first die after the second fabrication step (i.e., at inspection time 2).

[0056] In some examples, the same die are inspected at the first inspection time point and the second inspection time point. In some examples, the same number of die are inspected at the first inspection time point and the second inspection time point, but the die inspected at the second inspection time point need not all be the same as the die inspected at the first inspection time point.

[0057] The augmented matrix 60 (or other augmented array of the differentiated combined data subset) may be provided as an input to a machine learning model (e.g., a CNN) to predict the yield of the substrate after a second manufacturing step. The process can continue with matrix augmentation occurring after each of any desired number of manufacturing steps and corresponding inspections of the substrate. Thus, an updated yield prediction can be made after each manufacturing step using the further augmented data. As each augmentation is fed into the machine learning model, the machine learning model compares the input to a training matrix (or other data array) with the corresponding augmentation level after the same manufacturing step for the target substrate.

[0058] As an alternative to extending matrix 30 in FIG. 3 to include spectral data after another manufacturing step, a separate matrix can be generated for each inspection time point and input into the machine learning model as a separate matrix that is compared to the corresponding training matrix.

[0059] As a further alternative to extending matrix 30 of FIG. 3 to include spectral data after another manufacturing step, matrix 60 may be generated alone after both manufacturing steps without first generating matrix 30.

[0060] In some examples, a first predicted yield for the substrate is generated by inputting the matrix 30 (or other array of the combined spectral data) into a machine learning model. Then, another manufacturing process is performed on the substrate and another inspection is performed. Then, the spectral data from the other inspection is added to the matrix 30 (or other array of the combined spectral data) to generate an augmented matrix 60 (or other augmented array of the combined spectral data). Then, the augmented matrix (or other augmented array of the combined spectral data) is input into the machine learning model to modify the first predicted yield and output a second predicted yield for the substrate, which is a modification of the first predicted yield based on the augmentation of the matrix (or other array of the combined spectral data).

[0061] A matrix (e.g., matrix 30) (or other array of combined spectral data) or an augmented matrix (e.g., matrix 60) (or other augmented array of combined spectral data) having spectral data for the dies of the target substrate is input to the machine learning model, which performs one or more operations (e.g., matrix operations) on the input to classify the input. The one or more operations can generate one or more values ​​that can be compared to corresponding values ​​for training substrates having known yields to output a predicted yield for the target substrate.

[0062] A matrix (eg, matrix 30) or an expanded matrix (eg, matrix 60) may be constructed in a data file using delimiters to separate columns and rows.

[0063] By using the disclosed systems and methods to predict substrate yield during the manufacturing of substrates, advantageous efficiencies can be obtained. For example, manufacturing tools or processes that cause defects can be identified and corrected or resolved sooner, for example, before completing the manufacturing of a substrate or batch of substrates. In addition, an approximate number of substrates that may be affected by a defect can be determined relatively early in the manufacturing process.

[0064] Additionally, by re-evaluating the substrate yield prediction after each of multiple manufacturing steps of a substrate, the accuracy of the substrate yield prediction can be improved, which can further result in improved efficiency, minimization of unnecessary scrapping of acceptable substrates, and increased efficient scrapping of substrates that may have been given an unjustified pass score based on less inspection data. For example, the impact of a defect after a given manufacturing step may be affected (e.g., reduced or magnified) after a subsequent manufacturing step. Embodiments of the present disclosure are adapted to capture and evaluate such impacts and refine the substrate yield prediction using an extended spectral data matrix (or other data array) as described herein.

[0065] FIG. 5 illustrates an exemplary three-dimensional structure 70 of spectral data that can be used to generate a substrate yield prediction according to the present disclosure. The three-dimensional structure 70 includes a series of two-dimensional matrices, each of which includes spectral data from a die of a substrate acquired at a different time after the same fabrication process. That is, each two-dimensional matrix 30 represents data from a different channel. With reference to FIG. 5, the (x) axis corresponds to the spectral wavelength, as do the rows of each matrix 30 (FIG. 3). The (y) axis is a spatial axis in that it corresponds to a different die on the substrate, as do the columns of each matrix 30 (FIG. 3). The (z) axis is a channel axis, with each channel corresponding to a measurement made at a different time.

[0066] In some examples, the two-dimensional matrix of the three-dimensional structure 70 corresponds to a plurality of matrices 30 (FIG. 3), each of which is populated with signal intensity measurements taken at different time points, i.e., different channels, for a single inspection time point.

[0067] The completed three-dimensional structure 70 (e.g., with all cells of all two-dimensional matrices 30 containing signal intensity data) may be provided as input to a machine learning model to generate a prediction of the substrate yield. In some examples, the machine learning model may be trained via a supervised training method based on previous three-dimensional structures from substrates with known yields. In some examples, the machine learning model is a convolutional neural network. In some examples, the training structure is the same size as structure 70 (e.g., the same number and size of two-dimensional matrices).

[0068] FIG. 6 illustrates a schematic diagram of an exemplary system 100 for predicting yield of semiconductor substrates according to the present disclosure.

[0069] The system 100 includes a computing device 202. The computing device 202 may be a server and / or other computing device that performs operations described herein, such as the substrate yield prediction operations described herein. The computing device 202 may include a computing component 206. The computing component 206 includes at least one processor 208 and a memory 204. The memory 204 may include a non-transitory computer-readable medium. Depending on the exact configuration, the memory 204 (which stores, among other things, instructions for performing the substrate yield prediction instructions and other operations disclosed herein) may be volatile (such as RAM), non-volatile (such as ROM, flash memory, etc.), or some combination of the two. Additionally, the server 202 may also include storage (removable storage 210 and / or non-removable storage 212), including, but not limited to, a solid-state device, a magnetic or optical disk, or a tape. Additionally, computing device 202 may have input devices 216, such as a touch screen, keyboard, mouse, pen, voice input, etc., and / or output devices 214, such as a display, speakers, printer, etc. One or more communications connections 218, such as a local area network (LAN), a wide area network (WAN), point-to-point, Bluetooth, RF, etc., may also be incorporated into computing device 202.

[0070] System 100 may include one or more die inspection devices 102 in operative communication, such as being connected via a network, with a computing device 202. A non-limiting example of a die inspection device 102 is a reflectometer or spectrometer that measures the intensity of light or other waves (e.g., sound waves) reflected from a substrate at different wavelengths to generate spectral data, such as the spectral data shown in FIG.

[0071] In an alternative configuration, one or more components of the computing device 202 reside locally on one or more die inspection equipment 102. For example, one or more die inspection equipment 102 may itself be configured to perform one or more of the substrate yield operations described herein. That is, substrate yield prediction instructions may be executed directly on one or more die inspection equipment 102.

[0072] Figure 7 illustrates an example of using a machine learning model to predict substrate yield according to examples of the present disclosure using the system of Figure 6. In some examples, one or more of the operations performed in Figure 7 may be performed by the computing device 202 and / or one or more die inspection equipment 102 of Figure 6.

[0073] 7, input 302 is provided to a machine learning model 308, such as a CNN. The CNN is configured to analyze the input to generate an output 310 that may be provided to a technician (e.g., displayed on a display device) using the computing device 202 of FIG.

[0074] The inputs 302 include training data (e.g., a training matrix) and known yields 304, as well as test or subject data (e.g., a test or subject matrix) 306. The training data can include expanded and / or unaugmented combinations of spectral data for training substrates (e.g., in the form of a matrix as described above) with known substrate yields corresponding to each training data input.

[0075] The test data (eg, test matrix) 306 may be an expanded or unaugmented combination of spectral data for target substrates (eg, in the form of a matrix as described above) with unknown yields.

[0076] The machine learning model 308 is trained using the training data and the known yield 304. The machine learning model 308 runs on the test data 306 using the training matrix and the known yield 304 to generate an output 310.

[0077] The output 310 includes a predicted yield 312 for the target substrate corresponding to the test data 306. In some examples, the predicted yield may also include an indication as to whether the predicted yield passes or fails one or more predetermined operational acceptabilities, quality acceptabilities, or other acceptabilities.

[0078] In some examples, the output 310 may include one or more correction recommendations 314. For example, if the predicted yield 312 correlates with a reject score for the substrate, one or more corrective actions may be suggested. The recommended correction may be based on, for example, the extent to which the predicted yield does not reach a minimum acceptable threshold yield, the most recent manufacturing run performed, and / or other factors. For example, the correction recommendation 314 may be to recalibrate a substrate manufacturing tool that was used immediately prior to generating the test data 306. In another example, the correction recommendation 314 may be to discard the target substrate or a partially formed target substrate. Other correction recommendations 314 are possible.

[0079] Figure 8 illustrates a method 400 for predicting yield of semiconductor substrates using the system of Figure 6. It will be understood that different embodiments of the present disclosure can include different combinations of a subset of the steps of method 400, non-limiting examples of which are described herein.

[0080] In step 402 of method 400, a machine learning model is trained using spectral data (e.g., a spectral data matrix) generated by inspecting (e.g., using die inspection equipment 102 (FIG. 6)) dies of a training substrate having a known yield.

[0081] In step 404 of method 400, a target substrate manufacturing process is performed.

[0082] In step 406, a subset of the dies of the target substrate are inspected using, for example, die inspection equipment 102 (FIG. 6).

[0083] In step 408, the spectral data is combined (e.g., in the form of a matrix, such as spectral data matrix 30 (FIG. 3)) based on the testing performed in step 406. In some examples, step 408 includes accessing the spectral data generated in step 406 and generating combined spectral data using the accessed spectral data.

[0084] In some examples, following step 408, the method 400 proceeds to step 416, where a machine learning model uses the spectral data generated in step 408 and the training data obtained in step 402 to predict the yield of the target substrate. In some examples, following step 408, the method 400 proceeds to step 410.

[0085] In step 410, another manufacturing process is performed on the target substrate.

[0086] In step 412, a subset of the dies of the same target substrate as in step 406 is inspected using, for example, die inspection equipment 102 (FIG. 6).

[0087] In step 414, extended combined spectral data (e.g., an extended spectral data matrix, such as matrix 60 (FIG. 4) or matrix 70 (FIG. 5)) is generated based on the testing performed in step 412. In some examples, step 414 includes accessing the spectral data generated in step 412 and generating extended spectral data using the accessed spectral data generated in step 412.

[0088] In some examples, following step 414, the method sequentially repeats steps 410, 412, and 414 one or more times to further expand the spectral data after further manufacturing steps of the target substrate.

[0089] Following step 414 or the last instance of step 414, in step 416, the machine learning model uses the spectral data generated in the most recently performed iteration of step 414 and the training data obtained in step 402 to predict the yield of the target substrate. In some examples, steps 410, 412, 414, and 416 are repeated to predict the yield of the target substrate after each fabrication step or after a certain number of fabrication steps.

[0090] In other examples, a predicted yield may be generated after step 408, and in some examples, the predicted yield may be revised following one or more instances of step 414.

[0091] The embodiments described herein may be employed using software, hardware, or a combination of software and hardware to implement and execute the systems and methods disclosed herein. Although specific devices are recited as performing certain functions throughout this disclosure, those skilled in the art will appreciate that these devices are provided for illustrative purposes and that other devices may be used to perform the functions disclosed herein without departing from the scope of the present disclosure. In addition, some aspects of the present disclosure are described above with reference to block diagrams and / or operational diagrams of systems and methods according to aspects of the present disclosure. The functions, operations, and / or acts noted in the blocks may occur in a different order than the order shown in any respective flowchart. For example, two blocks shown in succession may in fact be executed or performed substantially simultaneously or in the reverse order, depending on the functionality and implementation involved.

[0092] This disclosure describes some embodiments of the technology with reference to the accompanying drawings, in which only some of the possible embodiments are shown. However, other aspects may be embodied in many different forms and should not be construed as being limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of possible embodiments to those skilled in the art. Furthermore, as used in the specification and claims, the phrase "at least one of element A, element B, or element C" is intended to convey any of element A, element B, element C, elements A and B, elements A and C, elements B and C, and elements A, B, and C. Furthermore, those skilled in the art will understand the extent to which terms such as "about" or "substantially" convey in light of the measurement techniques utilized herein. Unless such terms can be clearly defined or understood by those skilled in the art, the term "about" shall mean plus or minus 10%.

[0093] Although specific embodiments are described herein, the scope of the technology is not limited to those specific embodiments. Moreover, although different examples and embodiments may be described separately, such embodiments and examples may be combined with each other in implementing the technology described herein. Those skilled in the art will recognize other embodiments or improvements that are within the scope and spirit of the technology. Thus, specific structures, operations, or media are disclosed only as exemplary embodiments. The scope of the technology is defined by the following claims and any equivalents thereof.

Claims

1. 1. A method for predicting substrate yield, comprising: combining first spectral data for each of a plurality of dies on the substrate and second spectral data for each of the plurality of dies of the substrate into combined data, the first data being generated from a first inspection of the plurality of dies at a first time and the second data being generated from a second inspection of the plurality of dies at a second time; generating a predicted yield for the substrate, comprising providing the combined data as an input to a trained machine learning model; A method comprising:

2. The method of claim 1 , wherein the machine learning model is a convolutional neural network.

3. The method of claim 1 , wherein the combined data includes a first data subset representing the first spectral data for a first die of the plurality of dies.

4. 4. The method of claim 3, wherein the combined data is organized such that the first data subset is distinct from a second data subset representing the second spectral data for a second die of the plurality of dies.

5. The method of claim 4 , wherein each data value of the first data subset is associated with a wavelength of the first spectral data.

6. 6. The method of claim 5, wherein the values ​​in the first data subset represent signal intensities of the first spectral data at the wavelengths corresponding to the values ​​for the dies of the plurality of dies.

7. the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; the second point in time occurs after the second manufacturing step is performed on the substrate. The method of claim 1.

8. expanding the combined data by adding third spectral data for each of the plurality of dies, the third spectral data being generated from a third inspection of the die, to the combined data to form expanded combined data; generating another predicted yield for the substrate, comprising providing the augmented combined data as an input to the trained machine learning model; Further comprising: the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; the second point in time occurs after the second manufacturing step is performed on the substrate and before a third manufacturing step is performed on the substrate; the third point in time occurs after the third manufacturing step is performed on the substrate. The method of claim 1.

9. The method of claim 1 , wherein the die is less than all of the die on the substrate.

10. the combination data is extended combination data, the projected yield is a revised projected yield; The method further comprises, prior to the second test, generating unenhanced combined data including the first spectral data; providing the unaugmented combined data as an input to the trained machine learning model to generate a first predicted yield for the substrate; Further comprising: The method further comprises, after the second inspection, modifying the first predicted yield to generate the modified predicted yield. Further comprising: The method of claim 1.

11. generating modification recommendations based on said predicted yield; The method of claim 1 further comprising:

12. The method of claim 11 , wherein the modification recommendations include recommendations to adjust a substrate manufacturing tool.

13. A computing device, combining first spectral data for each of a plurality of dies on the substrate and second spectral data for each of the plurality of dies of the substrate into combined data, the first data being generated from a first inspection of the plurality of dies at a first time and the second data being generated from a second inspection of the plurality of dies at a second time; generating a predicted yield for the substrate, comprising providing the combined data as an input to a trained machine learning model; A non-transitory computer-readable medium comprising instructions to cause

14. 14. The non-transitory computer-readable medium of claim 13, wherein the machine learning model is a convolutional neural network.

15. the combined data includes a first data subset representing the first spectral data for a first die of the plurality of dies; the combined data is organized such that the first data subset is distinct from a second data subset representing the second spectral data for a second die; each data value of the first data subset is associated with a wavelength of the first spectral data; the value represents a signal intensity of the first spectral data for the first die of the plurality of dies at the wavelength corresponding to the value.

14. The non-transitory computer-readable medium of claim 13.

16. the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; the second point in time occurs after the second manufacturing step is performed on the substrate.

14. The non-transitory computer-readable medium of claim 13.

17. The computing device includes: expanding the combined data by adding third spectral data for each of the plurality of dies, the third spectral data being generated from a third inspection of the die, to the combined data to form expanded combined data; generating another predicted yield for the substrate, comprising providing the augmented combined data as an input to the trained machine learning model; further instructions configured to cause the the first time occurs after a first manufacturing step is performed on the substrate and before a second manufacturing step is performed on the substrate; the second point in time occurs after the second manufacturing step is performed on the substrate and before a third manufacturing step is performed on the substrate; the third point in time occurs after the third manufacturing step is performed on the substrate.

14. The non-transitory computer-readable medium of claim 13.

18. The non-transitory computer-readable medium of claim 13 , wherein the die is less than all of the die on the substrate.

19. the combination data is extended combination data, the projected yield is a revised projected yield; The non-transitory computer readable medium may further include, prior to the second test, generating unenhanced combined data including the first spectral data; providing the unaugmented combined data as an input to the trained machine learning model to generate a first predicted yield for the substrate; further instructions configured to cause the The non-transitory computer readable medium may further include, after the second testing, modifying the first predicted yield to generate the modified predicted yield. and further instructions configured to cause the 14. The non-transitory computer-readable medium of claim 13.

20. The computing device includes: generating modification recommendations based on said predicted yield; 14. The non-transitory computer-readable medium of claim 13, further comprising instructions configured to:

21. The non-transitory computer-readable medium of claim 20 , wherein the modification recommendations include recommendations to adjust a substrate manufacturing tool.

Citation Information

Patent Citations

  • Improved process control techniques for semiconductor manufacturing processes

    JP2017536584A

  • Model-based optical metrology of semiconductor structures using anisotropic dielectric constants

    JP2019532518A

  • Data Augmentation for Defect Inspection Based on Convolutional Neural Networks

    JP2020501154A

  • Dynamic yield prediction

    US20150253373A1

  • Methods of inspecting substrates and semiconductor fabrication methods incorporating the same

    US20170200658A1