Substrate defect analysis

Abstract

A method includes identifying property data of a substrate processed by a substrate processing system. The method further includes identifying a plurality of regions of the substrate corresponding to a first defect category based on a first subset of the property data. The method further includes finely classifying the plurality of regions of the substrate corresponding to the first defect category into a plurality of defect subcategories based on a second subset of the property data. The method further includes causing performance of a corrective action associated with the substrate processing system based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subclasses.

Description

Technical Field

[0001] The present disclosure relates to defect analysis, and more particularly to substrate (such as wafer) defect analysis and root cause analysis. Background Art

[0002] Manufacturing equipment is used to produce products (such as substrates). For example, semiconductor substrate processing equipment is used to produce semiconductor substrates (e.g., semiconductor substrates with integrated circuits). Summary of the Invention

[0003] The following is a simplified summary of the present disclosure to provide a basic understanding of some aspects of the present disclosure. This summary is not an extensive overview of the present disclosure. It is not intended to identify key or important elements of the present disclosure, nor is it intended to delineate any scope of specific implementations of the present disclosure or any scope of the claims. Its sole purpose is to present some concepts of the present disclosure in a simplified form as a prelude to the more detailed description presented later.

[0004] In one aspect of the present disclosure, a method includes identifying property data of a substrate processed by a substrate processing system. The method further includes identifying, based on a first subset of the property data, a plurality of regions of the substrate corresponding to a first defect category. The method further includes subclassifying, based on a second subset of the property data, the plurality of regions of the substrate corresponding to the first defect category into a plurality of defect subcategories. The method further includes causing the execution of a correction action associated with the substrate processing system based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subcategories.

[0005] In another aspect of the present disclosure, a non-transitory computer-readable storage medium stores instructions that, when executed, cause a processing device to perform operations. The operations include identifying property data of a substrate processed by a substrate processing system. The operations further include identifying, based on a first subset of the property data, a plurality of regions of the substrate corresponding to a first defect category. The operations further include subclassifying, based on a second subset of the property data, the plurality of regions of the substrate corresponding to the first defect category into a plurality of defect subcategories. The operations further include causing the execution of a correction action associated with the substrate processing system based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subcategories.

[0006] In another aspect of the present disclosure, a system includes a memory and a processing device coupled to the memory. The processing device is configured to identify property data of a substrate processed by a substrate processing system. The processing device is further configured to identify, based on a first subset of the property data, a plurality of regions of the substrate corresponding to a first defect class. The processing device is further configured to subclassify, based on a second subset of the property data, the plurality of regions of the substrate corresponding to the first defect class into a plurality of defect subclasses. The processing device is further configured to cause the execution of a correction action associated with the substrate processing system based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subclasses. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The present disclosure is illustrated by way of example and not limitation in the figures of the accompanying drawings.

[0009] Figure 1 is a block diagram illustrating an exemplary system architecture in accordance with certain embodiments.

[0010] Figure 2 illustrates a dataset generator for creating a dataset for a machine learning model in accordance with certain embodiments.

[0011] Figure 3 is a block diagram illustrating the determination of prediction data in accordance with certain embodiments.

[0012] Figure 4A is a block diagram illustrating defect classes in accordance with certain embodiments.

[0013] Figure 4B is a block diagram illustrating defect subclasses in accordance with certain embodiments.

[0014] Figure 4C is a block diagram associated with a defect subclass in accordance with certain embodiments.

[0015] Figure 4D is a block diagram illustrating substrate defects in accordance with certain embodiments.

[0016] Figures 5A - 5C is a flowchart of a method associated with defect analysis in accordance with certain embodiments.

[0017] Figure 6 is a block diagram illustrating a computer system in accordance with certain embodiments. DETAILED DESCRIPTION

[0018] Techniques for substrate defect analysis (e.g., defect source tracing and defect root cause identification and / or correction action recommendation) are described herein.

[0019] Fabrication equipment uses fabrication parameters to produce products. For example, substrate processing equipment uses fabrication parameters (such as temperature, pressure, etc.) during substrate processing operations (such as layer deposition, etching, etc.) to produce substrates. As a result of one or more operations, abnormalities (such as defects) may occur in the substrate (such as a finished substrate, a partially processed substrate). A substrate with an abnormality may have performance data that does not meet a threshold (for example, a defective wafer). This will result in substrate discard, poor substrate performance, reduced yield, waste of materials and energy, and so on.

[0020] Conventionally, an actual substrate or a test substrate is processed by substrate processing equipment and then manually inspected to try to identify defects and determine the source and root cause of the defects in order to reduce material exposure and improve the average time for equipment repair and recovery. Manual inspection is time-consuming, may be inaccurate, depends on the user performing the inspection, can damage the substrate processing equipment, and uses more energy and materials. Manually attempting to determine the defect source, root cause, and corrective actions associated with the defects is both time-consuming and inaccurate. This can lead to reduced throughput, production interruptions, substrates produced with performance data that does not meet the threshold, and so on.

[0021] The devices, systems, and methods disclosed herein provide substrate defect analysis (for example, defect source tracing and defect root cause identification, including suggestions for corrective actions to improve the average time for equipment repair).

[0022] A processing device identifies property data of a substrate processed by a substrate processing system. In some embodiments, the property data is metrology data received from metrology equipment.

[0023] Based on a first subset of the property data, the processing device identifies regions of the substrate corresponding to a first defect category. In some examples, the first subset of the property data includes scanning electron microscope (SEM) images, energy-dispersive x-ray microanalysis (EDX) images, and so on. In some examples, the first defect category includes broken lines, unfilled, bridging, stains, scratches, glass damage, foreign objects (such as particles), residues, resist collapse, via stress, cavities, pits, crystal defects, cracks, and so on.

[0024] The processing device further classifies a plurality of regions of the substrate corresponding to the first defect category into defect sub-categories based on a second subset of the property data. In some examples, the second subset of the property data includes morphological data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), and so on. In some examples, the defect sub-categories include spherical particles, random particles, rod-shaped particles, flake particles, post imprint fall-on particles, annular pits, micro-pits, macro-pits, mouse bites, chemical mechanical polishing (CMP) bridges, photoresist bridges, micro-bridges, organic stains, inorganic stains, lithography pinches, step bunching, stack tomography, shallow triangles, obtuse triangles, surface triangles, down falls, punctures, skip marks, crescents, microtubes, photoluminescence (PL) rings, basal plane dislocations, protrusions, mounds, and so on.

[0025] The processing device causes the execution of a correction action associated with the substrate processing system based on at least one of the plurality of defect sub-categories. In some embodiments, the correction action includes providing an alert, causing a cleaning process, causing a repair process, causing replacement of a substrate processing equipment part, causing further inspection, causing computational process control (CPC), causing statistical process control (SPC) (e.g., SPC compared with a 3-sigma chart, etc.), causing advanced process control (APC), causing model-based process control, causing preventive operational maintenance, causing design optimization, updating manufacturing parameters, causing wafer recipe modification, causing feedback control, causing machine learning modification, and so on.

[0026] Aspects of the present disclosure provide technical advantages. The present disclosure avoids the time, inaccuracy, and subjectivity of conventional systems. Compared with conventional solutions, the substrates produced by the present disclosure have property data that better meets the thresholds, less damage to the substrate processing equipment, higher throughput, fewer production interruptions, reduced use of test wafers, and so on.

[0027] Although some embodiments of the present disclosure relate to defects in substrate processing equipment and substrates, in some embodiments, the present disclosure can also be applied to other types of manufacturing equipment, other types of products, and other types of anomalies.

[0028] Figure 1According to certain embodiments, a block diagram of an exemplary system 100 (exemplary system architecture) is shown. The system 100 (e.g., the calibration action component 122 and / or the prediction component 114) may perform the methods described herein (e.g., Figures 5A to 5C methods 500A to 500C). The system 100 includes a client device 120, a manufacturing apparatus 124, sensors 126, metrology equipment 128, a prediction server 112, and a data store 140. In some embodiments, the prediction server 112 is part of a prediction system 110. In some embodiments, the prediction system 110 further includes server machines 170 and 180.

[0029] In some embodiments, one or more of the client device 120, the manufacturing apparatus 124, the sensors 126, the metrology equipment 128, the prediction server 112, the data store 140, the server machine 170, and / or the server machine 180 are coupled to each other via a network 130 for generating prediction data 160 to perform defect source tracing and defect root cause identification. In some embodiments, the network 130 is a public network that provides the client device 120 with access to the prediction server 112, the data store 140, and other publicly available computing devices. In some embodiments, the network 130 is a private network that provides the client device 120 with access to the manufacturing apparatus 124, the sensors 126, the metrology equipment 128, the data store 140, and other privately available computing devices. In some embodiments, the network 130 includes one or more wide area networks (WANs), local area networks (LANs), wired networks (e.g., Ethernet networks), wireless networks (e.g., 802.11 networks or Wi-Fi networks), cellular networks (e.g., Long Term Evolution (LTE) networks), routers, hubs, switches, server computers, cloud computing networks, and / or combinations of the foregoing.

[0030] In some embodiments, the client device 120 includes a computing device such as a personal computer (PC), a laptop computer, a mobile phone, a smartphone, a tablet computer, a netbook computer, etc. In some embodiments, the client device 120 includes a calibration action component 122. In some embodiments, the calibration action component 122 may also be included in the prediction system 110 (e.g., a machine learning processing system). In some embodiments, the calibration action component 122 is alternatively included in the prediction system 110 (e.g., instead of being included in the client device 120). The client device 120 includes an operating system that allows a user to perform one or more of the following operations: merge, generate, view, or edit data, provide instructions to the prediction system 110 (e.g., a machine learning processing system), etc.

[0031] In some embodiments, the corrective action component 122 receives one or more of the following: user input (e.g., via a graphical user interface (GUI) that is displayed via the client device 120), property data 142, performance data 152, etc. In some embodiments, the corrective action component 122 transmits data (e.g., user input, property data 142, performance data 152, etc.) to the prediction system 110, receives prediction data 160 from the prediction system 110, determines a corrective action based on the prediction data 160, and causes the corrective action to be implemented. In some embodiments, the corrective action component 122 stores data (e.g., user input, property data 142, performance data 152, etc.) in the data store 140, and the prediction server 112 retrieves the data from the data store 140. In some embodiments, the prediction server 112 stores the output of the trained machine learning model 190 (e.g., prediction data 160) in the data store 140, and the client device 120 retrieves the output from the data store 140. In some embodiments, the corrective action component 122 receives an indication of a corrective action (e.g., based on the prediction data 160) from the prediction system 110, and causes the execution of the corrective action.

[0032] In some embodiments, the prediction data 160 is associated with a corrective action. In some embodiments, the corrective action is associated with one or more of the following: cleaning one or more parts of the manufacturing equipment 124 (e.g., a processing chamber), repairing one or more parts of the manufacturing equipment 124, replacing one or more parts of the manufacturing equipment 124, computational process control (CPC), statistical process control (SPC) (e.g., SPC compared with a 3-sigma chart, etc.), advanced process control (APC), model-based process control, preventive operational maintenance, design optimization, updating manufacturing parameters, wafer recipe modification, feedback control, machine learning modification, and so on. In some embodiments, the corrective action includes providing an alert (e.g., providing a warning not to use one or more parts of the manufacturing equipment 124 if the prediction data 160 indicates a predicted anomaly). In some embodiments, the corrective action includes providing feedback control (e.g., cleaning, repairing, and / or replacing one or more parts of the manufacturing equipment 124 in response to prediction data 160 indicating a predicted anomaly). In some embodiments, the corrective action includes providing machine learning (e.g., causing a modification of one or more parts of the manufacturing equipment 124 based on the prediction data 160).

[0033] In some embodiments, the prediction server 112, the server machine 170, and the server machine 180 each include one or more computing devices, such as rack servers, router computers, server computers, personal computers, mainframe computers, laptop computers, tablet computers, desktop computers, graphics processing units (GPUs), application-specific integrated circuits (ASICs) for accelerators (e.g., tensor processing units (TPUs)), and the like.

[0034] The prediction server 112 includes a prediction component 114. In some embodiments, the prediction component 114 receives the property data 142 of the substrate (e.g., received from the client device 120, retrieved from the data store 140), and generates prediction data 160 associated with the execution of the correction action (e.g., defect analysis, defect source tracing, defect root cause identification, defect translation, defect evolution, etc.). In some embodiments, the prediction component 114 uses one or more trained machine learning models 190 to determine the prediction data 160. In some embodiments, the trained machine learning model 190 is trained using the historical property data 144 and the historical performance data 154.

[0035] In some embodiments, the prediction system 110 (e.g., the prediction server 112, the prediction component 114) generates the prediction data 160 using supervised machine learning (e.g., supervised data sets, historical property data 144 labeled with historical performance data 154, etc.). In some embodiments, the prediction system 110 uses semi-supervised learning to generate the prediction data 160 (e.g., semi-supervised data sets, the performance data 152 is a prediction percentage, etc.). In some embodiments, the prediction system 110 uses unsupervised machine learning (e.g., unsupervised data sets, clusters, clusters based on the historical property data 144, etc.) to generate the prediction data 160.

[0036] In some embodiments, the manufacturing equipment 124 (e.g., cluster tool, wafer backgrinding system, wafer saw equipment, die attach machine, wire bonder, die coating system, molding equipment, hermetic sealing equipment, metal can welding machine, DTFS machine, branding equipment, lead finishing equipment, etc.) is part of a substrate processing system (e.g., an integrated processing system). The manufacturing equipment 124 includes one or more of the following: a controller, a housing system (e.g., a substrate carrier, a front opening unified pod (FOUP), an automatic teach FOUP, a process kit housing system, a substrate housing system, a cassette, etc.), a side storage pod (SSP), an aligner device (e.g., an aligner chamber), a factory interface (e.g., an equipment front end module (EFEM)), a loadlock, a transfer chamber, one or more processing chambers, a robotic arm (e.g., disposed in the transfer chamber, disposed in the front end interface, etc.), and so on. The housing system, SSP, and loadlock mounted to the factory interface and the robotic arm disposed in the factory interface are used to transfer contents (e.g., substrates, process kit rings, carriers, verification wafers, etc.) between the housing system, SSP, loadlock, and factory interface. The aligner device is disposed in the factory interface to align the contents. The loadlock and processing chambers mounted to the transfer chamber and the robotic arm disposed in the transfer chamber are used to transfer contents (e.g., substrates, process kit rings, carriers, verification wafers, etc.) between the loadlock, processing chambers, and transfer chamber. In some embodiments, the manufacturing equipment 124 includes components of the substrate processing system. In some embodiments, the property data 142 of the substrate is obtained from a substrate that has undergone one or more processes (e.g., etching, heating, cooling, transferring, processing, flowing, etc.) performed by components of the manufacturing equipment 124.

[0037] In some embodiments, the sensor 126 provides the property data 142 (e.g., sensor values such as historical sensor values and current sensor values) of the substrate processed by the manufacturing equipment 124. In some embodiments, the sensor 126 includes one or more of the following: an imaging sensor (e.g., SEM, camera, imaging device, etc.), a pressure sensor, a temperature sensor, a flow sensor, a spectral sensor, and so on. In some embodiments, the property data 142 is used for equipment health status and / or product health status (e.g., product quality). In some embodiments, the property data 142 is received over a period of time.

[0038] In some embodiments, sensor 126 and / or metrology tool 128 provide property data 142, which includes one or more of the following: image data, SEM image, EDX image, morphology data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, or defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), temperature, pitch, current, power, voltage, and so on.

[0039] In some embodiments, the property data includes an SEM image (e.g., an image captured by scanning the surface of a substrate with a focused electron beam using a scanning electron microscope to generate a high-resolution image). In some embodiments, the property data includes an EDX image (e.g., an image generated from data collected using X-ray technology to identify the elemental composition of a material). In some embodiments, the property data includes morphology data (e.g., data related to the form of a substrate). In some embodiments, the property data includes size attribute data (e.g., data describing the attribute size of a substrate). In some embodiments, the property data includes dimensional attribute data (e.g., data describing the attribute dimensions of a substrate). In some embodiments, the property data includes defect distribution data (e.g., data describing the distribution of defects on a substrate (e.g., spatial distribution, temporal distribution, etc.)). In some embodiments, the property data includes spatial location data (e.g., data describing the spatial location of attributes, defects, components, etc. of a substrate). In some embodiments, the property data includes elemental analysis data (e.g., data describing the elemental composition of a substrate). In some embodiments, the property data includes wafer signature data (e.g., data describing the distribution of wafer defects in a substrate originating from a single manufacturing issue). In some embodiments, the property data includes chip layer data (e.g., a specific layer or operation in a substrate manufacturing process). In some embodiments, the property data includes chip layout data (e.g., data describing the layout of a substrate in planar geometry). In some embodiments, the property data includes edge data (e.g., data describing the edge of a wafer (such as a notch edge, wafer edge thickness, wafer bowing, and / or warping)). In some embodiments, the property data includes defect metadata, including but not limited to grayscale data (e.g., data describing the pixel brightness of a substrate image) and signal-to-noise ratio data (e.g., data describing the signal-to-noise ratio of a substrate measured using, for example, spectroscopic equipment).

[0040] In some embodiments, property data 142 (e.g., historical property data 144, current property data 146, etc.) is processed (e.g., by client device 120 and / or by prediction server 112). In some embodiments, processing of property data 142 includes generating features. In some embodiments, a feature is a pattern in property data 142 (e.g., slope, width, height, peak, etc.) or a combination of values from property data 142 (e.g., power derived from voltage and current, etc.). In some embodiments, property data 142 includes features used by prediction component 114 to obtain prediction data 160.

[0041] In some embodiments, metrology equipment 128 (such as imaging equipment, spectroscopic equipment, ellipsometry equipment, etc.) is used to determine metrology data (such as inspection data, image data, spectroscopic data, ellipsometry data, material composition, optical or structural data, etc.) corresponding to a substrate produced by manufacturing equipment 124 (such as substrate processing equipment). In some examples, after manufacturing equipment 124 processes a substrate, metrology equipment 128 is used to inspect a portion (such as a layer) of the substrate. In some embodiments, metrology equipment 128 performs scanning acoustic microscopy (SAM), ultrasonic inspection, x-ray inspection, and / or computed tomography (CT) inspection. In some examples, after manufacturing equipment 124 deposits one or more layers on a substrate, metrology equipment 128 is used to determine the quality of the processed substrate (e.g., thickness of a layer, uniformity of a layer, interlayer spacing of layers, etc.). In some embodiments, metrology equipment 128 includes an imaging device (e.g., SAM equipment, ultrasonic equipment, x-ray equipment, CT equipment, etc.). In some embodiments, property data 142 includes sensor data from sensor 126 and / or metrology data from metrology equipment 128. In some embodiments, performance data 152 includes user input via client device 120 and / or metrology data from metrology equipment 128. Property data 142 may include metrology data from a first subset of metrology equipment 128, and performance data 152 may include metrology data from a second subset of metrology equipment 128.

[0042] In some embodiments, data store 140 is a memory (e.g., random access memory), a drive (e.g., hard disk drive, flash drive), a database system, or another type of component or device capable of storing data. In some embodiments, data store 140 includes multiple storage components (e.g., multiple drives or multiple databases) spanning multiple computing devices (e.g., multiple server computers). In some embodiments, data store 140 stores one or more of property data 142, performance data 152, and / or prediction data 160.

[0043] The property data 142 includes historical property data 144 and current property data 146. In some embodiments, the property data 142 may include one or more of the following: image data, SEM images, EDX images, morphology data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), and so on. In some embodiments, the sensor data may include temperature data, temperature range, power data, comparison parameters for comparing inspection data with threshold data, threshold data, cooling rate data, cooling rate range, and so on. In some embodiments, at least a portion of the property data 142 is from the sensor 126 and / or metrology equipment 128.

[0044] The performance data 152 includes historical performance data 154 and current performance data 156. The performance data 152 may indicate whether the substrate is properly designed, properly manufactured, and / or properly functioning. In some embodiments, at least a portion of the performance data 152 is associated with the quality of the substrate produced by the manufacturing equipment 124. In some embodiments, at least a portion of the performance data 152 is based on metrology data from the metrology equipment 128 (e.g., the historical performance data 154 includes metrology data indicating a properly processed substrate, property data of the substrate, yield, etc.). In some embodiments, at least a portion of the performance data 152 is based on an inspection of the substrate (e.g., the current performance data 156 based on an actual inspection). In some embodiments, the performance data 152 includes user input indicating the quality of the substrate (e.g., via the client device 120). In some embodiments, the performance data 152 includes an indication of an absolute value (e.g., the inspection data of the bonding interface indicates a calculated value difference from the threshold data, the deformation value differs from the threshold deformation value by a calculated value) or a relative value (e.g., the inspection data of the bonding interface indicates a 5% difference from the threshold data, the deformation differs from the threshold deformation by 5%). In some embodiments, the performance data 152 indicates meeting a threshold error amount (e.g., at least 5% production error, at least 5% flow error, at least 5% deformation error, specification limits).

[0045] In some embodiments, the client device 120 provides performance data 152 (such as product data). In some examples, the client device 120 provides performance data 152 (e.g., based on user input) indicating a product anomaly (e.g., a defective product). In some embodiments, the performance data 152 includes the quantity of normal or abnormal products produced (e.g., 98% normal products). In some embodiments, the performance data 152 indicates the predicted quantity of produced products as normal or abnormal. In some embodiments, the performance data 152 includes one or more of the following: the yield rate of the previous batch of products, the average yield rate, the predicted yield rate, the predicted quantity of defective or non-defective products, and so on. In some examples, in response to the yield rate of the first batch of products being 98% (e.g., 98% of the products are normal and 2% are abnormal), the client device 120 provides performance data 152 indicating that the yield rate of the next batch of products will reach 98%.

[0046] In some embodiments, the historical data includes one or more of historical property data 144 and / or historical performance data 154 (e.g., at least a portion for training the machine learning model 190). The current data includes one or more of current property data 146 and / or current performance data 156 (e.g., at least a portion to be input into the trained machine learning model 190 after training the model 190 using the historical data). In some embodiments, the current data is used to retrain the trained machine learning model 190.

[0047] In some embodiments, the prediction data 160 is to be used to cause a corrective action to be performed on the manufacturing equipment, the substrate processing system, or the parts of the substrate processing equipment.

[0048] Performing multiple types of metrology on multiple product layers to determine whether to perform a corrective action is costly in terms of the time used, the metrology equipment 128 used, the energy consumed, the bandwidth for transmitting the metrology data, the processor overhead for processing the metrology data, and so on. By providing the property data 142 to the model 190 and receiving the prediction data 160 from the model 190, the system 100 has the technical advantage of avoiding the use of multiple types of metrology equipment 128 on multiple product layers and the expensive process of discarding substrates.

[0049] Performing a manufacturing process using the manufacturing equipment 124 and / or manufacturing parameters that cause product defects is costly in terms of time, energy, products, the manufacturing equipment 124, and the cost of identifying corrective actions to avoid product defects. By providing the property data 142 to the model 190, receiving the prediction data 160 from the model 190, and causing a corrective action based on the prediction data 160, the system 100 has the technical advantage of avoiding the costs of producing, identifying, and discarding defective substrates.

[0050] In some embodiments, the prediction system 110 further includes server machines 170 and 180. Server machine 170 includes a dataset generator 172 that is capable of generating a dataset (e.g., a set of data inputs and a set of target outputs) to train, validate, and / or test the machine learning model 190. The dataset generator 172 has data collection, compilation, reduction, and / or partitioning functions to convert the data into a form suitable for machine learning. In some embodiments (e.g., for small datasets), no partitioning (e.g., explicit partitioning) for post-training validation is used. During training, repeated cross-validation (such as five-fold cross-validation, leave-one-out-cross-validation) can be used, where during training, a given dataset is actually repeatedly partitioned into different training sets and validation sets. A model (e.g., the best model, the model with the highest accuracy, etc.) is selected from the model vectors on the automatically separated combinatoric subsets. In some embodiments, the dataset generator 172 can explicitly partition historical data (e.g., historical property data 144 and corresponding historical performance data 154) into a training set (e.g., sixty percent of the historical data), a validation set (e.g., twenty percent of the historical data), and a test set (e.g., twenty percent of the historical data). In this embodiment, some operations of the dataset generator 172 are described in detail below with respect to Figure 2 In some embodiments, the prediction system 110 (e.g., via the prediction component 114) generates multiple sets of features (e.g., training features). In some examples, the first set of features corresponds to a first set of types of property data (e.g., from a first set of sensors, a first combination of values from a first set of sensors, a first pattern of values from a first set of sensors), the first set of types of property data corresponding to each of the datasets (e.g., training set, validation set, and test set), and the second set of features corresponds to a second set of types of property data (e.g., from a second set of sensors different from the first set of sensors, a second combination of values different from the first combination, a second pattern different from the first pattern), the second set of types of property data corresponding to each of the datasets.

[0051] Server machine 180 includes a training engine 182, a validation engine 184, a selection engine 185, and / or a testing engine 186. In some embodiments, an engine (e.g., training engine 182, validation engine 184, selection engine 185, and testing engine 186) refers to hardware (e.g., circuitry, dedicated logic, programmable logic, microcode, processing device, etc.), software (e.g., instructions running on a processing device, a general computer system, or a dedicated machine), firmware, microcode, or a combination of the foregoing items. The training engine 182 is capable of training a machine learning model 190 using one or more sets of features associated with a training set from a dataset generator 172. In some embodiments, the training engine 182 generates multiple trained machine learning models 190, where each trained machine learning model 190 corresponds to a distinct set of parameters (e.g., property data 142) and a corresponding response (e.g., performance data 152) in the training set. In some embodiments, for the purpose of modeling multiple effects, multiple models are trained on the same parameters with different objectives. In some examples, a first trained machine learning model is trained using property data 142 from all sensors 126 (e.g., sensors 1-5), a second trained machine learning model is trained using a first subset of the property data (e.g., from sensors 1, 2, and 4), and a third trained machine learning model is trained using a second subset of the property data that partially overlaps with the first subset of features (e.g., from sensors 1, 3, 4, and 5).

[0052] The validation engine 184 is capable of validating a trained machine learning model 190 using a corresponding set of features from a validation set of the dataset generator 172. For example, a first trained machine learning model 190 trained using a first set of features of the training set is validated using a first set of features of the validation set. The validation engine 184 determines the accuracy of each trained machine learning model 190 based on the corresponding sets of features in the validation set. The validation engine 184 evaluates and flags (e.g., flags for discard) trained machine learning models 190 whose accuracy does not meet a threshold accuracy. In some embodiments, the selection engine 185 is capable of selecting one or more trained machine learning models 190 whose accuracy meets the threshold accuracy. In some embodiments, the selection engine 185 is capable of selecting the trained machine learning model 190 having the highest accuracy among the trained machine learning models 190.

[0053] The test engine 186 can test the trained machine learning model 190 using a corresponding set of features in the test set from the dataset generator 172. For example, a first trained machine learning model 190 trained using a first set of features of the training set is tested using a first set of features of the test set. The test engine 186 determines the trained machine learning model 190 with the highest accuracy among all the trained machine learning models based on the test set.

[0054] In some embodiments, the machine learning model 190 (e.g., for classification) refers to a model artifact created by the training engine 182 using a training set that includes data inputs and corresponding target outputs (e.g., correctly classifying the condition or ordinal level of the corresponding training inputs). Patterns in the dataset that map the data inputs to the target outputs (correct classifications or levels) can be found, and mappings capturing these patterns are provided to the machine learning model 190. In some embodiments, the machine learning model 190 uses one or more of the following: Gaussian process regression (GPR), Gaussian process classification (GPC), Bayesian neural network, neural network Gaussian process, deep belief network, Gaussian mixture model, or other probabilistic learning methods. Non-probabilistic methods can also be used, including one or more of the following: support vector machine (SVM), radial basis function (RBF), clustering, k-nearest neighbor algorithm (k-NN), linear regression, random forest, neural network (such as artificial neural network), etc. In some embodiments, the machine learning model 190 is a multivariate analysis (MVA) regression model.

[0055] The prediction component 114 provides the current property data 146 (e.g., as an input) to the trained machine learning model 190 and runs the trained machine learning model 190 (e.g., runs on the input to obtain one or more outputs). The prediction component 114 can determine (e.g., extract) the prediction data 160 from the trained machine learning model 190 and determine (e.g., extract) the uncertainty data indicating the confidence level corresponding to the prediction data 160 and the current performance data 156. In some embodiments, the prediction component 114 or the corrective action component 122 uses the uncertainty data (e.g., an uncertainty function or an acquisition function derived from the uncertainty function) to decide whether to perform a corrective action using the prediction data 160 or whether to further train the model 190.

[0056] For purposes of illustration and not limitation, aspects of the present disclosure describe training one or more machine learning models 190 using historical data (i.e., previous data, historical property data 144, and historical performance data 154), and providing current property data 146 into one or more trained probabilistic machine learning models 190 to determine prediction data 160. In other implementations, a heuristic model or rule-based model is used to determine prediction data 160 (e.g., without using a trained machine learning model). In other implementations, a non-probabilistic machine learning model may be used. The prediction component 114 monitors historical property data 144 and historical performance data 154. In some embodiments, any information described regarding Figure 2 the data input 210 can be monitored or otherwise used in a heuristic model or rule-based model.

[0057] In some embodiments, the functionality of the client device 120, prediction server 112, server machines 170, and server machines 180 is to be provided by a smaller number of machines. For example, in some embodiments, server machines 170 and 180 are integrated into a single machine, and in some other embodiments, server machines 170, server machine 180, and prediction server 112 are integrated into a single machine. In some embodiments, client device 120 and prediction server 112 are integrated into a single machine.

[0058] Generally, functions described in one embodiment as being performed by client device 120, prediction server 112, server machines 170, and server machines 180 may also be performed on prediction server 112 in other embodiments as appropriate. Additionally, functionality attributed to a particular component may also be performed by different components or multiple components operating together. For example, in some embodiments, prediction server 112 determines a corrective action based on prediction data 160. In another example, client device 120 determines prediction data 160 based on data received from a trained machine learning model.

[0059] Furthermore, the functionality of a specific component may also be performed by different components or multiple components operating together. In some embodiments, one or more of prediction server 112, server machine 170, or server machine 180 are accessed as a service provided to other systems or devices through an appropriate application programming interface (API).

[0060] In some embodiments, a "user" is represented as an individual person. However, other embodiments of the present disclosure include a "user" that is an entity controlled by multiple users and / or automated sources. In some examples, a group of individual users acting jointly as an administrator group are considered a "user".

[0061] Although embodiments of the present disclosure are discussed in terms of determining predictive data 160 for defect source tracking and defect root cause identification in substrate processing in a manufacturing facility (e.g., a substrate processing facility), in some embodiments, the present disclosure can also be generally applied to quality inspection. Embodiments can be generally applied to determining part quality based on different types of data.

[0062] Figure 2 Illustrated is a dataset generator 272 (e.g., Figure 1 dataset generator 172) for creating a dataset for a machine learning model (e.g., Figure 1 model 190) according to certain embodiments. In some embodiments, the dataset generator 272 is part of a server machine 170 of Figure 1 . The dataset generated by Figure 2 dataset generator 272 can be used to train a machine learning model (e.g., see Figure 5B ) to cause the execution of a corrective action (e.g., see Figure 5C ).

[0063] The dataset generator 272 (e.g., Figure 1 dataset generator 172) creates a dataset for a machine learning model (e.g., Figure 1 model 190). The dataset generator 272 creates the dataset using historical property data 244 (e.g., Figure 1 historical property data 144) and historical performance data 254 (e.g., Figure 1 historical performance data 154). Figure 2 System 200 of

[0064] illustrates the dataset generator 272, data input 210, and target output 220 (e.g., target data).

[0065] In some embodiments, the dataset generator 272 generates the data input 210 and the target output 220. In some embodiments, the data input 210 includes one or more sets of historical property data 244. In some embodiments, the historical property data 244 includes one or more of the following: property data from one or more types of sensors and / or metrology equipment, combinations of property data from one or more types of sensors and / or metrology equipment, patterns of property data from one or more types of sensors and / or metrology equipment, and the like.

[0066] In some embodiments, the dataset generator 272 generates a first data input corresponding to the first set of historical property data 244A to train, validate, or test the first machine learning model, and the dataset generator 272 generates a second data input corresponding to the second set of historical property data 244B to train, validate, or test the second machine learning model.

[0067] In some embodiments, the dataset generator 272 discretizes (e.g., segments) one or more of the data input 210 or the target output 220 (e.g., for a classification algorithm for a regression problem). The discretization of the data input 210 or the target output 220 (e.g., segmenting via a sliding window) converts continuous variable values into discrete values. In some embodiments, the discrete values of the data input 210 indicate discrete historical property data 244 used to obtain the target output 220 (e.g., discrete historical performance data 254).

[0068] The data input 210 and the target output 220 used to train, validate, or test the machine learning model include information about a specific facility (e.g., a specific substrate manufacturing facility). In some examples, the historical property data 244 and the historical performance data 254 are for the same manufacturing facility.

[0069] In some embodiments, the information used to train the machine learning model comes from a specific type of manufacturing equipment 124 in the manufacturing facility that has specific characteristics, and allows the trained machine learning model to determine results for a specific group of manufacturing equipment 124 based on the input of current parameters (e.g., current property data 146), where the current parameters are associated with one or more components sharing the characteristics of the specific group. In some embodiments, the information used to train the machine learning model is for components from two or more manufacturing facilities, and allows the trained machine learning model to determine results for the components based on the input from one manufacturing facility.

[0070] In some embodiments, after generating the dataset and using the dataset to train, validate, or test the machine learning model 190, the machine learning model 190 is further trained, validated, or tested (e.g., Figure 1current performance data 156) or adjustment (e.g., adjusting weights associated with input data of machine learning model 190, such as connection weights in a neural network).

[0071] Figure 3 is a block diagram of a system 300 for generating prediction data 360 (e.g., Figure 1 prediction data 160) according to certain embodiments. System 300 is used to determine prediction data 360 via a trained machine learning model (e.g., Figure 1 model 190) for defect source tracing and defect root cause identification (e.g., for performing corrective actions).

[0072] At block 310, system 300 (e.g., Figure 1 prediction system 110) performs data partitioning on historical data (e.g., historical property data 344 and / or historical performance data 354 for Figure 1 model 190) (e.g., performed by dataset generator 172 of server machine 170 of Figure 1 ) to generate a training set 302, a validation set 304, and a test set 306. In some examples, the training set is 60% of the historical data, the validation set is 20% of the historical data, and the test set is 20% of the historical data. System 300 generates multiple feature sets for each of the training set, validation set, and test set. In some examples, if the historical data includes features derived from 20 sensors (e.g., Figure 1 sensors 126, sensors of manufacturing equipment and / or metrology equipment) and 100 products (e.g., products each corresponding to property data from the 20 sensors), then the first set of features is sensors 1 - 10, the second set of features is sensors 11 - 20, the training set is products 1 - 60, the validation set is products 61 - 80, and the test set is products 81 - 100. In this example, the first set of features of the training set would be the parameters from sensors 1 - 10 for products 1 - 60.

[0073] At block 312, system 300 uses training set 302 to perform model training (e.g., via Figure 1The training engine 182 executes). In some embodiments, the system 300 uses multiple sets of features of the training set 302 (e.g., the first set of features of the training set 302, the second set of features of the training set 302, etc.) to train multiple models. For example, the system 300 trains a machine learning model to generate a first trained machine learning model using the first set of features in the training set (e.g., property data from sensors 1-10 for products 1-60), and generates a second trained machine learning model using the second set of features in the training set (e.g., property data from sensors 11-20 for products 1-60). In some embodiments, the first trained machine learning model and the second trained machine learning model are combined to generate a third trained machine learning model (in some embodiments, it is, for example, a better predictor than the first trained machine learning model or the second trained machine learning model itself). In some embodiments, the sets of features used to compare the models overlap (e.g., the first set of features is property data from sensors 1-15, and the second set of features is property data from sensors 5-20). In some embodiments, hundreds of models are generated, including models with various permutations of features and model combinations.

[0074] At block 314, the system 300 performs model validation using the validation set 304 (e.g., via Figure 1 the validation engine 184 executes). The system 300 validates each of the trained models using a corresponding set of features in the validation set 304. For example, the system 300 uses the first set of features in the validation set (e.g., parameters from sensors 1-10 for products 61-80) to validate the first trained machine learning model, and uses the second set of features in the validation set (e.g., parameters from sensors 11-20 for products 61-80) to validate the second trained machine learning model. In some embodiments, the system 300 validates hundreds of models generated at block 312 (e.g., models with various permutations of features, model combinations, etc.). At block 314, the system 300 determines the accuracy of each of the one or more trained models (e.g., determined via model validation), and determines whether one or more of the trained models have an accuracy that meets a threshold accuracy. In response to determining that none of the trained models have an accuracy that meets the threshold accuracy, the process returns to block 312, where the system 300 performs model training using different sets of features of the training set. In response to determining that one or more of the trained models have an accuracy that meets the threshold accuracy, the process continues to block 316. The system 300 discards the trained machine learning models that have an accuracy less than the threshold accuracy (e.g., based on the validation set).

[0075] At block 316, the system 300 performs model selection (e.g., viaFigure 1 The selection engine 185 (executed by determines which one of the one or more trained models that meet the threshold accuracy has the highest accuracy (e.g., the selected model 308 is selected based on the verification of block 314). In response to determining that two or more of the trained models that meet the threshold accuracy have the same accuracy, the process returns to block 312, where the system 300 performs model training using a further refined training set corresponding to further refined groups of features to determine the trained model with the highest accuracy.

[0076] At block 318, the system 300 performs model testing (e.g., via Figure 1 the test engine 186) using the test set 306 to test the selected model 308. The system 300 uses the first set of features in the test set (e.g., the property data from sensors 1-10 for products 81-100) to test the first trained machine learning model to determine that the first trained machine learning model meets the threshold accuracy (e.g., determined based on the first set of features of the test set 306). In response to the accuracy of the selected model 308 not meeting the threshold accuracy (e.g., the selected model 308 overfits the training set 302 and / or the validation set 304 and is not applicable to other data sets such as the test set 306), the process continues to block 312, where the system 300 performs model training (e.g., retrains) using a different training set corresponding to different groups of features (e.g., property data from different sensors). In response to determining that the selected model 308 has an accuracy that meets the threshold accuracy based on the test set 306, the process continues to block 320. At least in block 312, the model learns patterns in the model learning history data to make predictions, and in block 318, the system 300 applies the model to the remaining data (e.g., the test set 306) to test the predictions.

[0077] At block 320, the system 300 receives the current property data 346 (e.g., Figure 1 the current property data 146) using the trained model (e.g., the selected model 308), and determines (e.g., extracts) the prediction data 360 (e.g., Figure 1 the prediction data 160) from the trained model for defect source tracing and defect root cause identification, thereby performing corrective actions. In some embodiments, the current property data 346 corresponds to the same type of features in the historical property data 344. In some embodiments, the current property data 346 corresponds to a subset of the feature types in the historical property data 344 that are the same as the feature types used to train the selected model 308.

[0078] In some embodiments, current data is received. In some embodiments, the current data includes current performance data 356 (e.g., Figure 1 current performance data 156) and / or current property data 346. In some embodiments, at least a portion of the current data is received from metrology equipment (e.g., Figure 1 metrology equipment 128) or via user input. In some embodiments, the model 308 is retrained based on the current data. In some embodiments, the new model is trained based on the current performance data 356 and the current property data 346.

[0079] In some embodiments, one or more of blocks 310 to 320 occur in various orders and / or occur in conjunction with other operations not presented and described herein. In some embodiments, one or more of blocks 310 to 320 are not performed. For example, in some embodiments, one or more of the data partitioning of block 310, the model verification of block 314, the model selection of block 316, and / or the model testing of block 318 are not performed.

[0080] Figure 4A is a block diagram showing defect categories 410A to 410D according to certain embodiments. The property data may include different types of property data 400A to 400Z (e.g., different subsets of property data) from one or more types of sensors and / or one or more types of metrology equipment. In some embodiments, as Figure 4A shown, the processing device may identify regions of the substrate corresponding to the defect categories 410A to 410D based on the property data 400A (e.g., the first subset of the property data 400).

[0081] In some embodiments, the property data 400A includes SEM images captured by SEM metrology equipment, EDX images captured by EDX metrology equipment, etc., morphology data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), etc. The property data 400A may be associated with different regions of the substrate (e.g., captured at different regions of the substrate). The processing device may determine, based on the property data 400A, that a particular region of the substrate includes defects of a particular defect category 410A to 410D.

[0082] The defect categories 410 may include one or more of the following: broken wire, unfilled, bridge, stain, scratch, glass damage, foreign object (particle), residue, resist collapse, z-axis defect, large area, via stress, cavity, pit, crystal defect, crack, etc.

[0083] Figure 4B is a block diagram showing defect subcategories according to some embodiments. In some embodiments, as Figure 4B shown, the processing device, based on property data 400B to 400D (e.g., a second subset of property data), subdivides the regions of the substrate corresponding to defect category 410A into defect subcategories 410A1 to 410A4, subdivides the regions of the substrate corresponding to defect category 410B into defect subcategories 410B1 to 410B3, and subdivides the regions of the substrate corresponding to defect category 410C into defect subcategories 410C1 to 410C4.

[0084] Defect subcategories may include spherical particles, random particles, rod-shaped particles, flake particles, post imprint fall-on particles, annular pits, micro pits, macro pits, mouse bites, multi-line bridging, CMP bridging, photoresist bridging, micro bridging, organic stains, inorganic stains, lithography pinches, step bunching, stack tomography, shallow triangles, obtuse triangles, surface triangles, down falls, punctures, skip marks, crescents, microtubes, PL circles, basal plane dislocations, protrusions, mounds, and so on.

[0085] In some embodiments, property data 400B to 400C may include one or more of the following: SEM images, EDX images, morphology data, size attribute data, dimension attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), and so on. In some embodiments, property data 400A includes SEM images, property data 400B includes defect maps, property data 400C includes EDX data (e.g., EDX images generated via X-ray fluorescence (XRF), inductively coupled plasma mass spectrometry (ICP-MS), triple quadrupole ICP-MS (ICP-QQQ), etc.), and property data 400D includes chip layout.

[0086] Figure 4C is a block diagram associated with defect subcategories according to some embodiments. In some embodiments, as Figure 4C shown, the processing device identifies a subset of the defect subcategories corresponding to the execution of a correction action (e.g., defect subcategories of interest). In some embodiments, the subset of defect subcategories is determined based on user input 444 indicating one or more subcategories of interest. In some embodiments, the defect subcategories are determined based on machine learning. In some embodiments, the defect subcategories are determined based on a threshold.

[0087] In some embodiments, the processing device may compile defect data 420 and location data 430 to track each defect from layer to layer or from operation to operation in the substrate manufacturing process. For example, in some embodiments, the processing device may identify defects (e.g., nature data) and, based on defect evolution information, identify the previous process operation where the defect source or the root cause of the defect lies. The location data may include wafer map coordinates and identifications (e.g., layer number, or process operation number) to indicate the location and layer / operation of such data on the substrate. For example, wafer mapping (such as a KLA result file (KLARF)) may be used to store the location data. In some embodiments, the user identifies a subset of defect subcategories (e.g., 410A1, 410A2, 410A4, 410B1, 410B2, 410C1, 410C2, 410C4) through user input 444. Further, the processing device may add the location data corresponding to each identified defect subcategory to the location data 430 of the next layer or subsequent layers. The processing logic identifies the area corresponding to the identified defect subcategory and identifies further nature data of the area, including location data.

[0088] In some embodiments, one of the defect subcategories may be selected, and the defect mapping coordinates of the corresponding area of the defect subcategory may be appended to the corresponding location data (e.g., defect map) of one or more subsequent layers to generate location data (e.g., defect map) for the subsequent layers (such as a KLA result file (KLARF), etc.). In some embodiments, a part of the defect mapping coordinates of the corresponding area of the defect subcategory may be selected, and the selected part of the defect mapping coordinates of the corresponding area of the defect subcategory may be appended to the defect map of the subsequent layer.

[0089] In some embodiments, the location data of the subsequent layer of the substrate may be identified. The location data may correspond to the defect mapping coordinates of the previously selected defect subcategory of the corresponding area. In some embodiments, a defect evolution database entry may be created that includes nature data from each layer, the nature data including one or more defect images, location data, and defect evolution information of the defect including one or more defect evolution images. In some embodiments, the defect evolution database entry may be stored for subsequent access. In some embodiments, the nature data of the layer of the substrate may be identified, and based on the defect evolution database entry, the nature data may be matched with the database entry of the defect evolution information. Based on the defect evolution information, the defect may be traced back to the defect source, or the root cause of the defect may be identified, and based on the defect source or the root cause of the defect, the execution of a corrective action associated with the substrate processing system may be triggered. In some embodiments, the defect source or the root cause of the defect may be due to a defect or deficiency in a previous process corresponding to a previous layer of the substrate manufacturing process.

[0090] Defects that cause substrate failure can be referred to as killing the substrate. The kill ratio can be the percentage of substrates with a particular type of defect that experience failure. Defects that do not affect substrate functionality have a low kill ratio. For example, defects caused by a first substrate processing operation and etched off the substrate during a second substrate processing operation may have a low kill ratio. Defects that generally impede the normal functionality of the substrate have a high kill ratio. For example, defects caused by a substrate processing operation and not removed by subsequent operations may have a high kill ratio. The tolerance for defects with a low kill ratio can be higher compared to defects with a high kill ratio. In some embodiments, a defect subcategory can include defects that do not cause the performance data 152 of the manufactured substrate to fall below a certain threshold (e.g., low kill ratio, etc.).

[0091] Figure 4D is a block diagram showing substrate defects (e.g., adder defects and common defects) according to certain embodiments. In some embodiments, as Figure 4D shown, the processing device identifies the property data 400A of the substrate 401 (e.g., the first subset of the property data 400). In some embodiments, the substrate 401 has undergone operation 1 of the manufacturing process. In some embodiments, the substrate 402 can be the substrate 401 after undergoing operation 2 of the manufacturing process. In some embodiments, the substrate 403 can be the substrate 402 after undergoing operation n of the manufacturing process. In some embodiments, the property data 400A includes the location data 400B1 of the regions of the substrate corresponding to the defect categories 450A to 450C. Such location data 400B1 can be identified as an adder defect (e.g., a new defect or a defect not left over from a previous layer or operation in the substrate manufacturing process). In some embodiments, as Figure 4D shown, the processing device can identify the regions of the substrate corresponding to the defect categories 450A to 450C based on the property data 400A (e.g., the first subset of the property data 400). In some embodiments, the processing logic maps the regions of the substrate corresponding to the defect categories to the location data 400B1 (e.g., defect mapping, KLARF, etc.).

[0092] In some embodiments, subsequent operations (e.g., operation 2) are performed on the substrate 401. In some embodiments, the processing device identifies the property data 400E of the substrate 402 (e.g., the second subset of the property data 400) (e.g., the substrate 401 after undergoing operation 2 of the substrate manufacturing process). In some embodiments, the property data 400E includes the location data 400B2 of the regions of the substrate corresponding to the defect categories 460A to 460C. In some embodiments, as Figure 4DAs shown, the processing device can identify regions of the substrate corresponding to defect classes 460A to 460C based on property data 400E (e.g., the second subset of property data 400). In some embodiments, location data 400B2 can be identified as an additive defect (e.g., a new defect or a defect not left over from a previous layer or operation in the substrate manufacturing process). In some embodiments, location data 400B1 can be a common defect (e.g., an old defect or a defect left over from a previous layer or operation in the substrate manufacturing process). In some embodiments, the processing logic does not map regions of the substrate corresponding to defect classes that are common defects. In some embodiments, the processing device maps regions of the substrate corresponding to defect classes to location data 400B2 (e.g., defect mapping, KLARF, etc.).

[0093] In some embodiments, subsequent operations (e.g., operation n) are performed on substrate 402. In some embodiments, the processing device identifies property data 400F (e.g., the third subset of property data 400) of substrate 403 (e.g., substrate 402 after undergoing operation n of the substrate manufacturing process). In some embodiments, property data 400F includes location data 400B3 of regions of the substrate corresponding to defect classes 470A to 470C. In some embodiments, as Figure 4D shown, the processing device can identify regions of the substrate corresponding to defect classes 470A to 470C based on property data 400F (e.g., the third subset of property data 400). In some embodiments, location data 400B3 can be identified as an additive defect (e.g., a new defect or a defect not left over from a previous layer or operation in the substrate manufacturing process). In some embodiments, location data 400B1 to 400B2 can be common defects (e.g., old defects or defects left over from a previous layer or operation in the substrate manufacturing process). In some embodiments, the processing logic does not map regions of the substrate corresponding to defect classes that are common defects. In some embodiments, the processing logic maps regions of the substrate corresponding to defect classes to location data 400B3 (e.g., defect mapping, KLARF, etc.).

[0094] In some embodiments, as Figure 4CAs described above, a subset of defect sub - categories (e.g., 450A of operation 1) can be identified (e.g., via user input 444, via the processing device). Further, the processing device can add the position data corresponding to each identified defect sub - category (e.g., 450A) corresponding to the substrate region to the position data 430 of the next layer (e.g., operation 2), even if such position data may be common defects. In some embodiments, adding the position data corresponding to the identified defect sub - categories corresponding to the substrate region to the position data of the next layer allows for the observation of defect evolution from operation to operation and from layer to layer.

[0095] In some embodiments, property data corresponding to the selected defect sub - categories can be identified (e.g., via user input 444, via the processing device) (e.g., defect images, mapped coordinates of the corresponding regions of the defect sub - categories, etc.). In some embodiments, the processing device can create a database entry (defect database entry, defect evolution database entry, etc.) including property data from each layer (e.g., one or more defect images and defect evolution information of the defect including one or more defect evolution images). In some embodiments, the defect evolution image can include multiple images of defects from multiple layers of the substrate, where these layers correspond to different operations in the manufacturing process. The processing device can store the database entry in a database (e.g., defect database, defect evolution database, knowledge database, etc.) for subsequent access.

[0096] In some embodiments, the processing device can identify the property data of the substrate processed by the substrate processing system. In some embodiments, the processing device can identify the region of the substrate corresponding to the first defect category based on a first subset of the property data. In some embodiments, the processing device can subdivide the region of the substrate corresponding to the first defect category into defect sub - categories based on a second subset of the property data. In some embodiments, the processing device can cause the execution of a correction action associated with the substrate processing system based on one or more regions corresponding to at least one of the defect sub - categories. In some embodiments, the processing device can determine the defect source (e.g., defect source tracing) based on at least one of the defect sub - categories, and the correction action corresponds to the defect source. In some embodiments, the processing device can determine the defect root cause (e.g., defect root cause identification) based on at least one of the defect sub - categories, and the correction action corresponds to the defect root cause. In some embodiments, the defect source or the defect root cause can correspond to a previous layer of the substrate and / or a previous process operation in the substrate manufacturing process. In some embodiments, the processing device can determine the defect source and / or the defect root cause based on the defect evolution information. In some embodiments, the defect evolution information can be stored in the defect evolution database.

[0097] In some embodiments, the processing device may search a database (e.g., a defect evolution database) for database entries having historical property data that is substantially similar (e.g., most similar) to the current property data (e.g., defect image), and match the current property data corresponding to the defect with a substantially similar defect evolution entry (e.g., the most similar defect evolution database entry). In some embodiments, the processing device may associate the search results (e.g., similar defect evolution entries on the substrate, the most similar database entries) with, for example, a defect source, a defect root cause, and / or a corrective action within a particular confidence metric. In some embodiments, the processing device may identify the defect source, the defect root cause, and / or the corrective action. In some embodiments, the defect source and / or the defect root cause may correspond to a previous operation in the substrate manufacturing process. In some embodiments, the defect source, the defect root cause, and / or the corrective action may correspond to manufacturing equipment previously used in the substrate manufacturing process.

[0098] In some embodiments, previous operations in the substrate manufacturing process may be wet cleaning, surface passivation, lithography, ion implantation, etching, dry etching, reactive ion etching (RIE), deep reactive ion etching, atomic layer etching (ALE), wet etching, buffered oxide etching, plasma ashing, heat treatment, rapid thermal annealing, furnace annealing, thermal oxidation, chemical vapor deposition (CVD), atomic layer deposition (ALD), physical vapor deposition (PVD), molecular beam epitaxy (MBE), laser lift-off, electrochemical deposition (ECD), chemical mechanical polishing (CMP), wafer testing, die preparation, via manufacturing, wafer mounting, wafer back grinding and polishing, wafer bonding and stacking, redistribution layer manufacturing, wafer bumping, die singulation or wafer dicing, IC packaging, die attachment, molding, baking, electroplating, laser marking or screen printing, IC testing, and so on.

[0099] In some embodiments, during a defect out-of-control (OOC) event in a substrate manufacturing system (e.g., a substrate manufacturing facility), the processing device may identify property data corresponding to the OOC defect. The processing device may search a database (e.g., a knowledge database) for the database entry that is most similar to the property data (e.g., a defect image or a defect evolution image), and match the property data corresponding to the OOC defect with the most similar database entry. The processing device may associate the search results (e.g., the most similar database entry) with, for example, a defect source, a defect root cause, and / or a corrective action within a particular confidence metric. In some embodiments, the processing device or the user may review the recommendations and develop an action plan.

[0100] Figures 5A to 5CFIG. 0 is a flowchart of methods 500A through 500C associated with defect analysis. In some embodiments, methods 500A through 500C are performed by processing logic that includes hardware (e.g., circuitry, dedicated logic, programmable logic, microcode, processing device, etc.), software (e.g., instructions running on a processing device, general-purpose circuitry, or a dedicated machine), firmware, microcode, or a combination of the foregoing. In some embodiments, methods 500A through 500C are performed at least in part by prediction system 110. In some embodiments, method 500A is performed by client device 120 (e.g., calibration action component 122) and / or prediction system 110 (e.g., prediction component). In some embodiments, method 500B is performed by server machine 180 (e.g., training engine 182, etc.). In some embodiments, method 500C is performed by prediction server 112 (e.g., prediction component 114) and / or client device 120 (e.g., calibration action component 122). In some embodiments, a non-transitory storage medium stores instructions that, when executed by a processing device (e.g., the processing device of prediction system 110, the processing device of server machine 180, the processing device of prediction server 112, the processing device of client device 120, etc.), cause the processing device to perform one or more of methods 500A through 500C.

[0101] For ease of explanation, methods 500A through 500C are depicted and described as a series of operations. However, the operations in accordance with the present disclosure may occur in various orders and / or in parallel, and may occur in conjunction with other operations not presented and described herein. Additionally, in some embodiments, not all of the operations shown are performed to implement methods 500A through 500C in accordance with the disclosed subject matter. Further, those skilled in the art will appreciate and understand that methods 500A through 500C may alternatively be represented as a series of interrelated states via a state diagram or events.

[0102] Figure 5A FIG. 7 is a flowchart of method 500A for defect analysis (e.g., defect source tracing and defect root cause identification) according to certain embodiments.

[0103] Referring Figure 5A , in some embodiments, at block 502, the processing logic implementing method 500A identifies property data (e.g., Figure 1 property data 142) of a substrate processed by a substrate processing system (e.g., Figure 1 fabrication equipment 124). The substrate may be a wafer, semiconductor, display, etc.

[0104] In some embodiments, the property data can be at least one of, for example, SEM images, EDX images, etc. In some embodiments, the property data can be at least one of the following: morphological data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), etc.

[0105] After one or more substrate processing operations, the property data can be captured via metrology equipment.

[0106] After one or more substrate processing operations, the property data can be captured via a sensor.

[0107] At block 504, the processing logic identifies a region of the substrate corresponding to a first defect class based on a first subset of the property data. In some embodiments, the first subset of the property data 142 can be at least one of the following: SEM images, EDX images, morphological data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), etc. In some embodiments, the first subset of the property data 142 can be at least one of SEM images, EDX images, etc. The defect class can include one or more of the following: broken line, unfilled, bridge, stain, scratch, glass damage, foreign object (particle), residue, resist collapse, via stress, cavity, pit, crystal defect, crack, etc.

[0108] At block 506, the processing logic subdivides the regions of the substrate corresponding to the first defect category into defect subcategories based on a second subset of the property data. The defect subcategories can include one or more of the following: globular particles, random particles, rod-shaped particles, flake particles, post imprint fall-on particles, annular pits, micro-pits, macro-pits, mouse bites, multi-line bridging, CMP bridging, photoresist bridging, micro-bridging, organic stains, inorganic stains, lithography pinches, step bunching, stack faults, shallow triangles, obtuse triangles, surface triangles, down falls, punctures, skip marks, crescents, micro-tubes, PL circles, basal plane dislocations, protrusions, mounds, and / or similar defects.

[0109] In some embodiments, the processing logic can perform one or more of blocks 502 to 508 using a machine learning model (e.g., see Figures 5B to 5C ).

[0110] At block 508, the processing logic causes the execution of a corrective action associated with the substrate processing system based on at least one of the defect subcategories. In some embodiments, block 508 includes determining a defect root cause based on at least one of the plurality of defect subcategories, and the corrective action corresponds to the defect root cause. For example, the corrective action corresponding to the defect root cause can be a corrective action that corrects aspects of a subset of the manufacturing equipment that caused the defect. Tracking the root cause can be determining the defect caused by a subset of the manufacturing equipment based on the property data of the substrate. In some embodiments, the defect root cause can be a specific part of the substrate processing equipment (e.g., Figure 1 manufacturing equipment 124).

[0111] In some embodiments, block 508 includes determining a defect source based on at least one of the plurality of defect subcategories (e.g., the corrective action corresponds to the defect source). For example, when a corrective action prevents further defects from the defect source, the corrective action corresponds to the defect source. For example, the corrective action can be replacing a defective component of the manufacturing equipment. In some embodiments, defect source tracking can be used to identify the defect source. Defect source tracking can determine the origin of the defect based on the defect property data. In some embodiments, the defect source can be a component of the substrate processing equipment (e.g., Figure 1 manufacturing equipment 124) to be cleaned, repaired, and / or replaced.

[0112] In some embodiments, the corrective action includes providing an alert, causing a cleaning action, causing a repair action, causing a replacement component, shutting down one or more parts of the substrate processing equipment, determining a predicted end-of-life of a component of the substrate processing equipment, and so on. In some embodiments,Figure 1 The corrective action component 122 receives an indication of a corrective action (e.g., based on the prediction data 160) from the prediction system 110 and causes the corrective action to be executed.

[0113] In some embodiments, the processing logic further determines that at least one of the defect subcategories corresponds to a defect evolution associated with the execution of a corrective action. In some embodiments, the defect evolution may be information about defects in multiple operations (e.g., layers) of a substrate manufacturing process. The defect evolution can track defects from layer to layer and record changes in the defects (such as property data). Defect translation uses the defect evolution information to track defects from layer to layer. In some embodiments, the determination is based on the defect evolution information of at least one of the defect subcategories. In some embodiments, the defect evolution information may be the property data 142 of the substrate (e.g., historical property data 144, current property data 146), which shows the same coordinates in various layers of the substrate (e.g., at least one layer has a defect or anomaly at the coordinates). In some embodiments, the lethality associated with at least one of the multiple defect subcategories can be determined using the defect evolution data. In some embodiments, the lethality can be an estimated proportion of defects that cause substrate failure (e.g., die failure on the substrate). For example, defects found on a certain number of dies result in a certain number of die failures. The ratio of the number of dies with defects to the number of failed dies can be the lethality of such defects. In some embodiments, the corrective action associated with the substrate processing system can be executed or not executed based on the defect evolution information and / or the associated lethality.

[0114] In some embodiments, the processing logic further identifies a first subset of the defect subcategories (e.g., including at least one defect subcategory corresponding to a corrective action in the defect subcategories). The first subset of the defect subcategories corresponds to the substrate property data that meets a threshold level (e.g., a thickness that meets a threshold thickness, a conductivity that meets a threshold conductivity, etc.). In some embodiments, the threshold level can correspond to property data (e.g., SEM image, EDX image, morphology data, size attribute data, dimension attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, or defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), etc.) or performance data (e.g., values of one or more of image data, morphology data, size attribute data, dimension attribute data, temperature, pitch, current, power, voltage, etc.).

[0115] In some embodiments, the identification of the first subset of the defect subcategories is based on one or more of the following: user input indicating one or more subcategories of interest (e.g., Figure 4Cuser input 444), or prediction data associated with the output received from the trained machine learning model, the prediction data being based on data inputs including historical defect subcategories and / or target outputs including historical property data.

[0116] In some embodiments, the processing logic further provides a first subset of property data as an input to the trained machine learning model. The processing logic obtains an output associated with the prediction data from the trained machine learning model. Identifying the region of the substrate corresponding to the first defect category is based on the prediction data.

[0117] In some embodiments, the processing logic further provides a second subset of property data as an input to the trained machine learning model. The processing logic further obtains an output associated with the prediction data from the trained machine learning model. Subclassifying the plurality of regions into defect subcategories is based on the prediction data.

[0118] In some embodiments, the defect source and / or defect root cause may be formed during a previous operation in the substrate manufacturing process. In some embodiments, the previous operation in the substrate manufacturing process may be related to a previous layer or the same layer in the substrate manufacturing process. In some embodiments, the previous operation may be one of the following: wet cleaning, surface passivation, lithography, ion implantation, etching, dry etching, reactive ion etching (RIE), deep reactive ion etching, atomic layer etching (ALE), wet etching, buffered oxide etching, plasma ashing, heat treatment, rapid thermal annealing, furnace annealing, thermal oxidation, chemical vapor deposition (CVD), atomic layer deposition (ALD), physical vapor deposition (PVD), molecular beam epitaxy (MBE), laser lift-off, electrochemical deposition (ECD), chemical mechanical polishing (CMP), wafer testing, die preparation, via manufacturing, wafer mounting, wafer back grinding and polishing, wafer bonding and stacking, redistribution layer manufacturing, wafer bumping, die singulation or wafer dicing, IC packaging, die attachment, molding, baking, electroplating, laser marking or screen printing, IC testing, and so on.

[0119] In some embodiments, the defect source or defect root cause may originate from a previous layer in the substrate manufacturing process. In some embodiments, the previous layer may correspond to a previous operation in the substrate manufacturing process. For example, in some embodiments, a defect may originate from Operation 1 of substrate 401. After Operation n of substrate 403, property data (e.g., property data 142, historical property data 144) may be identified. The defect source or defect root cause may be identified based on the property data 400F of substrate 403 and the defect evolution entries in the defect evolution database. In further embodiments, the processing device may cause a corrective action based on the defect source or defect root cause, which is identified based on the defect evolution entries in the defect evolution database.

[0120] Figure 5B is a method for training a machine learning model (e.g., Figure 1 model 190) to determine prediction data (e.g., Figure 1 prediction data 160) for defect source tracking and defect root cause identification.

[0121] Refer to Figure 5B , at block 510 of method 500B, the processing logic identifies the historical property data of the substrate (e.g., Figure 1 historical property data 144, historical input property data). The historical property data may include data from historical substrates, such as image data, SEM images, EDX images, morphological data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layers, chip layout data, edge data, or defect metadata (including but not limited to grayscale data and signal-to-noise ratio data), temperature, pitch, current, power, voltage, and so on.

[0122] In some embodiments, at block 512, the processing logic identifies the historical performance data of the substrate (e.g., Figure 1 historical performance data 154). The historical performance data may include values of one or more of data from historical substrates, such as image data, morphological data, size attribute data, dimensional attribute data, temperature, pitch, current, power, voltage, and so on. Performance data (including historical performance data) may include metrology data or user input that indicates that the performance of the substrate meets specific parameters or reaches a specific performance level (e.g., the ability to pass a probe test by measuring voltage). At least a portion of the historical property data and the historical performance data may be associated with new substrate processing equipment parts (which are used, for example, for benchmarking). At least a portion of the historical property data and the historical performance data may be associated with the manufactured substrate.

[0123] At block 514, processing logic trains a machine learning model using a data input that includes historical property data 144 and / or a target output that includes historical performance data 154 to generate a trained machine learning model.

[0124] In some embodiments, the historical property data is historical property data of a historical substrate, and / or the historical performance data corresponds to the historical substrate. In some embodiments, the historical property data includes a historical image of a historical substrate, and / or the historical performance data corresponds to the historical substrate. The historical performance data may be associated with substrate quality, such as metrology data of the substrate, substrate throughput, substrate defects, etc. The historical performance data may be associated with the quality of substrate processing equipment parts, such as test data, metrology data of the substrate, time to failure of the substrate, etc.

[0125] At block 514, the machine learning model may be trained using a data input that includes historical property data and / or a target output that includes historical performance data to generate a trained machine learning model configured to identify the source or root cause of an identified defect based on property data (e.g., Figure 5A the property data of block 502). At Figure 5A block 508, processing logic may cause the execution of a corrective action associated with the substrate processing system using the source or root cause identified via the trained machine learning model.

[0126] In some embodiments, the historical property data of block 510 is historical property data of a historical substrate, and the historical performance data of block 512 corresponds to the historical substrate.

[0127] At block 514, the machine learning model may be trained using a data input that includes historical property data 144 and a target output that includes historical property 144 data to generate a trained machine learning model configured to predict performance data 152 (e.g., performance data of the substrate, prediction data 160) based on property data 144 (e.g., Figure 5A the property data of block 502). In response to the predicted performance data meeting a first threshold (e.g., a threshold lethality rate for a defect of interest (DOI)), processing logic may cause a corrective action (e.g., shutting down, cleaning, repairing, replacing a substrate processing equipment part, etc.). In response to the predicted performance data meeting a second threshold (e.g., a lethality rate for DOI), processing logic may cause the substrate processing equipment part to be used in the substrate processing system.

[0128] In some embodiments, the historical property data 144 of block 510 includes historical property data 144 of a historical substrate, and the historical performance data 144 of block 512 includes historical property data 144.

[0129] At block 514, a machine learning model can be trained using a data input that includes historical property data 144 and a target output that includes historical property data 144 to generate a trained machine learning model configured to predict performance data 152 (e.g., performance data of a substrate processing equipment part) based on property data 144 (e.g., Figure 5A property data of blocks 502 and 504). In response to the predicted performance data meeting a first threshold, the processing logic can cause a corrective action (e.g., shutting down, cleaning, repairing, or replacing a substrate processing equipment part). In response to the predicted performance data meeting a second threshold, the processing logic can cause the substrate processing equipment part to be used in the substrate processing system.

[0130] Figure 5C is method 500C for performing defect analysis using a trained machine learning model (e.g., Figure 1 model 190) to cause the execution of a corrective action.

[0131] Reference Figure 5C , at block 520 of method 500C, the processing logic identifies property data 144. In some embodiments, the property data 144 of block 510 includes an image of the substrate.

[0132] At block 522, the processing logic provides the property data 144 as a data input to the trained machine learning model (e.g., the machine learning model trained via Figure 5B block 514).

[0133] At block 524, the processing logic receives an output associated with the prediction data from the trained machine learning model.

[0134] At block 526, the processing logic causes the execution of a corrective action based on the prediction data 160.

[0135] In some embodiments, the property data 144 is an image of the substrate, and the trained machine learning model of block 522 is trained using a data input that includes historical substrate images and a target output that includes historical performance data 154 (e.g., the substrate quality using historical substrate processing equipment parts).

[0136] In some embodiments, the property data 144 is an image of a substrate, and the trained machine learning model of block 522 is trained using data inputs including historical substrate images and target outputs including historical performance data 154, the historical performance data including historical property data 144 of defects corresponding to the historical substrates. The predicted data 160 of block 524 may be associated with performance data predicted based on the image (e.g., performance data of the substrate). In response to the predicted performance data meeting a first threshold (e.g., lethality), the processing logic may cause a corrective action (e.g., shutting down, cleaning, repairing, or replacing a substrate processing equipment part). In response to the predicted performance data meeting a second threshold (e.g., lethality), the processing logic may cause the substrate processing equipment part to be used in the substrate processing system.

[0137] In some embodiments, Figure 5A block 502 of includes training a machine learning model to identify property data of substrates processed by a substrate processing system and using the trained machine learning model to identify property data of substrates processed by the substrate processing system.

[0138] In some embodiments, Figure 5A block 504 of includes training a machine learning model to identify multiple regions of a substrate corresponding to a first defect category based on a first subset of the property data and using the trained machine learning model to identify multiple regions of the substrate corresponding to the first defect category based on the first subset of the property data.

[0139] In some embodiments, Figure 5A block 506 of includes training a machine learning model to subdivide multiple regions of a substrate corresponding to a first defect category into multiple defect subcategories based on a second subset of the property data and using the trained machine learning model to subdivide multiple regions of the substrate corresponding to the first defect category into multiple defect subcategories based on the second subset of the property data.

[0140] In some embodiments, Figure 5A block 508 of includes training a machine learning model to cause the execution of a corrective action associated with the substrate processing system based on at least one of the multiple defect subcategories and using the trained machine learning model to cause the execution of a corrective action associated with the substrate processing system based on at least one of the multiple defect subcategories.

[0141] In some embodiments, the processing logic trains a machine learning model to identify a subtype of interest and uses the trained machine learning model to identify a subcategory of interest.

[0142] Figure 6According to some embodiments, a block diagram of a computer system 600 is shown. In some embodiments, the computer system 600 is one or more of a client device 120, a prediction system 110, a server machine 170, a server machine 180, a prediction server 112, and the like.

[0143] In some embodiments, the computer system 600 is connected to other computer systems (e.g., via a network connection such as a local area network (LAN), an internal network, an external network, or the Internet). In some embodiments, the computer system 600 operates as a server or a client computer in a client-server environment, or as a peer computer in a peer-to-peer or distributed network environment. In some embodiments, the computer system 600 is provided by a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, a web appliance, a server, a network router, a switch, or a bridge, or any device capable of executing a set of instructions (sequentially or otherwise) that specify actions to be taken by the device. Further, the term "computer" shall include any collection of computers that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methods described herein.

[0144] In another aspect, the computer system 600 includes a processing device 602, a volatile memory 604 (e.g., random access memory (RAM)), a non-volatile memory 606 (e.g., read-only memory (ROM) or electrically erasable programmable ROM (EEPROM)), and a data storage device 618, which communicate with each other via a bus 608.

[0145] In some embodiments, the processing device 602 is provided by one or more processors such as a general-purpose processor (e.g., a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a microprocessor implementing other types of instruction sets, or a combination of microprocessors implementing multiple types of instruction sets) or a special processor (e.g., an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), or a network processor).

[0146] In some embodiments, the computer system 600 further includes a network interface device 622 (which is coupled to a network 674, for example). In some embodiments, the computer system 600 also includes a video display unit 610 (e.g., an LCD), an alphanumeric input device 612 (e.g., a keyboard), a cursor control device 614 (e.g., a mouse), and a signal generation device 620.

[0147] In some implementations, the data storage device 618 includes a non-transitory computer-readable storage medium 624 having instructions 626 stored thereon that encode any one or more of the methods or functions described herein, including encoding Figure 1 components (e.g., the calibration action component 122, the prediction component 114, etc.) and instructions for implementing the methods described herein (e.g., one or more of methods 500A to 500C).

[0148] In some embodiments, the instructions 626 also reside, at least partially, within the volatile memory 604 and / or the processing device 602 during the time they are executed by the computer system 600. Thus, in some embodiments, the volatile memory 604 and the processing device 602 also constitute machine-readable storage media.

[0149] Although the computer-readable storage medium 624 is shown as a single medium in the illustrative examples, the term "computer-readable storage medium" should also include a single medium or multiple media (e.g., a centralized or distributed database and / or associated caches and servers) that store one or more sets of executable instructions. The term "computer-readable storage medium" should also include any tangible medium capable of storing or encoding a set of instructions for execution by a computer, which cause the computer to perform any one or more of the methods described herein. The term "computer-readable storage medium" should include, but is not limited to, solid-state memory, optical media, and magnetic media.

[0150] In some embodiments, the methods, components, and features described herein are implemented by discrete hardware components or are integrated into the functionality of other hardware components such as ASICs, FPGAs, DSPs, or similar devices. In some embodiments, the methods, components, and features are implemented by firmware modules or functional circuitry within a hardware device. In some embodiments, the methods, components, and features are implemented in any combination of a hardware device and computer program components or as a computer program.

[0151] Unless otherwise specifically stated, terms such as "identify", "sub - classify", "cause", "provide", "obtain", "determine", "mask", "resize", "perform", "transform", "apply", "associate", "compare", "train", "receive", "update", etc. refer to actions and processes performed or implemented by a computer system that manipulate and transform data represented as physical (electronic) quantities within a computer system's registers and memory into other data similarly represented as physical quantities within a computer system's memory or registers or other such information storage, transmission, or display devices. In some embodiments, the terms "first", "second", "third", "fourth", etc. as used herein are intended as labels to distinguish different elements and do not have an order significance based on their numerical labels.

[0152] The examples described herein also relate to an apparatus for performing the methods described herein. In some embodiments, this apparatus is specifically constructed to perform the methods described herein or comprises a general - purpose computer system selectively programmed by a computer program stored in a computer system. Such a computer program is stored in a computer - readable tangible storage medium.

[0153] The methods and illustrative examples described herein have no inherent association with any particular computer or other apparatus. In some embodiments, various general - purpose systems are used in accordance with the teachings described herein. In some embodiments, more specialized apparatuses are constructed to perform the methods described herein and / or each of their respective functions, routines, sub - routines, or operations. Examples of the structures for various such systems are set forth in the foregoing description.

[0154] The foregoing description is intended to be illustrative, not restrictive. While the present disclosure has been described with reference to specific illustrative examples and implementations, it will be recognized that the present disclosure is not limited to the examples and implementations described. The scope of the present disclosure should be determined with reference to the following claims and the full scope of the equivalents given by those claims.

Claims

1. A method, the method comprising: Identifying property data of a substrate processed by a substrate processing system; Based on a first subset of the property data, identifying a plurality of regions of the substrate corresponding to a first defect category; Based on a second subset of the property data, subdividing the plurality of regions of the substrate corresponding to the first defect category into a plurality of defect subcategories; And Based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subcategories, causing the execution of a correction action associated with the substrate processing system.

2. The method of claim 1, wherein the first subset of the property data comprises at least one of a scanning electron microscope (SEM) image or an energy dispersive x-ray microanalysis (EDX) image.

3. The method of claim 1, wherein the second subset of the property data comprises at least one of morphological data, size attribute data, dimensional attribute data, defect distribution data, spatial location data, elemental analysis data, wafer signature data, chip layer, chip layout data, edge data, or defect metadata.

4. The method of claim 1, the method further comprising: Providing the first subset of the property data as an input to a trained machine learning model; And Obtaining an output associated with prediction data from the trained machine learning model, wherein the identifying the plurality of regions of the substrate corresponding to the first defect category is based on the prediction data.

5. The method of claim 1, the method further comprising: Providing the second subset of the property data as an input to a trained machine learning model; And Obtaining an output associated with prediction data from the trained machine learning model, wherein the subdividing the plurality of regions into the plurality of defect subcategories is based on the prediction data.

6. The method of claim 1, wherein the causing the execution of the correction action comprises one or more of the following: Determining a defect source based on at least one of the plurality of defect subcategories, the correction action corresponding to the defect source; or Determining a defect root cause based on at least one of the plurality of defect subcategories, the correction action corresponding to the defect root cause.

7. The method of claim 6, wherein at least one of the defect source or the defect root cause is formed during a previous operation of a substrate manufacturing process.

8. The method of claim 6, wherein at least one of the defect source or the defect root cause originates from a previous layer provided by a substrate manufacturing process.

9. The method of claim 1, the method further comprising: Identify a first subset of the plurality of defect subcategories, the first subset including at least one of the plurality of defect subcategories, wherein the first subset of the plurality of defect subcategories corresponds to substrate property data that meets a threshold level.

10. The method of claim 9, wherein identifying the first subset of the plurality of defect subcategories is based on one or more of the following: User input indicating one or more subcategories of interest; or Prediction data associated with an output received from a trained machine learning model, the prediction data being based on a data input including historical defect subcategories and a target output including historical property data.

11. The method of claim 1, the method further comprising determining that at least one of the plurality of defect subcategories corresponds to a defect evolution associated with the execution of the corrective action.

12. A non-transitory computer-readable storage medium storing instructions that, when executed, cause a processing device to perform operations including the following: Identify property data of a substrate processed by a substrate processing system; Based on a first subset of the property data, identify a plurality of regions of the substrate corresponding to a first defect category; Based on a second subset of the property data, subdivide the plurality of regions of the substrate corresponding to the first defect category into a plurality of defect subcategories; and Based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subcategories, cause the execution of a corrective action associated with the substrate processing system.

13. The non-transitory computer-readable storage medium of claim 12, the operations further comprising: Provide the first subset of the property data as an input to a trained machine learning model; and Obtain, from the trained machine learning model, an output associated with prediction data, wherein identifying the plurality of regions of the substrate corresponding to the first defect category is based on the prediction data.

14. The non-transitory computer-readable storage medium of claim 12, the operations further comprising: Provide the second subset of the property data as an input to a trained machine learning model; and Obtain, from the trained machine learning model, an output associated with prediction data, wherein subdividing the plurality of regions into the plurality of defect subcategories is based on the prediction data.

15. The non-transitory computer-readable storage medium of claim 12, wherein causing the execution of the corrective action includes one or more of the following: Determine a defect source based on at least one of the plurality of defect subcategories, the corrective action corresponding to the defect source; or Determine a defect root cause based on at least one of the plurality of defect subcategories, the corrective action corresponding to the defect root cause.

16. The non-transitory computer-readable storage medium of claim 12, wherein the operations further include: Identify a first subset of the plurality of defect subcategories, the first subset including at least one of the plurality of defect subcategories, wherein the first subset of the plurality of defect subcategories corresponds to substrate property data that meets a threshold level.

17. The non-transitory computer-readable storage medium of claim 16, wherein the identifying the first subset of the plurality of defect subcategories is based on one or more of: user input indicating one or more subcategories of interest; or prediction data associated with an output received from a trained machine learning model, the prediction data being based on a data input including historical defect subcategories and a target output including historical property data.

18. A system, the system comprising: a memory; and a processing device coupled to the memory, the processing device configured to: identify property data of a substrate processed by a substrate processing system; identify, based on a first subset of the property data, a plurality of regions of the substrate corresponding to a first defect category; subclassify the plurality of regions of the substrate corresponding to the first defect category into a plurality of defect subcategories based on a second subset of the property data; and cause execution of a correction action associated with the substrate processing system based on one or more of the plurality of regions corresponding to at least one of the plurality of defect subcategories.

19. The system of claim 18, wherein the processing device is further configured to: provide the first subset of the property data as an input to a trained machine learning model; and obtain, from the trained machine learning model, an output associated with prediction data, wherein the processing device is configured to identify the plurality of regions of the substrate corresponding to the first defect category based on the prediction data.

20. The system of claim 18, wherein the processing device is further configured to identify the first subset of the plurality of defect subcategories based on one or more of: user input indicating one or more subcategories of interest; or prediction data associated with an output received from a trained machine learning model, the prediction data being based on a data input including historical defect subcategories and a target output including historical property data.

Images