Chip-level spatial label grouping using transfer learning
By using a machine learning-based transfer learning method to automatically classify wafer patterns, the problem of inflexible defect pattern recognition in existing technologies is solved, enabling fast and accurate defect identification and root cause analysis, thereby improving the yield of semiconductor manufacturing.
Patent Information
- Application Number
- CN202180050094.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2021-08-17
- Filing Date
- 2021-08-24
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2041-08-24
AI Technical Summary
Existing spatial marker analysis methods are inflexible in semiconductor manufacturing, prone to over-analysis or omission of defect patterns, resulting in time-consuming and non-scalable determination of defect properties, which affects the yield of semiconductor manufacturers.
A machine learning-based approach is adopted, which utilizes transfer learning to automatically classify the markings on the wafer image. Through deep neural networks and logistic regression models, the root cause of defects is quickly identified, a sample marking library is generated, and confidence is assessed.
It enables rapid and accurate identification and classification of wafer defects, reducing wafer scrap and improving semiconductor manufacturing yield and production efficiency.
Smart Images

Figure CN116113982B_ABST
Abstract
Description
[0001] Cross-reference of related applications
[0002] This application claims priority to Indian Patent Application No. 202041036573, filed August 25, 2020, and designated as U.S. Provisional Patent Application No. 63 / 089,036, filed October 8, 2020, the disclosures of which are hereby incorporated by reference. Technical Field
[0003] This disclosure relates to semiconductor manufacturing. Background Technology
[0004] The evolution of the semiconductor manufacturing industry is placing higher demands on yield management, and specifically on metrology and inspection systems. Critical dimensions continue to shrink, but the industry needs to reduce the time required to achieve high-yield, high-value production. Minimizing the total time from detecting a yield problem to resolving it maximizes the semiconductor manufacturer's return on investment.
[0005] Manufacturing semiconductor devices (such as logic and memory devices) typically involves processing semiconductor wafers using a wide range of manufacturing processes to form the different features and multiple layers of the semiconductor devices. For example, photolithography is a semiconductor manufacturing process that involves transferring a pattern from a photomask to a photoresist disposed on a semiconductor wafer. Additional examples of semiconductor manufacturing processes include (but are not limited to) chemical mechanical polishing (CMP), etching, deposition, and ion implantation.
[0006] Most monolithic integrated circuits are manufactured as multiple batches of devices on a single wafer. Having many such devices on a single wafer makes the devices easier to handle and reduces manufacturing costs. Because the functional elements within each device tend to be extremely small, they are susceptible to damage. For example, particulate matter can be detrimental to a device regardless of whether the particulate matter is airborne or liquid-borne. Any particle that lands on the surface of the wafer can interfere with the manufacturing process and cause device failure if it is not removed in a timely manner. By analogy, scratches and other physical disturbances to the manufacturing process can also cause device failure.
[0007] Because it is so important to identify and correct the causes of such physical anomalies (e.g., defects) as quickly as possible, physical inspection of wafers is typically performed at various stages of the manufacturing cycle. Inspection systems attempt not only to determine the type of defect (e.g., scratches or particulate matter) but also its source. In many cases, the type and pattern of the defect can provide clues about its origin. Typically, such inspections are performed manually by trained technicians or engineers who examine one or more wafers under a particular type of microscope, searching for defects and attempting to determine their source based on experience.
[0008] Manual inspection of each wafer is cumbersome and prone to inaccurate and inconsistent results due to factors such as fatigue, lack of experience, or carelessness. It is also impractical in manufacturing environments due to the large number of wafers involved. Spatial marker analysis provides the ability to automatically track problems in integrated circuit wafer processes. Spatial marker analysis can be performed on wafers at different stages of the manufacturing process to detect certain patterns of defects on the wafer. Identified patterns can be mapped to the different processes the wafer has undergone. For example, a defective CMP process can cause long bending scratches. Therefore, process problems can be automatically detected without resorting to detailed studies of a subset of microscopic defects (which typically require scanning electron microscopy). This, in turn, leads to faster corrective actions, improving yield and increasing profitability.
[0009] Unfortunately, spatial marker analysis is an inflexible process and often tends to overanalyze individual defects, repeating this process without identifying patterns of possible defects. Therefore, spatial marker analysis typically has the problem of determining the nature of defects, tends to be too time-consuming, and is not scalable or reconfigurable beyond its original implementation capabilities and parameters.
[0010] Previous systems used parametric rules with wafer markings. In previous examples, pixel-level marking analysis platforms detected markings by setting a series of parametric rules or by using pattern templates. Parameter searches were limited to shapes that could be described by basic mathematics. Configuring independent rules for each shape led to overfitting or missed detections. Furthermore, most process markings cannot be defined using simple rules such as lines, arcs, and circles. Template searches were used to address complex shapes, but this was limited by user-defined templates. Key markings encompass several variations in shape, size, density, and integrity. These previous parametric rules delayed the discovery of affected wafers and any root cause analysis, reducing semiconductor manufacturers' yields.
[0011] Therefore, improved methods and systems are needed. Summary of the Invention
[0012] A method is provided in a first embodiment. The method includes receiving a wafer map at a processor, wherein the wafer map is for the entire surface of a wafer. The processor is configured to run a machine learning-based model for classifying the wafer map. The wafer map is classified using the machine learning-based model and labels on the wafer map. The machine learning-based model uses transfer learning.
[0013] The method may further include determining a confidence level for the classification. The confidence level is based on the criticality of the defects in the markers.
[0014] The method may further include sending a warning when the flag is outside the confidence level.
[0015] The method may further include: receiving a plurality of sample wafer images at the processor; classifying the sample wafer images based on the root cause of defects using the processor; and generating a sample label library from the sample wafer images using the processor. In one example, the processor may be used to extract images from a plurality of defect images, hierarchical ranking images, and / or metric images, and the processor may be used to augment the images. In one example, the machine learning-based model is trained using extracted features from the images.
[0016] The method may include grouping the wafer pattern together with wafer patterns that have the same classification.
[0017] A second embodiment provides a system. The system includes a semiconductor wafer inspection system and a processor in electronic communication with the semiconductor wafer inspection system. The processor is configured to receive a wafer image and classify the wafer image using the machine learning-based model and labels on the wafer image. The wafer image is for the entire surface of a wafer. The processor is configured to run a machine learning-based model for classifying the wafer image. The machine learning-based model uses transfer learning.
[0018] The semiconductor wafer inspection system may include a light source or an electron beam source.
[0019] The semiconductor wafer inspection system can generate the wafer pattern.
[0020] The processor can be further configured to determine a confidence level for the classification. The confidence level may be based on the criticality of the defects in the flags.
[0021] The processor may be further configured to send a warning if the flag is outside the confidence level.
[0022] The processor may be further configured to: receive a plurality of sample wafer images; classify the sample wafer images based on the root cause of defects; and generate a sample label library from the sample wafer images. In one example, the processor is further configured to extract images from a plurality of defect images, hierarchical ranking images, and / or metric images, and augment the images. In one example, the processor is further configured to train the machine learning-based model using extracted features from the images.
[0023] The processor may be further configured to group the wafer pattern together with wafer patterns of the same classification.
[0024] A third embodiment provides a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium includes one or more programs for performing the following steps on one or more computing devices: classifying the wafer pattern using a machine learning-based model and labels on the wafer pattern; the machine learning-based model using transfer learning; and the wafer pattern covering the entire surface of a wafer.
[0025] The steps may further include: receiving multiple sample wafer images; classifying the sample wafer images based on the root cause of the defects; and generating a sample label library from the sample wafer images.
[0026] The steps may further include extracting images from multiple defect maps, hierarchical sorting maps, and / or metric maps, and augmenting the images.
[0027] The step may further include training the machine learning-based model using extracted features from the image. Attached Figure Description
[0028] For a more complete understanding of the nature and objectives of this disclosure, reference should be made to the following detailed description taken in conjunction with the accompanying drawings, wherein:
[0029] Figure 1 This is a flowchart of the operation according to this disclosure;
[0030] Figure 2 Explanatory classification of the logo;
[0031] Figure 3 This is a flowchart of the classification of marks according to this disclosure;
[0032] Figure 4 This is a prime example of a problem with chip processors;
[0033] Figure 5 It is a model of the CMP problem;
[0034] Figure 6 Other exemplary signs illustrating various semiconductor manufacturing problems;
[0035] Figure 7 A flowchart illustrating an embodiment of the method according to this disclosure; and
[0036] Figure 8 This is an embodiment of the system according to the present disclosure. Detailed Implementation
[0037] Although the claimed subject matter will be described with reference to specific embodiments, other embodiments (including those that do not provide all the benefits and features set forth herein) are also within the scope of this disclosure. Various structural, logical, process, and electronic changes may be made without departing from the scope of this disclosure. Therefore, the scope of this disclosure is defined only with reference to the appended claims.
[0038] Transfer learning and machine learning can be used to automatically classify wafer markers at the wafer level. Wafer markers provide information to semiconductor manufacturers because they aid in root cause analysis. Figure 4 It serves as a demonstrative example of a chip processor problem that demonstrates three points of impact. Figure 5 It serves as an exemplary marker for CMP problems that demonstrate long-arc scratches. Figure 6 Other exemplary signs illustrating various semiconductor manufacturing problems. Each of these examples has a different origin, and quickly identifying the signs on the wafer can help determine the root cause of the problem.
[0039] Existing applications are manual, cumbersome, and insufficient, especially given the distribution of marker shapes falling into a single category. Transfer learning can be used to classify multi-marker, multi-class markers on chip maps. This helps to modify newly introduced markers because retraining is fast and tends to require low computational power.
[0040] Figure 7 This is an embodiment of method 100. Some or all of the steps of method 100 may be performed using a processor. At 101, a wafer pattern is received at the processor. The wafer pattern is for the entire surface of the wafer, which is... Figures 4 to 6 The example is shown. The surface can be a flat surface of the wafer or a surface containing devices on the wafer (e.g., a bare die or integrated circuit). The processor is configured to run a machine learning-based model for classifying the wafer map.
[0041] In section 102, a machine learning-based model and markers on the wafer pattern are used to classify the wafer pattern. The machine learning-based model uses transfer learning. Wafer markers can help semiconductor manufacturers. These can be used to find the root causes of errors. Early detection of root causes means fewer wafer scraps. This increases the yield of semiconductor manufacturers. By automating and quantifying the markers, semiconductor manufacturers can also react more quickly to in-line excursions and run root cause analysis.
[0042] exist Figure 3 In the embodiment shown, sample wafer patterns are received. These sample wafer patterns are classified based on the root cause of the defects. For example, Figure 2 Explain the exemplary classification of the logo. For example... Figure 3The example demonstrates how a sample label library is generated from sample wafer images. Machine learning-based models can be trained using the extracted features from these images.
[0043] exist Figure 3 In this embodiment, images are extracted from defect maps, hierarchical sorting maps, and / or metric maps. A defect map is a wafer image showing the location of detected defects. A hierarchical sorting map is a wafer image showing the pass / fail status of dies within the wafer. A metric map is a wafer image showing the measured properties of actual components within the wafer or specific target markers placed within the wafer. These images can then be augmented.
[0044] Extraction can be performed. A deep neural network can be trained to map images to categories. The intermediate layers of the deep neural network provide an abstract representation of the image. During extraction, the image is processed and features are extracted from the output of one of the intermediate layers. If features are generated using a deep neural network that has been trained on different image datasets, then transfer learning is performed.
[0045] Augmentation can be performed. Augmentation involves transforming the original image (i.e., a small dataset) in a way that allows the original image to be considered a different example, resulting in a larger dataset. It does not transform the image into a portion that can be considered a completely different dataset.
[0046] use Figure 1 The flowchart further explains Figure 3 An example is provided. Features are extracted from the augmented image. These features are then classified and added to a sample label library. Users can validate and / or retrain the machine learning-based model based on instances in the sample label library. Manual classification can be used to provide real-world data. For example, a semiconductor manufacturer can classify one or two instances, and then a machine learning-based model can determine the remainder. Figure 1 In this example, images are classified into category An. Machine learning-based models can examine trends, track defect sources, and monitor production. Machine learning-based models can also identify complex indicators used in specific wafer processing steps or tools.
[0047] Embodiments of method 100 can address the range of flags found in a semiconductor manufacturing environment that are attributable to wafer scrap and affect yield loss. Flag categories that cannot be defined using existing methods can be classified because the input is an image rather than data.
[0048] In one example, a confidence level for the classification can be determined. The confidence level can be based on the criticality of the defects in the flag. Deviations from the confidence level can be monitored. For example, if a flag falls outside the confidence level, a warning can be issued. Confidence metrics enable semiconductor manufacturers to control the purity of flag classification based on the criticality of the flag classification.
[0049] In one example, chip images can be grouped together with chip images belonging to the same category. Labels can be grouped in an organized manner based on different shapes, sizes, and formats.
[0050] Transfer learning can be used with machine learning-based models. It is a machine learning technique where a model trained on one task is retrained for a second, related task. When modeling the second task, transfer learning can achieve rapid progress or improved performance.
[0051] The embodiments disclosed herein enable automated wafer-level flag analysis using a combination of transfer learning and deep learning techniques. In the prototype, flag categories were detected with greater than 80% accuracy and recalled based on the initial analysis. In a specific instance, approximately 3000 production wafers were analyzed. The purity of flag categories can be controlled by a confidence matrix. The embodiments disclosed herein are applicable to semiconductor manufacturing across wafers, layers, and devices. Tool stability can be monitored throughout the cycle.
[0052] To train a model to detect CMP problems on a chip, a model already trained on a similar domain (e.g., chip disposal problems) can be used. Figure 4 and 5 These types of labels are shown in the diagram. However, in practice, performance degradation or breakdown can occur because the model inherits biases from its training data and doesn't know how to generalize to new domains. Existing models cannot be reused to train for new tasks (such as detecting etching defects) because the labels differ between tasks.
[0053] In practice, it can be beneficial to transfer as much knowledge as possible from the source setting to the target task or domain. Depending on the data, this knowledge can take various forms. For example, it could be about how objects are constructed to identify novel objects.
[0054] Transfer learning can address these cases by leveraging existing labeled data from related tasks or domains. This knowledge, gained while solving the source task in the source domain, can be stored and applied to new chip labels.
[0055] Transfer learning can generally be defined as an improvement on learning in a new task (or target task) through the transfer of knowledge from related tasks (one or more source tasks) that have already been learned. In the embodiments described herein, training a machine learning-based model using only nominal examples involves learning one or more source tasks, and retraining the machine learning-based model using non-nominal examples transfers knowledge from the source tasks (nominal examples) to the target task (non-nominal examples). In transfer learning, the agent is unaware of the target task (or even unaware that a target task will exist) when it learns the source tasks. For example, in the embodiments described herein, the machine learning-based model is unaware of new labels when it is trained using previously received data. However, in general, transfer learning as described herein can be performed in any suitable manner known in the art.
[0056] Inductive transfer can be viewed not only as a way to improve learning in standard supervised learning tasks, but also as a way to offset the difficulties posed by tasks involving relatively small datasets. That is, if a task has a relatively small amount of data or class labels, treating it as the target task and performing an inductive transfer from the relevant source task can lead to a more accurate model. Although it is assumed that the two datasets come from different probability distributions, these methods can use source task data to augment target task data. Transfer learning as described herein can be further performed as described in U.S. Patent Application No. 20170193400, which is incorporated herein by reference as if its entirety were stated. The embodiments described herein can be further configured as described in this patent application and related applications disclosed herein.
[0057] The retraining (and training) architecture used in the embodiments described herein can be designed to converge to the real data (for validation samples) using a minimum number of samples. In the embodiments, one or more components include one or more additional components used to perform retraining. The one or more additional components may include a common parent network for all layers on the sample, a large common parent network for all layers on the sample, or other components. For example, transfer learning methods can be applied to training sets generated using any of the transfer learning training input generation methods described herein to train machine learning-based models.
[0058] In one example, a deep neural network (e.g., a VGG19 deep neural network) is used to extract feature vectors from a wafer image. The deep neural network can then be followed by a series of logistic regression models to perform classification. The deep neural network can be a machine learning-based model.
[0059] Generally speaking, "deep learning" (also known as deep structured learning, hierarchical learning, or deep machine learning) is a branch of machine learning based on a set of algorithms that attempt to model higher-order abstractions of data. In a simple case, there can be two sets of layers: one set that receives input signals and another set that sends output signals. When an input layer receives input, it passes a modified version of the input to the next layer. In deep networks, there are many layers between input and output, allowing algorithms to use multiple processing layers consisting of multiple linear and nonlinear transformations.
[0060] Deep learning is part of a broader family of machine learning methods that learn representations based on data. Observations (e.g., images) can be represented in many ways (e.g., vectors of intensity values per pixel) or in more abstract ways (as a set of edges, regions of a specific shape, etc.). Some representations are better than others at simplifying learning tasks (e.g., face recognition or facial expression recognition). A promise of deep learning is to replace handcrafted features with efficient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction.
[0061] Research in this field attempts to create better representations and generate models to learn these representations from large-scale unlabeled data. Some representations are inspired by advances in neuroscience and are loosely based on the interpretation of information processing and communication patterns in the nervous system, such as attempts to define neural codes that define the relationship between various stimuli and the responses of associated neurons in the brain.
[0062] In some embodiments, the deep learning model is a generative model. A “generative” model can be broadly defined as a model that is inherently probabilistic. A generative model can be learned based on a suitable training dataset (where its parameters can be learned). In embodiments, the deep learning model is configured as a deep generative model. For example, the model can be configured to have a deep learning architecture, wherein the model may contain multiple layers that perform several algorithms or transformations.
[0063] In another embodiment, the deep learning model is configured as a neural network. In a further embodiment, the deep learning model may be a deep neural network with a set of weights that model the world based on data that has been fed to train the model. Neural networks can be broadly defined as computational methods based on relatively large assemblies of neural units, loosely modeling how the biological brain solves problems using relatively large clusters of biological neurons connected by axons. Each neural unit is connected to many other neural units, and the links can enforce or inhibit their effects on the firing states of the connected neural units. These systems are learned and trained rather than explicitly programmed and have advantages in domains where solutions or feature detection are difficult to express in traditional computer programs.
[0064] Neural networks typically consist of multiple layers, with signal paths traversing from front to back. The goal of neural networks is to solve problems in a manner similar to the human brain, although some neural networks are far more abstract. Modern neural network projects often utilize thousands to millions of neural units and millions of connections. Neural networks can have any suitable architecture and / or configuration known in their field.
[0065] In one embodiment, the information includes a classification of defects detected on the sample. In one such embodiment, the deep learning model is configured as AlexNet. For example, AlexNet includes several convolutional layers (e.g., 5) followed by several fully connected layers (e.g., 3), which are configured and trained in combination to classify images.
[0066] In another embodiment, the deep learning model is configured as GoogleNet. For example, GoogleNet may contain layers such as convolutional layers, pooling layers, and fully connected layers, such as those configured and trained to classify images, as further described herein. While the GoogleNet architecture can contain a relatively high number of layers, some layers can operate in parallel, and groups of layers operating in parallel with each other are often referred to as starter modules. Other layers can operate sequentially. Therefore, GoogleNet differs from other neural networks described herein in that not all layers are arranged in a sequential structure.
[0067] In a further embodiment, the deep learning model is configured as a VGG network. For example, a VGG network is formed by increasing the number of convolutional layers while keeping other parameters of the architecture fixed. Adding convolutional layers to increase depth is possible by using generally small convolutional filters across all layers. Like other neural networks described herein, the VGG network is formed and trained to classify images. The VGG network also includes convolutional layers followed by fully connected layers. An example of a neural network configured as VGG is described in “Very Deep Convolutional Networks for Large-Scale Image Recognition” by Simonyan et al., ICLR 2015, which is incorporated herein by reference in its entirety. The deep learning models described herein can be further configured as described in the references.
[0068] In some such embodiments, the deep learning model is configured as a deep residual network. For example, as with some other networks described herein, a deep residual network may contain convolutional layers followed by fully connected layers, which are configured and trained in combination for image classification. In a deep residual network, these layers are configured to learn residual functions from reference layer inputs rather than learning unreferenced functions. Specifically, instead of expecting each stacked layer to directly fit the desired underlying mapping, these layers are explicitly allowed to fit residual mappings, which is achieved by a feedforward neural network with shortcut connections. Shortcut connections are connections that skip one or more layers. A deep residual network can be formed by taking a regular neural network structure containing convolutional layers and inserting shortcut connections, thereby taking a regular neural network and transforming it into its residual learning counterpart.
[0069] In a further embodiment, the deep learning model includes one or more fully connected layers configured to classify defects on a sample. A “fully connected layer” can be broadly defined as a layer in which each node is connected to each node in a previous layer. The fully connected layer can perform classification based on features extracted by the convolutional layers, and it can be configured as further described herein. The fully connected layer is configured for feature selection and classification. In other words, the fully connected layer selects features from the feature map and then classifies defects in the image based on the selected features. The selected features may include all features in the feature map (if appropriate) or only some features in the feature map.
[0070] If a deep learning model outputs a classification for defects detected on a sample, the deep learning model can output an image classification, which may include a classification result for each image and a confidence level associated with each classification result. The results of the image classification can also be used as further described herein. The image classification can have any suitable format (e.g., image or defect ID, defect description such as "pattern," "bridge," etc.). The image classification results can be stored and used as further described herein.
[0071] In some embodiments, the information determined by the deep learning model includes features of the image extracted by the deep learning model. In one embodiment, the deep learning model includes one or more convolutional layers. The convolutional layers may have any suitable configuration known in the field and are typically configured to determine features (i.e., feature maps) of the image that vary across the image location by applying a convolutional function to the input image using one or more filters. In this way, the deep learning model (or at least a portion of the deep learning model) may be configured as a convolutional neural network (CNN). For example, the deep learning model may be configured as a CNN for extracting local features, which is typically a stack of convolutional and pooling layers. The embodiments described herein can leverage deep learning concepts (e.g., CNNs) to solve the generally difficult problem of representation inversion. The deep learning model may have any CNN configuration or architecture known in the field. The one or more pooling layers may also have any suitable configuration known in the field (e.g., max pooling layers) and are typically configured to reduce the dimensionality of the feature maps generated by the one or more convolutional layers while preserving the most important features.
[0072] Features determined by a deep learning model may include any suitable features known in the field that can be inferred from the inputs described herein (and may be used to produce the outputs further described herein). For example, features may include a vector of intensity values for each pixel. Features may also include any other types of features described herein, such as vectors of scalar values, vectors of independent distributions, vectors of joint distributions, or any other suitable feature types known in the field.
[0073] Generally, the deep learning models described herein are trained deep learning models. For example, a deep learning model can be trained beforehand to perform one or more diagnostic functions using one or more other systems and / or methods. Additionally, a deep learning model can be trained using one or more of the embodiments described herein before performing one or more diagnostic functions on the deep learning model. In this way, the diagnostic functions described herein differ from the deep learning model characterization performed during the training of the deep learning model, because in the embodiments described herein, a deep learning model has already been generated and trained, and its functionality has then been determined as described herein, which can then be used to perform one or more additional functions of the deep learning model.
[0074] exist Figure 8An embodiment of system 200 is shown. System 200 includes an optically based subsystem 201. Generally, the optically based subsystem 201 is configured to generate an optically based output for sample 202 by directing light to (or scanning light across) sample 202 and detecting the light from sample 202. In one embodiment, sample 202 includes a wafer. The wafer may include any wafer known in the art. In another embodiment, sample 202 includes a photomask. The photomask may include any photomask known in the art.
[0075] exist Figure 8 In the embodiment of system 200 shown, the optical-based subsystem 201 includes an illumination subsystem configured to direct light to sample 202. The illumination subsystem includes at least one light source. For example, as... Figure 8 As shown, the illumination subsystem includes a light source 203. In one embodiment, the illumination subsystem is configured to direct light onto the sample 202 at one or more incident angles (which may include one or more tilt angles and / or one or more normal angles). For example, as... Figure 8 As shown, light from light source 203 is guided at an angle of incidence through optical element 204 and then through lens 205 to sample 202. The angle of incidence can include any suitable angle of incidence, which can vary depending on, for example, the characteristics of sample 202.
[0076] The optical-based subsystem 201 can be configured to guide light to the sample 202 at different incident angles at different times. For example, the optical-based subsystem 201 can be configured to modify one or more characteristics of one or more elements of the illumination subsystem, such that the light can be directed to the sample 202 at different incident angles. Figure 8 The different incident angles shown are guided to sample 202. In one example, the optical-based subsystem 201 can be configured to move the light source 203, optical element 204 and lens 205 such that light is guided to sample 202 at different tilted incident angles or normal (or near-normal) incident angles.
[0077] In some examples, the optical-based subsystem 201 may be configured to guide light to the sample 202 simultaneously at more than one incident angle. For example, the illumination subsystem may include more than one illumination channel, one of which may include, for example, Figure 8The light source 203, optical element 204, and lens 205 shown herein, and another of the illumination channels (not shown), may contain similar elements that may be configured differently or identically, or may contain at least one light source and one or more other components, such as those further described herein. If this light and another light are simultaneously directed to the sample, then one or more characteristics (e.g., wavelength, polarization, etc.) of the light directed to the sample 202 at different incident angles may be different, so that the light generated by illuminating the sample 202 at different incident angles can be distinguished from each other at the detector.
[0078] In another example, the lighting subsystem may contain only one light source (e.g., Figure 8 The light source 203 shown in the diagram can be separated into different optical paths (e.g., based on wavelength, polarization, etc.) by one or more optical elements (not shown) of the illumination subsystem. Light in each of the different optical paths can then be directed to sample 202. Multiple illumination channels can be configured to direct light to sample 202 at the same time or at different times (e.g., when using different illumination channels to sequentially illuminate the sample). In another example, the same illumination channel can be configured to direct light to sample 202 at different times using different characteristics. For example, in some examples, optical element 204 can be configured as a spectral filter, and the properties of the spectral filter can be changed in various different ways (e.g., by substituting the spectral filter) such that light of different wavelengths can be directed to sample 202 at different times. The illumination subsystem can have any other suitable configuration known in the art for sequentially or simultaneously directing light with different or identical characteristics to sample 202 at different or identical incident angles.
[0079] In one embodiment, light source 203 may comprise a broadband plasma (BBP) source. In this way, the light generated by light source 203 and directed to sample 202 may comprise broadband light. However, the light source may comprise any other suitable light source, such as a laser. The laser may comprise any suitable laser known in the art and may be configured to produce light of any suitable wavelength or wavelength known in the art. Furthermore, the laser may be configured to produce monochromatic or near-monochromatic light. In this way, the laser may be a narrowband laser. Light source 203 may also comprise a multicolor light source that produces light of multiple discrete wavelengths or bands.
[0080] Light from optical element 204 can be focused onto sample 202 by lens 205. Although lens 205 is... Figure 8 While shown as a single refractive optical element, it should be understood that, in reality, lens 205 may comprise several refractive and / or reflective optical elements that combine to focus light from said optical elements onto the sample. Figure 8The illumination subsystem shown and described herein may include any other suitable optical elements (not shown). Examples of such optical elements include (but are not limited to) polarizing components, spectral filters, spatial filters, reflective optics, apodizers, beam splitters (e.g., beam splitter 213), apertures, and the like, which may include any such suitable optical elements known in the art. Furthermore, the optical-based subsystem 201 may be configured to modify one or more elements of the illumination subsystem to produce an optical-based output based on the type of illumination.
[0081] The optical-based subsystem 201 may also include a scanning subsystem configured to scan light across the sample 202. For example, the optical-based subsystem 201 may include a stage 206 on which the sample 202 is placed during optical-based output generation. The scanning subsystem may include any suitable mechanical and / or robotic assembly (which includes the stage 206) configured to move the sample 202 such that light can scan across it. Alternatively, the optical-based subsystem 201 may be configured such that one or more optical elements of the optical-based subsystem 201 perform a scan of light across the sample 202. Light can scan across the sample 202 in any suitable manner (e.g., in a serpentine or helical path).
[0082] The optical-based subsystem 201 further includes one or more detection channels. At least one of the detection channels includes a detector configured to detect light originating from the sample 202 attributable to illumination of the sample 202 by the subsystem and to generate an output in response to the detected light. For example, Figure 8 The optical-based subsystem 201 shown includes two detection channels: one formed by a light collector 207, an element 208, and a detector 209, and the other formed by a light collector 210, an element 211, and a detector 212. Figure 8 As shown, the two detection channels are configured to collect and detect light at different focusing angles. In some examples, the two detection channels are configured to detect scattered light, and the detection channels are configured to detect light scattered from sample 202 at different angles. However, one or more detection channels are configured to detect another type of light from sample 202 (e.g., reflected light).
[0083] like Figure 8The diagram further illustrates that the two detection channels are positioned within the plane of the paper, and the illumination subsystem is also positioned within the plane of the paper. Therefore, in this embodiment, the two detection channels are positioned (e.g., centered) within the plane of incidence. However, one or more detection channels may be positioned outside the plane of incidence. For example, a detection channel formed by the light collector 210, element 211, and detector 212 may be configured to collect and detect light scattered from the plane of incidence. Therefore, this detection channel may be collectively referred to as a "side" channel, and this side channel may be centered in a plane substantially perpendicular to the plane of incidence.
[0084] although Figure 8 An embodiment of an optically based subsystem 201 comprising two detection channels is shown, but the optically based subsystem 201 may comprise a different number of detection channels (e.g., only one detection channel or two or more detection channels). In one example, the detection channel formed by the light collector 210, element 211, and detector 212 may form a side channel as described above, and the optically based subsystem 201 may include an additional detection channel (not shown) formed as another side channel positioned on the opposite side of the incident plane. Thus, the optically based subsystem 201 may include a detection channel comprising a light collector 207, element 208, and detector 209 and centered in the incident plane and configured to collect and detect light at a scattering angle normal to or near normal to the surface of sample 202. Therefore, this detection channel may be collectively referred to as the “top” channel, and the optically based subsystem 201 may also include two or more side channels configured as described above. Therefore, the optical-based subsystem 201 may include at least three channels (i.e., a top channel and two side channels), and each of the at least three channels has its own light collector, each of which is configured to collect light at a scattering angle different from that of each of the other light collectors.
[0085] As further described above, each of the detection channels included in the optical-based subsystem 201 can be configured to detect scattered light. Therefore, Figure 8 The optical-based subsystem 201 shown herein can be configured for generating a dark-field (DF) output for a sample 202. However, the optical-based subsystem 201 may also, or alternatively, include a detection channel configured for generating a bright-field (BF) output for a sample 202. In other words, the optical-based subsystem 201 may include at least one detection channel configured to detect light reflected from the specular surface of the sample 202. Therefore, the optical-based subsystem 201 described herein can be configured for imaging DF only, BF only, or both DF and BF. Although each of the light collectors in Figure 8While shown as a single refractive optical element, it should be understood that each of the light collectors may contain one or more refractive optical slabs and / or one or more refractive optical elements.
[0086] The one or more detection channels may comprise any suitable detector known in the art. For example, the detector may comprise a photomultiplier tube (PMT), a charge-coupled device (CCD), a time-delay integration (TDI) camera, and any other suitable detector known in the art. The detector may also comprise a non-imaging detector or an imaging detector. In this way, if the detector is a non-imaging detector, each of the detectors may be configured to detect specific characteristics (e.g., intensity) of the scattered light, but may not be configured to detect such characteristics that vary depending on position within the imaging plane. Therefore, the output generated by each of the detectors in each of the detection channels of the optical-based subsystem may be a signal or data, rather than an image signal or image data. In such examples, a processor (e.g., processor 214) may be configured to generate an image of sample 202 from the non-imaging output of the detector. However, in other examples, the detector may be configured as an imaging detector, configured to generate an imaging signal or image data. Therefore, the optical-based subsystem may be configured to generate optical images or other optical-based outputs described herein in several ways.
[0087] It should be noted that the information provided in this article... Figure 8 The configuration of the optical-based subsystem 201 described herein is generally illustrated. This subsystem may be included in the system embodiments described herein, or it may produce optical-based output used by the system embodiments described herein. The configuration of the optical-based subsystem 201 described herein may be modified to optimize the performance of the optical-based subsystem 201, as is typically done when designing commercial output acquisition systems. Furthermore, the systems described herein can be implemented using existing systems (e.g., by adding the functionality described herein to an existing system). For some such systems, the methods described herein may be provided as optional functionality of the system (e.g., in addition to other functionalities of the system). Alternatively, the systems described herein may be designed as entirely new systems.
[0088] Processor 214 may be coupled to components of system 200 in any suitable manner (e.g., via one or more transmission media, which may include wired and / or wireless transmission media), enabling processor 214 to receive output. Processor 214 may be configured to perform several functions using said output. System 200 may receive instructions or other information from processor 214. Processor 214 and / or electronic data storage unit 215 may optionally communicate electronically with wafer inspection tools, wafer measurement tools, or wafer viewing tools (not described) to receive additional information or send instructions. For example, processor 214 and / or electronic data storage unit 215 may communicate electronically with a scanning electron microscope.
[0089] The processor 214, other systems, or other subsystems described herein may be part of a variety of systems, including personal computer systems, graphics computers, mainframe computer systems, workstations, network devices, Internet devices, or other devices. A subsystem or system may also include any suitable processor known in the art, such as a parallel processor. Furthermore, a subsystem or system may include a platform and software (as a standalone or network tool) with high-speed processing capabilities.
[0090] The processor 214 and electronic data storage unit 215 may be housed in or be part of system 200 or another device. In an example, the processor 214 and electronic data storage unit 215 may be part of a separate control unit or may be in a centralized quality control unit. Multiple processors 214 or electronic data storage units 215 may be used.
[0091] Processor 214 can be implemented by any combination of hardware, software, and firmware. Furthermore, its functions as described herein can be executed by a single unit or divided among different components, each of which can be implemented by any combination of hardware, software, and firmware. Program code or instructions for processor 214 to implement various methods and functions can be stored in a readable storage medium, such as the memory in electronic data storage unit 215 or other memory.
[0092] If system 200 includes more than one processor 214, then different subsystems can be coupled to each other, allowing images, data, information, instructions, etc., to be transmitted between the subsystems. For example, a subsystem can be coupled to an additional subsystem via any suitable transmission medium, which may include any suitable wired and / or wireless transmission medium known in the art. Two or more of such subsystems can also be effectively coupled by sharing a computer-readable storage medium (not shown).
[0093] Processor 214 may be configured to perform several functions using the output of system 200 or other outputs. For example, processor 214 may be configured to send the output to electronic data storage unit 215 or another storage medium. Processor 214 may be configured according to any of the embodiments described herein. Processor 214 may also be configured to perform other functions or additional steps using the output of system 200 or using images or data from other sources.
[0094] The various steps, functions, and / or operations of System 200 and the methods described herein are implemented by one or more of the following: electronic circuits, logic gates, multiplexers, programmable logic devices, ASICs, analog or digital controllers / switches, microcontrollers, or computing systems. Program instructions for implementing methods such as those described herein may be transmitted via or stored on a carrier medium. The carrier medium may include storage media such as read-only memory, random access memory, magnetic disks or optical disks, non-volatile memory, solid-state memory, magnetic tape, and the like. The carrier medium may include transmission media such as wires, cables, or wireless transmission links. For example, the various steps described throughout this disclosure may be implemented by a single processor 214 or alternatively by multiple processors 214. Furthermore, different subsystems of System 200 may include one or more computing or logic systems. Therefore, the above description should not be construed as a limitation of this disclosure, but is merely illustrative.
[0095] In one example, processor 214 communicates with system 200. Processor 214 is configured to execute an embodiment of method 100. In one example, processor 214 can operate a machine learning-based model. System 200 can inspect a wafer for defects, the results of which can be used to generate a wafer pattern received at processor 214.
[0096] Additional embodiments relate to a non-transitory computer-readable medium storing program instructions executable on a controller to perform a computer-implemented method for classifying wafer patterns, as disclosed herein. Specifically, as Figure 8 As shown herein, electronic data storage unit 215 or other storage media may contain non-transitory computer-readable media containing program instructions executable on processor 214. The computer-implemented method may include any step of any method described herein (including method 100).
[0097] The program instructions can be implemented in any of a variety of ways, including program-based technologies, component-based technologies, and / or object-oriented technologies. For example, the program instructions can be implemented using ActiveX controls, C++ objects, JavaBeans, Microsoft Foundation Classes (MFC), Streaming SIMD Extensions (SSE), or other technologies or methodologies.
[0098] Although system 200 uses light, method 100 can be performed using different semiconductor testing tools. For example, method 100 can be performed using results from a system that uses an electron beam (e.g., a scanning electron microscope) or a system that uses an ion beam. Therefore, the system can have an electron beam source or an ion beam source.
[0099] Although this disclosure has been described with respect to one or more specific embodiments, it will be understood that other embodiments of this disclosure may be made without departing from the scope of this disclosure. Therefore, this disclosure is to be considered limited only by the appended claims and their reasonable interpretation.
Claims
1. A method comprising: A wafer map is received at a processor, wherein the wafer map is for the entire surface of a wafer, and wherein the processor is configured to run a machine learning-based model for classifying the wafer map. The wafer image is classified using the machine learning-based model and the labels on the wafer image, wherein the machine learning-based model uses transfer learning, and wherein the confidence matrix is used to control the purity of the label categories in the machine learning-based model. Determine the confidence level for the classification, wherein the confidence level is based on the criticality of the defect in the label; and Determine whether the flag is outside the confidence level.
2. The method of claim 1, further comprising sending a warning if the flag is outside the confidence level.
3. The method according to claim 1, further comprising: Multiple sample wafer diagrams are received at the processor; The processor is used to classify the sample wafer images based on the root cause of the defects; and The processor is used to generate a sample label library from the sample wafer diagram.
4. The method of claim 3, further comprising: The processor is used to extract images from multiple defect maps, hierarchical sorting maps, and / or metric maps; and The image is augmented using the processor.
5. The method of claim 4, further comprising training the machine learning-based model using extracted features from the image.
6. The method of claim 1, further comprising grouping the wafer pattern together with wafer patterns having the same classification.
7. A system comprising: Semiconductor wafer inspection system; and A processor that communicates electronically with the semiconductor wafer inspection system, wherein the processor is configured to: Receive a wafer map, wherein the wafer map is for the entire surface of a wafer, wherein the processor is configured to run a machine learning-based model for classifying the wafer map; The wafer image is classified using the machine learning-based model and the labels on the wafer image, wherein the machine learning-based model uses transfer learning, and wherein the confidence matrix is used to control the purity of the label categories in the machine learning-based model. Determine the confidence level for the classification, wherein the confidence level is based on the criticality of the defect in the label; and Determine whether the flag is outside the confidence level.
8. The system of claim 7, wherein the semiconductor wafer inspection system comprises a light source or an electron beam source.
9. The system of claim 7, wherein the semiconductor wafer inspection system generates the wafer pattern.
10. The system of claim 7, wherein the processor is further configured to send a warning if the flag is outside the confidence level.
11. The system of claim 7, wherein the processor is further configured to: Receive multiple sample wafer images; The sample wafer images are classified based on the root cause of the defects; and A sample label library is generated from the sample wafer image.
12. The system of claim 11, wherein the processor is further configured to: Images are extracted from multiple defect maps, hierarchical sorting maps, and / or metric maps; and The image is amplified.
13. The system of claim 12, wherein the processor is further configured to train the machine learning-based model using extracted features from the image.
14. The system of claim 7, wherein the processor is further configured to group the wafer pattern together with wafer patterns having the same classification.
15. A non-transitory computer-readable storage medium comprising one or more programs for performing the following steps on one or more computing devices: The wafer image is classified using a machine learning-based model and labels on the wafer image, wherein the machine learning-based model uses transfer learning, wherein the wafer image is for the entire surface of the wafer, and wherein a confidence matrix is used to control the purity of the label categories in the machine learning-based model. Determine the confidence level for the classification, wherein the confidence level is based on the criticality of the defect in the label; and Determine whether the flag is outside the confidence level.
16. The non-transitory computer-readable storage medium of claim 15, wherein the step further comprises: Receive multiple sample wafer images; The sample wafer images are classified based on the root cause of the defects; and A sample label library is generated from the sample wafer image.
17. The non-transitory computer-readable storage medium of claim 16, wherein the step further comprises: Extract images from multiple defect maps, hierarchical ranking maps, and / or metric maps; and The image is amplified.
18. The non-transitory computer-readable storage medium of claim 17, wherein the step further comprises training the machine learning-based model using extracted features from the image.
