Method and system for classifying defects in semiconductor samples

Abstract

A system, method and computer-readable medium for classifying defects, the method comprising: receiving a classified first defect and a potential defect, each of the first defect and the potential defect having an attribute value; processing the first defect and the potential defect to select a subset of attributes that distinguish the first defect from the potential defect; obtaining a first function and a second function based on the first defect and the potential defect, respectively; obtaining a first threshold for the first function and a second threshold for a combination of the first function and the second function; applying the first function and the second function to each potential defect, respectively, to obtain a first score and a second score; and determining a combined score of the first score and the second score; and indicating the potential defect as a potential new category defect when the first score is lower than the first threshold or the combined score exceeds the second threshold.

Description

Technical Field

[0001] The presently disclosed subject matter relates generally to the field of sample inspection, and more particularly to methods and systems capable of automatically detecting defects belonging to new classes. Background Art

[0002] The current demands for high density and performance associated with ultra-large scale integration of manufactured devices require sub-micron features, increased transistors and increased circuit speeds, and improved reliability. These demands require that device features be formed with high precision and uniformity, which in turn requires careful monitoring of the manufacturing process, including frequent and detailed inspection of the devices while they are still in semiconductor sample form.

[0003] The term "sample" used in this specification should be broadly interpreted to cover any kind of wafers, masks and other structures, combinations and / or components thereof used to manufacture semiconductor integrated circuits, magnetic heads, flat panel displays and other semiconductor products.

[0004] The term "defect" as used in this specification should be interpreted broadly to cover any kind of abnormal or undesirable feature formed on or in a sample.

[0005] The complex manufacturing process of the sample is not error-free, and such errors may cause the manufactured device to malfunction. These faults may include defects that may impair the operation of the device, and hazards that may be defects but do not cause any damage to the manufactured device or cause the manufactured device to malfunction. As a non-limiting example, defects may be caused in the manufacturing process due to faults in the raw materials, mechanical, electrical or optical errors, human errors or other reasons. In addition, defects may be caused by spatiotemporal factors, such as changes in wafer temperature that occur after one or more manufacturing stages in the inspection process, which may cause some deformation of the wafer. The inspection process may also introduce other so-called errors, such as optical, mechanical or electrical problems in the inspection equipment or process, which thereby provide imperfect acquisition results. Such errors may produce false positive results, which may appear to contain defects, but there are no actual defects in the area.

[0006] In many applications, the type or category of defects is important. For example, defects can be classified into one of a number of categories, such as particles, scratches, processing, or the like.

[0007] Unless otherwise specifically stated, the term "inspection" used in this specification should be broadly interpreted as covering any kind of detection and / or classification of defects in an object. Inspection is provided by using a non-destructive inspection tool during or after the process of manufacturing the object to be inspected. As a non-limiting example, the inspection process may include using one or more inspection tools to provide scanning (in the form of a single or multiple scans), sampling, reviewing, measuring, classification and / or other operations for the object or part thereof. Similarly, the inspection can be provided before the object to be inspected is manufactured, and may include, for example, generating (multiple) inspection plans. It should be noted that, unless otherwise specifically stated, the term "inspection" or its derivatives used in this specification are not limited in terms of the size of (multiple) inspection areas, the speed or resolution of the scan, or the type of inspection tool. As non-limiting examples, various non-destructive inspection tools include optical tools, scanning electron microscopes, atomic force microscopes, and the like.

[0008] The inspection process may include multiple inspection steps. The inspection step may be performed multiple times during the manufacturing process, for example after the manufacture or processing of certain layers, or the like. Additionally or alternatively, each inspection step may be repeated multiple times, for example for different sample positions or for the same sample position with different inspection settings.

[0009] As a non-limiting example, a run-time examination may employ a two-step procedure, such as inspecting a sample and then reviewing the sampled defects. During the inspection step, the surface of a sample or a portion of a sample (e.g., an area of ​​interest, a hot spot, etc.) is scanned, typically at a relatively high speed and / or a relatively low resolution. The captured inspection images are analyzed to detect defects and obtain defect locations and other inspection attributes. In the review step, images of at least some of the defects detected in the inspection stage are captured, typically at a relatively low speed and / or a relatively high resolution, to enable classification and optional other analysis of the at least some of the defects. In some cases, the two stages may be implemented by the same inspection tool, while in some other cases, the two stages are implemented by different inspection tools. Summary of the invention

[0010] According to certain aspects of the presently disclosed subject matter, a system capable of classifying defects in a sample into a plurality of categories is provided, the system comprising a processing and memory circuitry (PMC), the PMC being configured to: receive a first plurality of first defects and a second plurality of potential second defects classified into a plurality of categories, each first defect and each potential second defect being associated with a corresponding value of each of a plurality of attributes, the first plurality of first defects being characterized as a first attribute space, and the second plurality of potential second defects being characterized as a second attribute space; process the first plurality of first defects and the second plurality of potential second defects to select a subset of attributes from the plurality of attributes, comprising: for each attribute: determining a first distribution of attribute values ​​of defects from the first plurality of first defects; determining a second distribution of attribute values ​​of defects from the second plurality of potential second defects; and determining a difference between the first distribution and the second distribution; selecting the subset of attributes as the attribute having the largest difference between the first distribution and the second distribution; Obtaining a first density estimation function based on values ​​assigned to a subset of attributes of defects from the first plurality of first defects; and determining a second density estimation function based on values ​​assigned to a subset of attributes of defects from the second plurality of potential second defects; obtaining a first threshold associated with the first density estimation function, and a second threshold associated with a combination of the first density estimation function and the second density estimation function; applying the first density estimation function to each defect of the second plurality of potential second defects to obtain a first score; applying the second density estimation function to each defect of the second plurality of potential second defects to obtain a second score; determining a combined score of the first score and the second score; and indicating a defect from the second plurality of potential second defects as a potential new class defect when the first score is below the first threshold or when the combined score exceeds the second threshold. As a non-limiting example, when the first score is below the first threshold, the potential defect is optionally indicated as an outlier. As another non-limiting example, when the second score exceeds the second threshold, the potential defect is optionally indicated as an excursion. As another non-limiting example, the combination of the first density estimation function and the second density estimation function is a ratio between a first score for a defect in the first density estimation function and a second score for a defect in the second density estimation function.

[0011] According to another aspect of the presently disclosed subject matter, a method for classifying a plurality of defects using a processor operably connected to a memory is provided, the method comprising: receiving, by the processor, a first plurality of first defects classified into a plurality of categories, and a second plurality of potential second defects, each first defect and each potential second defect being associated with a corresponding value of each of a plurality of attributes, the first plurality of first defects being characterized as a first attribute space, and the second plurality of potential second defects being characterized as a second attribute space; processing, by the processor, the first plurality of first defects and the second plurality of potential second defects to select a subset of attributes from the plurality of attributes, comprising: for each attribute: determining, by the processor, a first distribution of attribute values ​​of defects from the first plurality of first defects; determining, by the processor, a second distribution of attribute values ​​of defects from the second plurality of potential second defects; and determining, by the processor, a difference between the first distribution and the second distribution; selecting, by the processor, the subset of attributes as the attribute value in the The processor determines a first density estimation function based on the values ​​of the attribute subset assigned to the defects from the first plurality of first defects, and determines a second density estimation function based on the values ​​of the attribute subset assigned to the defects from the second plurality of potential second defects; the processor obtains a first threshold value associated with the first density estimation function and a second threshold value associated with the combination of the first density estimation function and the second density estimation function; the processor applies the first density estimation function to each defect of the second plurality of potential second defects to obtain a first score; the processor applies the second density estimation function to each defect of the second plurality of potential second defects to obtain a second score; the processor determines a combined score of the first score and the second score; and when the first score is below the first threshold or the combined score exceeds the second threshold, the defect from the second plurality of potential second defects is indicated as a potential new category defect. As a non-limiting example, when the first score is below the first threshold, the potential defect is optionally indicated as an outlier. As another non-limiting example, when the second score exceeds the second threshold, the potential defect is optionally indicated as an offset. As another non-limiting example, the combination of the first density estimation function and the second density estimation function is a ratio between a first score for a defect in the first density estimation function and a second score for a defect in the second density estimation function.

[0012] According to another aspect of the subject matter disclosed herein, a non-transitory computer-readable medium is provided, the non-transitory computer-readable medium comprising instructions, which when executed by a computer causes the computer to perform a method for automatically classifying defects in a sample into a plurality of categories, the method comprising: receiving, by a processor, a first plurality of first defects and a second plurality of potential second defects classified into a plurality of categories, each first defect and each potential second defect being associated with a corresponding value of each attribute in a plurality of attributes, the first plurality of first defects being characterized as a first attribute space, and the second plurality of potential second defects being characterized as a second attribute space; processing, by the processor, the first plurality of first defects and the second plurality of potential second defects to select an attribute subset from the plurality of attributes, comprising: for each attribute: determining, by the processor, a first distribution of attribute values ​​of defects from the first plurality of first defects; determining, by the processor, a second distribution of attribute values ​​of defects from the second plurality of potential second defects; and determining, by the processor, a difference between the first distribution and the second distribution; The processor selects the attribute subset as the attribute with the largest difference between the first distribution and the second distribution; the processor obtains a first density estimation function based on the values ​​of the attribute subset assigned to defects from the first plurality of first defects, and determines a second density estimation function based on the values ​​of the attribute subset assigned to defects from the second plurality of potential second defects; the processor obtains a first threshold associated with the first density estimation function and a second threshold associated with a combination of the first density estimation function and the second density estimation function; the processor applies the first density estimation function to each defect of the second plurality of potential second defects to obtain a first score; the processor applies the second density estimation function to each defect of the second plurality of potential second defects to obtain a second score; the processor determines a combined score of the first score and the second score; and when the first score is lower than the first threshold or the combined score exceeds the second threshold, indicates the defect from the second plurality of potential second defects as a potential new category defect. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to understand the invention and to see how it may be carried out in practice, embodiments will now be described, by way of non-limiting example only, with reference to the accompanying drawings, in which:

[0014] Figure 1 shows a generalized block diagram of an inspection system according to certain embodiments of the subject matter disclosed herein;

[0015] Figure 2 shows an exemplary two-dimensional case of a new class of defects represented in a test set; and

[0016] Figure 3A general flow chart illustrating a method of operating a new category identification engine according to certain implementations of the subject matter disclosed herein is shown. DETAILED DESCRIPTION

[0017] In the following detailed description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be appreciated by those skilled in the art that the subject matter disclosed herein may be practiced without these specific details. In other cases, well-known methods, processes, components, and circuits are not described in detail to avoid obscuring the subject matter disclosed herein.

[0018] Unless otherwise expressly stated, as will be apparent from the following discussion, it should be understood that in the discussion throughout the specification, terms such as "processing," "computing," "representing," "operating," "generating," "distributing," "selecting," "determining," "receiving," "training," "applying," "outputting," or the like are used to refer to the action(s) and / or processing(s) of a computer to manipulate and / or transform data into other data, the data being represented physically, such as electronically, quantitatively, and / or the data representing physical objects. The term "computer" should be broadly interpreted to cover any kind of hardware-based electronic device with data processing capabilities, including (as a non-limiting example) the classifier disclosed in this application and the PMC therein.

[0019] The terms "non-transitory memory" and "non-transitory storage medium" as used herein should be broadly interpreted to cover any volatile or non-volatile computer memory suitable for use with the presently disclosed subject matter.

[0020] It should be understood that, unless otherwise specifically stated, certain features of the currently disclosed subject matter described in the context of separate embodiments may also be provided in combination in a single embodiment. Conversely, various features of the currently disclosed subject matter described in the context of a single embodiment may also be provided individually or in any suitable sub-combination. In the following detailed description, many specific details are set forth in order to provide a thorough understanding of the method and apparatus.

[0021] The term "defect" in a sample as used in this specification should be interpreted broadly to cover any kind of abnormal or undesirable feature formed on or in the sample.

[0022] The term "known to belong to a class" as used in this specification should be broadly interpreted to cover any classification of defects into a class that is considered to be ground truth, such as classification by a human user, classification by a trusted classifier, or the like.

[0023] When inspecting an object, each defect may be characterized by values ​​assigned to a plurality of attributes, such as size, color, background color, or the like. The distance or proximity between defects may be defined based on an applicable distance function, such as a distance defined in an attribute space and possible values ​​for each attribute.

[0024] The term "classification" as used in this specification should be broadly interpreted to cover any assignment of defects to one of a plurality of predetermined categories (such as particles, scratches, processing or the like), such that each defect is automatically classified into one of the plurality of categories based on its attribute value, and items in the same category are generally adjacent to each other in the attribute space according to a defined distance function. These categories may be user defined or received in other ways.

[0025] Typically, a classifier is trained based on a collection of defects (called a "training set"), where each defect in the training set is classified into one of the defined categories, for example by a human user. Once the training is complete, the classifier can be used to classify further defects (called a "test set") into these categories.

[0026] It is common to try to obtain a training set that represents the expected defects as closely as possible, i.e., the distribution of defects of each class in the feature space is similar to the distribution in the future (multiple) test sets. However, this is not always possible, and some classes may be under-represented or not represented at all in the training set, which means that there are not enough defects of these classes for training. As a result, at runtime, defects of these classes will not be correctly classified. Such misclassification may have a negative impact on the quality of process control, such as statistical process control (SPC).

[0027] Therefore, given a training set and a test set, it is necessary to detect new categories of defects in the test set that have not been observed or are not sufficiently observed in the training set.

[0028] Such defects can be of two types: outlier defects, where a new class manifests itself as a statistical outlier, and excursion defects, where a new class appears within the boundaries of known classes and manifests itself as a significant change in the defect distribution.

[0029] Some known techniques for detecting outlier defects include local outlier factor (LOF), one-class support vector machines (SVM), and isolation forests. Known techniques for handling excursion detection typically rely on Kullback-Leibler (KL) divergence or similar metrics.

[0030] However, there is no known solution to detect both aspects: new class defects as well as shifted defects.

[0031] Therefore, under-representation of defects of certain groups (eg, minority classes) when training classifiers is a serious problem in manufacturing and inspection processes for a number of reasons.

[0032] First, misclassification may lead to an inaccurate assessment of the quality of the inspection process. Due to the low representation of a few defects, the impact of a few defects on the process assessment is small, so the quality of the entire process may be incorrectly assessed.

[0033] Second, the stability of the inspection process may be low: since smaller groups are more susceptible to variations when defects are misclassified, a few defects, or even a single defect, may significantly affect the standard deviation of the process evaluation. Therefore, it is difficult to evaluate the inspection process stably.

[0034] Third, pure misclassification can result in defects that may be too severe to be processed. In addition, significant effort is put into eliminating some very severe defects, resulting in fewer remaining defects of these types. Therefore, if such defects are not adequately represented in the training process, additional defects will not be classified into their class, which may cause severe problems in the sample to go undetected; and

[0035] Fourth, some types of defects, while referred to as belonging to the same class, may have different characteristics. For example, a defect of type "particle" may be significantly different in metal and silicone. By not identifying and training on specific minority classes, important information about specific defects may be lost.

[0036] The disclosed subject matter provides a method and apparatus for detecting new classes of defects. Using the present disclosure, a user can detect defects that may be new classes in a test set. The user can then examine the defects and classify them as belonging to one or more new classes. Additionally or alternatively, if a defect is detected with a high probability to be a new class, such defects can be clustered (in order to distinguish between multiple new classes) and automatically assigned to the new class.

[0037] Once a sufficient number of such defects are detected for one or more new classes (usually several dozen, e.g., at least 20-50 defects per class), a classifier can be trained with the defects so that other defects can be correctly classified.

[0038] With this in mind, please note Figure 1 , Figure 1 A general block diagram of an inspection system according to certain embodiments of the presently disclosed subject matter is shown. As part of sample fabrication, a Figure 1The inspection system 100 shown is used to inspect samples (e.g., wafers and / or portions thereof). The inspection can be part of the manufacture of an object and can be performed during or after the manufacture of the object. The inspection system may include various inspection tools, such as one or more inspection inspection tools 101, which are configured to capture inspection images (typically, at a relatively high speed and / or low resolution); and one or more review inspection tools 102, which are configured to capture review images of at least some defects detected by the inspection tool 101 (typically at a relatively low speed and / or high resolution). The inspection system 100 shown also includes a computer-based automatic defect classification tool (also referred to as a classifier hereinafter) 103, which is capable of automatically classifying defects into multiple categories based on the defect type. As non-limiting examples, classification can have different purposes, and the classification results can be used to identify specific defects of interest (DoI), filter false defects from real defects, build Pareto charts to identify deviations in statistical process control (SPC), and / or other aspects depending on the classification purpose.

[0039] The user can define the categories and the defects assigned to each category to train the classification engine 126 with the help of the GUI 108. In addition, the GUI 108 can be configured to enable the user (and / or the management system) to set the purity, accuracy, extraction requirements or other parameters for each category separately and optimize the classification results according to the user requirements.

[0040] The classifier 103 may be operably connected to one or more viewing tools 101 and / or one or more review tools 102. Optionally, the classifier 103 may be fully or partially integrated with one or more viewing tools 101 or with one or more review tools 102. The classifier 103 may further be operably connected to the design server 110 and the database 109.

[0041] The sample may be inspected by an inspection tool 101 (eg, an optical inspection system, a low resolution SEM, etc.) The acquired images and / or derivatives (optionally with other data such as design data and / or defect classification data) may be processed to select potential defects for review.

[0042] The subset of potential defect locations selected for review may be reviewed by a review tool 102 (e.g., a scanning electron microscope (SEM) or an atomic force microscope (AFM), etc.). For example, the subset may be selected based on categories associated with the potential defects output by the inspection tool 101 and possibly as part of additional considerations, where the categories may be determined by the classifier 103 based on the inspection data 128. For example, it may be necessary to review at least a predetermined number or percentage of defects in each category.

[0043] Additionally or alternatively, data providing information of the review image and / or its derivatives (hereinafter referred to as review data 122) and corresponding associated metadata may be sent directly or via one or more intermediate systems to the classifier 103. It should be noted that the review data may include data generated by one or more intermediate systems as derivatives of the review image.

[0044] The classifier 103 includes a processor and memory circuit (PMC) 104, which is operably connected to a hardware-based input interface 105 and / or a hardware-based output interface 106. The PMC 104 is configured to provide operations such as reference Figure 3 The processing required by the classifier is further described in detail, and includes a processor and memory (not shown separately in the PMC). The operation of the classifier 103 and the PMC 104 will be referred to in detail. Figures 2 to 3 Described in further detail.

[0045] The processing result of the classifier 103 outputted by the output interface 106 may be provided as the inspection instruction related data 123 or the review instruction related data 124 to the inspection tool 101 or the review tool 102 respectively, so as to change their operations, for example, by changing the values ​​of the operation parameters.

[0046] Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are not limited to Figure 1 Constraints of the system shown; equivalent and / or modified functionality may be combined or divided in another manner and implemented in any suitable combination of software and hardware with firmware.

[0047] It should be noted that Figure 1 The inspection system shown can be implemented in a distributed computing environment, where Figure 1The above-mentioned functional modules shown can be distributed on several local and / or remote devices and can be linked through a communication network. It should also be noted that in other embodiments, at least a portion of the inspection tools 101 and / or 102, the database 109, the storage system 107, the design server 110 and / or the GUI 108 can be outside the inspection system 100 and operate in a manner of data communication with the classifier 103 via the input interface 105 and the output interface 106. The classifier 103 can be implemented as (multiple) independent computers used in conjunction with one or more inspection tools. Optionally, the classifier 103 can operate on pre-acquired review data stored in the database 109 and / or the storage system 107. In addition or alternatively, the corresponding functions of the classifier 103 can be at least partially integrated with one or more inspection tools, process control tools, solution generation tools, systems for automatic defect review and / or classification, and / or other systems related to inspection.

[0048] PMC 104 may include new class identification engine 117, training engine 120, and one or more classification engines 121. New class identification engine 117 may be used to identify new class defects so that training engine 120 may train one or more classification engines 121 to classify future defects as new defects as well.

[0049] The new class identification engine 117 may include an attribute subset identification engine 118. The attribute subset identification engine 118 may be configured to identify a subset of attributes such that defects of unknown types during training are distinguished from defects associated with other classes by the values ​​assigned to these attributes.

[0050] The new category identification engine 117 may include a density function acquisition engine 119 for determining a first density function for the training set and a second density function for the test set. The density function is determined based on the attribute subset identified by the attribute subset identification engine 118.

[0051] The new category identification engine 117 may include a threshold deriving engine 120 for associating a first threshold associated with a first density function and a second threshold associated with a combination (e.g., a ratio) between the first density function and the second density function, such as a ratio. The new category identification engine 117 may include a density function application engine 126 for applying the first density function and the second density function to the defect to obtain a first score and a second score. If the first score exceeds the first threshold, the defect may potentially be an "outlier", i.e., have an attribute value that is different from the attribute values ​​commonly found in the training set. If the second score exceeds the second threshold, the defect may be a "shift", i.e., may indicate a different distribution between the defect density in the training set and the test set.

[0052] If the first score exceeds a first threshold, or the second score exceeds a second threshold, the defect may be output to a user, added to a set of potential new categories of defects, or utilized in any other manner.

[0053] Reference now Figure 2 , provides a visualization of examples of new classes of defects that manifest in the test set. For visualization purposes, Figure 2 The example of is two-dimensional, ie represents defects with two attributes, although defects in a typical system include significantly more attributes, eg, hundreds of attributes.

[0054] The training set includes two types of defects: a first type of defects 200 contained in a subspace 201 and a second type of defects 204 contained in a subspace 202 .

[0055] Similar to the training set, the test set includes defect 200 and defect 204. However, the test set also includes defect 208 contained in subspace 209, which are outlier defects because their subspace does not significantly overlap with the subspace of known classes (such as the class of defect 200 or 204).

[0056] The test set also includes defects 212 contained in subspace 213, which does have significant overlap with subspace 202 and can therefore be initially classified as belonging to the same category. However, the presence of these defects in the test set is identified by the change in the relative frequency of defects assumed to be defects 204 (i.e., defects 204 and defects 212 in a subspace that is substantially the same as subspace 202).

[0057] refer to Figure 3 , shows a generalized flow chart of operating the new category identification engine 117 according to certain implementations of the presently disclosed subject matter. The PMC 104 is configured to perform the respective operations detailed below according to computer-readable instructions implemented on a non-transitory computer-readable storage medium.

[0058] The PMC 104 receives a first plurality of first defects classified into a plurality of categories by a first classifier (step 300). The received defects have been pre-classified into predefined categories (e.g., by a human expert and / or another classifier and / or a version of a previous classifier, etc.), such that the pre-classification is considered ground truth. The PMC 104 also receives a second plurality of second defects (step 300), for which no ground truth is available. However, the second plurality of second defects may include defects of a type that is not present or insufficiently present in the first plurality of first defects.

[0059] Each first defect and each second defect can be described as a collection of values ​​assigned to various attributes, such as a vector. Typically, each defect can be described as values ​​assigned to hundreds or more attributes, such as its type, color, size, location, cause, or the like.

[0060] The attribute subset identification engine 117 may then identify a subset of attributes (step 304) that best distinguishes the first plurality of first defects from the second plurality of second defects. Thus, a plurality of attributes, which may but not necessarily include all attributes, are examined as distinguishing attributes. An attribute that better distinguishes between the two pluralities of defects is considered if the overlap between the range of values ​​for the first plurality of first defects and the range of values ​​for the second plurality of second defects is smaller. Thus, a first distribution of attribute values ​​for points of the first plurality of first defects may be determined, and a second distribution of attribute values ​​for points of the second plurality of second defects may be determined. The greater the difference between the first distribution and the second distribution, the better the attribute. In some embodiments, the lower the sum of the density scores for points in the second plurality of second defects, as given by the density function of the first plurality of first defects, the better the attribute, because this indicates that points in the second plurality of second defects are less likely to occur in the first plurality of first defects when using that attribute. A predetermined number of attributes may then be selected, wherein the best attribute may be selected as described above.

[0061] The density function obtaining engine 119 may obtain first and second density functions for a first plurality of first defects and a second plurality of second defects, respectively (step 308). Each such density function is configured to receive as input a defect described as a set of values ​​of attributes in an attribute subset, and output a score indicating whether the defect may be included in the corresponding plurality of defects. For example, a first defect having many neighbors (e.g., defects whose attribute values ​​are close to the attribute value of the first defect) will have a higher score than a defect having fewer such neighbors. The density function may be determined, retrieved from a storage device, received over a network, or the like.

[0062] The threshold obtaining engine 119 may obtain first and second thresholds of the first density function and the second density function, respectively (step 312).

[0063] The first threshold is associated with the first density function. A score below the first threshold is assigned to a defect indicating that the defect is likely not part of the training set and can therefore be considered an outlier.

[0064] The second threshold is related to the difference between the first density function and the second density function.A score above the second threshold assigned to a defect indicates that the defect is more likely to belong to the test set than to the training set and can therefore be considered a shift.

[0065] The density function application engine (121) can be configured to apply the density function to defects (e.g., defects from a test set) and obtain a first score and a second score, and a combination of the first score and the second score, such as a ratio between the first score and the second score (step 316).

[0066] If the first score of the defect is below a first threshold, then the defect may be suspected and indicated as an outlier (step 320). If the defect has a second score greater than a second threshold, then the defect may be suspected and indicated as an excursion (step 320). Both cases may be suspected as new categories and may therefore be indicated to the user.

[0067] PMC 104 may optionally collect defects that potentially belong to new categories over a period of time. The collected defects may be manually exported and verified as one or more new categories by a user (step 324). Alternatively, the defects may be clustered to identify one or more potential new categories.

[0068] Once a sufficient number of defects of each such new category are collected (e.g., at least 20-50 defects), the training engine 125 can be trained with multiple defects, including defects that are identified as indeed belonging to the new category. The resulting classification engine can then classify defects into the new category as well.

[0069] Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are not limited to Figure 3 Constraints of the presented approach. To obtain a small number of defects and train the classifier accordingly, other appropriate algorithms and their combinations can be implemented.

[0070] It should be understood that the application of the present invention is not limited to the details set forth in the description contained herein or shown in the accompanying drawings. The present invention is capable of other embodiments and can be practiced and implemented in various ways. Therefore, it should be understood that the wording and terminology employed herein are for descriptive purposes and should not be considered restrictive. Therefore, it should be appreciated by those skilled in the art that the concepts upon which the present disclosure is based can be easily used as the basis for designing other structures, methods and systems for achieving the several purposes of the subject matter disclosed herein.

[0071] It should also be understood that the system according to the present invention can be implemented at least in part on a suitably programmed computer. Likewise, the present invention contemplates a computer program readable by a computer to perform the method of the present invention. The present invention also contemplates a non-transitory computer readable memory to perform the method of the present invention, the non-transitory computer readable memory tangibly embodying an instruction program executable by a computer.

[0072] Those skilled in the art will readily appreciate that various modifications and changes may be applied to the embodiments of the invention described above without departing from the scope of protection of the invention as defined in and by the appended claims.

Claims

1. A system capable of classifying defects in a semiconductor sample into a plurality of categories, the system comprising a processing and memory circuit (PMC), the processing and memory circuit being configured to: receiving a first plurality of first defects and a second plurality of potential second defects categorized into a plurality of categories, each first defect and each potential second defect being associated with a respective value of each of a plurality of attributes, the first plurality being characterized as a first attribute space, and the second plurality being characterized as a second attribute space; Processing the first plurality and the second plurality to select a subset of attributes from the plurality of attributes comprises: For each attribute: determining a first distribution of property values ​​of defects from the first plurality; determining a second distribution of property values ​​of defects from the second plurality; and determining a difference between the first distribution and the second distribution; selecting the subset of attributes as attributes having the greatest difference between the first distribution and the second distribution; obtaining a first density estimation function based on values ​​assigned to the subset of attributes of defects from the first plurality; and determining a second density estimation function based on values ​​assigned to the subset of attributes of defects from the second plurality; Obtaining a first threshold associated with the first density estimation function and a second threshold associated with a combination of the first density estimation function and the second density estimation function; applying the first density estimate function to each defect of the second plurality to obtain a first score; applying the second density estimate function to each defect of the second plurality to obtain a second score; determining a combined score of the first score and the second score; and When the first score is below the first threshold or the combined score exceeds the second threshold, a defect from the second plurality is indicated as a potential new class defect. 2 . The system of claim 1 , wherein the potential second defect is indicated as an outlier when the first score is below the first threshold. 3 . The system of claim 1 , wherein the potential second defect is indicated as an excursion when the second score exceeds the second threshold.

4. The system of claim 1, wherein a combination of the first density estimation function and the second density estimation function is a ratio between a first score for a defect in the first density estimation function and a second score for a defect in the second density estimation function.

5. A method of classifying a plurality of defects in a semiconductor sample using a processor operably connected to a memory, the method comprising: receiving a first plurality of first defects and a second plurality of potential second defects categorized into a plurality of categories, each first defect and each potential second defect being associated with a respective value of each of a plurality of attributes, the first plurality of first defects being characterized as a first attribute space, and the second plurality of potential second defects being characterized as a second attribute space; The first and second pluralities are processed by the processor to select a subset of attributes from the plurality of attributes, the step of selecting the subset of attributes from the plurality of attributes comprising: For each attribute: determining, by the processor, a first distribution of attribute values ​​of defects from the first plurality; determining, by the processor, a second distribution of property values ​​of defects from the second plurality; and determining, by the processor, a difference between the first distribution and the second distribution; selecting, by the processor, the subset of attributes as attributes having a maximum difference between the first distribution and the second distribution; obtaining, by the processor, a first density estimation function based on values ​​assigned to the subset of attributes of defects from the first plurality; and determining a second density estimation function based on values ​​assigned to the subset of attributes of defects from the second plurality; Obtaining, by the processor, a first threshold associated with the first density estimation function and a second threshold associated with a combination of the first density estimation function and the second density estimation function; applying, by the processor, the first density estimate to each defect of the second plurality to obtain a first score; applying, by the processor, the second density estimation function to each defect of the second plurality to obtain a second score; determining, by the processor, a combined score of the first score and the second score; and A defect from the second plurality is indicated, by the processor, as a potential new class defect when the first score is below the first threshold or the combined score exceeds the second threshold. The method of claim 5 , wherein the potential second defect is indicated as an outlier when the first score is below the first threshold.

7. The method of claim 5, wherein the potential second defect is indicated as an excursion when the second score exceeds the second threshold.

8. The method of claim 5, wherein a combination of the first density estimation function and the second density estimation function is a ratio between a first score for a defect in the first density estimation function and a second score for a defect in the second density estimation function.

9. A non-transitory computer readable medium comprising instructions that, when executed by a computer, cause the computer to perform a method of automatically classifying defects in a semiconductor sample into a plurality of categories, the method comprising: receiving, by a processor, a first plurality of first defects and a second plurality of potential second defects categorized into a plurality of categories, each first defect and each potential second defect being associated with a respective value of each of a plurality of attributes, the first plurality of first defects being characterized as a first attribute space, and the second plurality of potential second defects being characterized as a second attribute space; The first and second pluralities are processed by the processor to select a subset of attributes from the plurality of attributes, the step of selecting the subset of attributes from the plurality of attributes comprising: For each attribute: determining, by the processor, a first distribution of attribute values ​​of defects from the first plurality; determining, by the processor, a second distribution of property values ​​of defects from the second plurality; and determining, by the processor, a difference between the first distribution and the second distribution; selecting, by the processor, the subset of attributes as attributes having a maximum difference between the first distribution and the second distribution; obtaining, by the processor, a first density estimation function based on values ​​assigned to the subset of attributes of defects from the first plurality; and determining a second density estimation function based on values ​​assigned to the subset of attributes of defects from the second plurality; Obtaining, by the processor, a first threshold associated with the first density estimation function and a second threshold associated with a combination of the first density estimation function and the second density estimation function; applying, by the processor, the first density estimation function to each defect of the second plurality to obtain a first score; applying, by the processor, the second density estimation function to each defect of the second plurality to obtain a second score; determining, by the processor, a combined score of the first score and the second score; and A defect from the second plurality is indicated, by the processor, as a potential new class defect when the first score is below the first threshold or the combined score exceeds the second threshold.

Images