Computer implemented method for the detection of defects in an imaging dataset of an object comprising integrated circuit patterns, computer-readable medium, computer program product and a system making use of such methods

Abstract

A first embodiment involves a computer implemented method for reducing the memory space demand of a process of detecting defects an object comprising integrated circuit patterns, the method comprising: (a) obtaining an imaging dataset of the object through an image acquisition method; (b) loading a reference dataset of the object from a database; (c) aligning the imaging dataset with the reference dataset; (d) applying a machine learning model capable of encoding and decoding a dataset to the imaging dataset to obtain an encoded imaging dataset; (e) storing the encoded imaging dataset on a data storage; (f) accessing the data storage to retrieve the encoded imaging dataset; (g) applying the machine learning model to the reference dataset to obtain an encoded reference dataset; and (h) applying a decoding operation of the machine learning model to a combined input comprising the encoded imaging dataset and the encoded reference dataset to obtain a reconstruction of the imaging dataset. A second embodiment involves a computer implemented method for detecting defects on an object comprising integrated circuit patterns, comprising steps (a) to (g) and additionally (h) applying a decoding operation of the machine learning model to a combined input comprising the encoded imaging dataset and the encoded reference dataset to obtain a reconstruction of the imaging dataset.

Description

[0001] Computer implemented method for the detection of defects in an imaging dataset of an object comprising integrated circuit patterns, computer-readable medium, computer program product and a system making use of such methods

[0002] Field of the invention

[0003] The invention relates to systems and methods for reducing the memory space demand of workflows for quality assurance of objects comprising integrated circuit patterns, more specifically to a computer implemented method, a computer-readable medium, a computer program product and a corresponding system for reducing memory space demand of a method of detecting defects in an imaging dataset of such an object. By using an autoencoder neural network for encoding and decoding an imaging dataset, the memory space demand of a defect detection method, in particular a die-to-database workflow, can be reduced. The method, computer-readable medium, computer program product and system can be utilized for quantitative metrology, process monitoring, defect detection and defect review in objects comprising integrated circuit patterns, e.g., in photolithography masks, reticles or wafers.

[0004] Background of the invention

[0005] A wafer made of a thin slice of silicon serves as the substrate for microelectronic devices containing semiconductor structures built in and upon the wafer. The semiconductor structures are constructed layer by layer using repeated processing steps that involve repeated chemical, mechanical, thermal, and optical processes. Dimensions, shapes and placements of the semiconductor structures and patterns are subject to several influences. One of the most crucial steps is the photolithography process.

[0006] Photolithography is a process used to produce patterns on the substrate. The patterns to be printed on the surface of the substrate are generated by computer-aided-design (CAD). From the design, for each layer a photolithography mask is generated, which contains a magnified image of the computer-generated pattern to be etched into the substrate. During the printing process an illuminated image projected from the photolithography mask is focused onto a photoresist thin film formed on the substrate. A semiconductor chip powering mobile phones or tablets comprises, for example, approximately between 80 and 120 patterned layers.

[0007] Due to the growing integration density in the semiconductor industry, photolithography masks have to image increasingly smaller structures onto wafers. The aspect ratio and the number of layers of integrated circuits constantly increases and the structures are growing into 3rd (vertical) dimension. The currentheight of the memory stacks is exceeding a dozen of microns. In contrast, the feature size is becoming smaller. The minimum design node can be below 10 nm, for example 7 nm or 5 nm, and is approaching feature sizes below 3 nm in near future, which corresponds to a feature size below 50 nm, such as below 20nm. While the complexity and dimensions of the semiconductor structures are growing into the 3rd dimension, the lateral dimensions of integrated semiconductor structures are becoming smaller. Producing the small structure dimensions imaged onto the wafer requires photolithographic masks or templates for nanoimprint photolithography with ever smaller structures or pattern elements. The production process of photolithographic masks and templates for nanoimprint photolithography is, therefore, becoming increasingly more complex and, as a result, more time-consuming and ultimately also more expensive. With the advent of EUV photolithography scanners, the nature of masks changed from transmission based to reflection-based patterning.

[0008] On account of the tiny structure sizes of the pattern elements of photolithographic masks or templates, it is not possible to exclude errors during mask or template production. The resulting defects can, for example, arise from degeneration of photolithography masks or particle contamination. Of the various defects occurring during semiconductor structure manufacturing, photolithography related defects make up nearly half of the number of defects. Hence, in semiconductor process control, photolithography mask inspection, review, and metrology play a crucial role to monitor systematic defects. Defects detected during quality assurance processes can be used for root cause analysis, for example, to modify or repair the photolithography mask. The defects can also serve as feedback to improve the process parameters of the manufacturing process, e.g., exposure time, focus variation, etc.

[0009] Chips are predominantly of two types: memory or logic. Therefore, for generating memory structures photolithography masks contain mainly simple, repetitive structures such as lines, spaces, contact holes, comers, etc., and for generating logic operations photolithography masks contain complex polygon structures, often without repetitions or re-occurring sub-patterns.

[0010] Photolithography mask inspection needs to be done at multiple points in time in order to improve the quality of the photolithography masks and to maximize their usage cycles. Once the photolithography mask is fabricated according to the requirements, an initial quality assessment of the photolithography mask is done at the mask house before it is shipped to the wafer fab. Semiconductor device design and photolithography mask manufacturing quality are verified by different procedures before the photolithography mask enters a semiconductor fabrication facility to begin production of integrated circuits. The semiconductor device design is checked by software simulation to verify that all features print correctly after photolithography in manufacturing. The photolithography mask is inspected for defects and measured to ensure that the features are within specification. The data gathered during this process becomes the golden baseline or reference for further inspections to be performed at the maskhouse or wafer fab. Any defect candidates found on the photolithography mask are validated using a review tool followed by a decision of sending the photolithography mask for repair or decommissioning the mask and ordering a new one.

[0011] At the wafer fab, the photolithography mask is scanned to find additional defects called “adders” compared to the last scan performed at the mask house. Each of these adders is analyzed using a review tool. In case of a particle defect, the particle is removed. In case of a pattern-based defect the photolithography mask is either repaired, if possible, or replaced by a new one. The inspection process is repeated after every few photolithography cycles. Each defect in the photolithography mask can lead to unwanted behavior of the produced wafer, or a wafer can be significantly damaged. Therefore, each defect must be found, classified and repaired if possible and necessary. Tighter pitch constraints imposed by the mask designer rules make the inspection of photolithography masks very difficult due to the limited physical resolution of the optical inspection tools with the defect size being close to the resolution limits, and due to the limited throughput of the charged particle beam systems. Reliable and fast defect detection methods are, therefore, important for photolithography masks.

[0012] Apart from defect detection in photolithography masks, defect detection in wafers is also crucial for quality management. During the manufacturing of wavers many defects apart from photolithography mask defects can occur, e.g., during etching or deposition. For example, bridge defects can indicate insufficient etching, line breaks can indicate excessive etching, consistently occurring defects can indicate a defective mask and missing structures hint at non-ideal material deposition etc. Therefore, a quality assurance process and a quality control process is important for ensuring high quality standards of the manufactured wafers.

[0013] Apart from quality assurance and quality control, defect detection in wafers is also important during process window qualification (PWQ). This process serves for defining windows for a number of process parameters mainly related to different focus and exposure conditions in order to prevent systematic defects. In each iteration a test wafer is manufactured based on a number of selected process parameters, e.g., exposure time, focus variation, etc., with different dies of the wafer being exposed to different manufacturing conditions. By detecting and analyzing the defects in the different dies based on a quality assurance process, the best manufacturing process parameters can be selected, and a window or range can be established for each process parameter from which the respective process parameter can be selected. In addition, a highly accurate quality control process and device for the metrology of semiconductor structures in wafers is required. The recognized defects can, thus, be used for monitoring the quality of wafers during production or for process window establishment. Reliable and fast defect detection methods are, therefore, important for objects comprising integrated circuit patterns.An object comprising integrated circuit patterns can refer, for example, to a photolithography mask, a reticle or a wafer. In a photolithography mask or reticle the integrated circuit patterns are mask structures used to generate semiconductor patterns in a wafer during the photolithography process. In a wafer the integrated circuit patterns are semiconductor structures, which are imprinted on the wafer during the photolithography process.

[0014] In order to analyze large amounts of data requiring large amounts of measurements to be taken, machine learning methods can be used. Machine learning is a field of artificial intelligence. Machine learning methods generally build a parametric machine learning model based on training data consisting of a large number of samples. After training, the method is able to generalize the knowledge gained from the training data to new previously unencountered samples, thereby making predictions for new data. There are many machine learning methods, e.g., linear regression, k-means, support vector machines, shallow neural networks or deep learning approaches.

[0015] Deep learning is a class of machine learning that uses artificial neural networks with numerous hidden layers between the input layer and the output layer. Due to this complex internal structure the networks are able to progressively extract higher- and lower-level features from the raw input data. Each level learns to transform its input data into a slightly more abstract and composite representation, thus deriving low- and high-level knowledge from the training data. The hidden layers can have differing sizes and tasks such as convolutional or pooling layers.

[0016] Methods for the automatic detection of defects in objects comprising integrated circuit patterns include anomaly detection algorithm , which are often based on a die-to-die, die-to-database or intra-die principle. The die-to-die principle compares images of portions of an object with reference images of the same portions of another identical object. An identical object can comprise, for example, another photolithography mask with an identical design of integrated circuit patterns, the same structure on a neighboring die of the same object, or the same photolithography mask having been imaged earlier. The discovered deviations are treated as defects. However, this method requires the availability and timeconsuming scanning of two corresponding portions of objects and exact knowledge about their relative position. In addition, it fails in case of repeater defects.

[0017] The die-to-database principle compares an image location of an object with a reference image from a database, e.g., a previously recorded image or a simulated image or a CAD file, thereby discovering deviations from the ideal data. Unexpected or irregular patterns in the imaging dataset are detected due to large differences.Summary of the invention

[0018] Embodiments of the invention concern computer implemented methods, computer readable media, computer program products and systems implementing defect detection methods for objects comprising integrated circuit patterns. Furthermore, embodiments of the invention concern a method of manufacturing a computer chip and a manufactured computer chip received from that method.

[0019] An embodiment of the invention involves a computer implemented method for reducing the memory space demand of a process of detecting defects an object comprising integrated circuit patterns, the method comprising: (a) obtaining an imaging dataset of the object through an image acquisition method; (b) loading a reference dataset of the object from a database; (c) aligning the imaging dataset with the reference dataset; (d) applying a first encoding operation of a machine learning model capable of encoding and decoding a dataset to the imaging dataset to obtain an encoded imaging dataset; (e) storing the encoded imaging dataset on a data storage; (f) accessing the data storage to retrieve the encoded imaging dataset; (g) applying a second encoding operation of the machine learning model to the reference dataset to obtain an encoded reference dataset; and (h) applying a decoding operation of the machine learning model to a combined input comprising the encoded imaging dataset and the encoded reference dataset to obtain a reconstruction of the imaging dataset.

[0020] An embodiment of the invention involves a computer implemented method for detecting defects on an object comprising integrated circuit patterns, the method comprising: (a) obtaining an imaging dataset of the object through an image acquisition method; (b) loading a reference dataset of the object from a database; (c) aligning the imaging dataset with the reference dataset; (d) applying a first encoding operation of a machine learning model capable of encoding and decoding a dataset to the imaging dataset to obtain an encoded imaging dataset; (e) storing the encoded imaging dataset on a data storage; (f) accessing the data storage to retrieve the encoded imaging dataset; (g) applying a second encoding operation of the machine learning model to the reference dataset to obtain an encoded reference dataset; and (h) applying a decoding operation of the machine learning model to a combined input comprising the encoded imaging dataset and the encoded reference dataset to obtain a reconstruction of the imaging dataset.

[0021] An embodiment of the invention involves a computer-readable medium, having stored thereon a computer program executable by a computing device, the computer program comprising code for executing a computer implemented method for reducing the memory space demand of a process of detecting defects an object comprising integrated circuit patterns as defined above and in further detail below.An embodiment of the invention involves a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out a computer implemented method for reducing the memory space demand of a process of detecting defects an object comprising integrated circuit patterns as defined above and in further detail below.

[0022] An embodiment of the invention involves a system for inspecting an object comprising integrated circuit patterns to detect defects, the system comprising: an imaging device adapted to provide an imaging dataset of an object comprising integrated circuit patterns; one or more processing devices; one or more machine-readable hardware storage devices comprising instructions that are executable by the one or more processing devices to perform operations comprising a computer implemented method for reducing the memory space demand of a process of detecting defects an object comprising integrated circuit patterns as defined above and in further detail below.

[0023] An embodiment of the invention involves a computer chip manufactures using an object comprising integrated circuit patterns what was checked for the absence of defects by a method as described within this specification.

[0024] Throughout this specification, the term “imaging dataset” can refer to images comprising the integrated circuit patterns of the whole object. It can also refer to images of only a subset of the integrated circuit patterns of the object, e.g., to a spatial subset, for example to an aera of interest of the object. The imaging dataset can refer to a single image, in particular to an area of interest of a single image. The imaging dataset can refer to two or multiple images, in particular to an area of interest within each of the images. For example, the imaging dataset can comprise several hundred or several thousand or several ten thousand images.

[0025] Throughout this specification, the term “integrated circuit patterns” comprises both patterns of existing integrated circuits, such as integrated circuits already being present on a wafer or reticle, as well as patterns for integrated circuits, which can be present, for example, on a photolithography mask or on a wafer.

[0026] Throughout this specification, the term “comprising” can define a device or a method which can implicitly contain or involve features or steps in addition to the features or steps explicitly mentioned. The term “consisting essentially of’ can define a device or a method which can implicitly contain or involve features or steps in addition to the features or steps explicitly mentioned, which do not materially affect the essential characteristics of the defined object or method. The term “consisting of’ can define a device or a method which cannot contain or involve any features or steps in addition to the features or steps explicitly mentioned. Within the context of the present specification, embodiments, examples or aspectsof the invention using the term “comprising” also disclose corresponding embodiments, examples or aspects of the invention using the terms “consisting of’ and / or “consisting essentially of’.

[0027] Throughout this specification, any dataset can be referred to as “not encoded” or “pre-encoded”, “encoded”, or “decoded”. A “decoded” dataset can also be referred to as the “reconstruction” of the (“not encoded”) dataset. The terms “not encoded” or “pre-encoded” in this context refer to datasets which have not previously been encoded during the computer implemented method according to the present invention. The term “encoded” in this context means that the dataset has been compressed by means of an encoding operation, for example using an autoencoder neural network capable of performing encoding and decoding operations and has a lower file size compared to a corresponding as “not encoded”, “preencoded” or “decoded” dataset. The terms “decoded” or “reconstruction” refer to a dataset which was previously in an “encoded” state and which has been restored by means of a decoding operation. The decoding operation can use one or more individual input signals. A “decoded” or “reconstructed” dataset is not necessarily equivalent or equal to a “not encoded” or “pre-encoded” dataset, as data identity relies on the type of compression and decompression methods used, which can be lossy or loss-less.

[0028] Furthermore, a “decoded” or “reconstructed” dataset can intentionally comprise a lower amount of irrelevant information, such as imaging information related to noise.

[0029] The imaging dataset can be acquired in different ways, e.g., through an image acquisition method or a combination of multiple image acquisition methods. Suitable image acquisition methods can comprise, for example, the application of a charged particle beam system such as a scanning electron microscope (SEM) or a focused ion beam (FIB) microscope, or an atomic force microscope (AFM), through an aerial image measurement system, e.g., equipped with a staring array sensor or a line-scanning sensor or a time-delayed integration (TDI) sensor, or by using a camera adapted for acquiring images at predetermined wavelengths. The camera can be an EUV camera, and / or a camera comprising a TDI sensor. In preferred embodiments, an image acquisition method comprises the use of an EUV camera comprising a TDI sensor. The camera’s image sensor can accordingly be an EUV image sensor, i.e., an image sensorthat is sensitive to EUV light. EUV light is light in the extreme ultraviolet spectral range with wavelengths between 5 nm and 100 nm, in particular with wavelengths between 5 nm and 30 nm. Especially the EUV light can have a wavelength of 13.5 nm. The EUV camera can be adapted for use in a photomask inspection system, wherein the photomask is projected onto an EUV image sensor of the EUV camera. In preferred embodiments, the image acquisition method comprises illuminating a photolithography mask with actinic radiation within the EUV wavelength range. EUV radiation reflected from the mask is then projected on an imaging sensor of an EUV camera via accordingly adapted projection optics. The image acquisition method can further comprise one or more image simulation methods as known in the art. Image simulation methods can in this context comprise computer implemented methods of calculating an image of the object based upon, for example, design data.A photolithography mask can have an aspect ratio of between 1:1 and 1:4, preferably between 1:1 and 1:2, most preferably of 1 : 1 or 1 :2. The photolithography mask can have a nearly rectangular shape. The photolithography mask can be preferably 12.7 cm (5 inch) to 17.8 cm (7 inch) long and wide, most preferably 15.2 cm (6 inch) long and wide. Alternatively, the photolithography mask may be 12.7 cm (5 inch) to 17.8 cm (7 inch) long and 25.4 cm (10 inch) to 35.6 cm (14 inch) wide, preferably 15.2 cm (6 inch) long and 30.5 cm (12 inch) wide.

[0030] The term “a number of’ can refer to one or more. The term “a subset of a dataset” can refer to a portion of the dataset or to the whole dataset. The terms “a” or “an” can throughout this disclosure refer to the singular or the plural. This means that, for example, when referring to “an autoencoder”, both one autoencoder and more than one autoencoders are disclosed.

[0031] As previously mentioned, the method comprises loading a reference dataset of the object from a database. The reference dataset can comprise a design dataset or an imaging dataset of an identical object having no defects. A design dataset can comprise a model design represented by a CAD file of a photolithography mask or reticle, or a CAD file of the wafer structures to be printed. Using these CAD files, an aerial image of the object can be simulated using computational methods known in the art. Alternatively, the reference dataset can comprise an imaging dataset of an object being known to not have any defects, which was previously obtained using an imaging method as described above. Such a dataset is also known as a golden reference.

[0032] Loading in this context means electronically obtaining a copy the reference dataset from the database and temporarily storing it on a memory storage, such as working memory, RAM, VRAM or cache. The physical location of the database is not limited. It can therefore be physically located in close proximity to the storage as mentioned before, such as in the same room or building, in another building of the same fab, or in a location outside the fab. The reference dataset can be transferred by wired and / or wireless connections. According to variants of embodiments of the present invention, the method can further comprise unloading the reference dataset after step (c) and loading the reference dataset again between steps (f) and (g). Accordingly, the copy of the reference dataset can be deleted from the memory storage when not required, such as between steps (c) and (g), in order to save memory storage space.

[0033] As defined above, step (c) of the computer implemented method according to the present invention involves aligning (or registration) of the imaging dataset with the reference dataset. Alignment is often necessary as the imaging dataset may, due to efficiency reasons of the imaging method, not be acquired in the same angle as the reference dataset. The step of aligning is therefore crucial in order to avoid for a step of computing for the misalignment. Such a step would require more computing resources andmemory space than step (c) as defined above. Furthermore, registration of both images ensures to obtain a correct reconstruction by decoding of the encoded imaging and reference datasets during step (h).

[0034] Alignment or registration can typically be carried out using one or more alignment algorithm(s) known in the art which recognize specific structural areas in the imaging data of the imaging and reference datasets and bring both images into alignment.

[0035] A machine learning model capable of encoding and decoding a dataset (or “machine learning model”) as used during steps (d), (g) and (h) can comprise a diffusion model, an autoencoder neural network, or a variational autoencoder. These machine learning models do not necessarily need to have a symmetrical structure. Therefore, these machine learning models do not necessarily need to comprise one encoder unit and one decoder unit which is symmetrical to the encoder unit. Instead, the machine learning model capable of encoding and decoding a dataset can comprise several (sub-)units, such as more than one, preferably two, encoder and / or decoder units, wherein each partial model being accordingly trained. The encoder unit(s) can carry out the first and second encoding operations, while the decoder unit can carry out the decoding operation as defined above. A preferred variant of embodiments of the present invention comprises the application of two separate encoder units and one decoder unit. In case more than one encoder or decoder sub-units are used, these sub-units can optionally be not identical. Not identical in this context means that the more than one encoder (sub-)units can have different sizes, numbers of layers, etc. Accordingly, these sub-units can have different layer sizes and / or may have been trained using different training data.

[0036] According to variants of embodiments of the present invention, the machine learning model capable of encoding and decoding a dataset can comprise an autoencoder neural network. In examples of these variants, steps (d) and (g) can comprise applying encoder layers of the autoencoder neural network, and wherein during step (h) the decoding operation can comprise applying decoder layers of the autoencoder neural network.

[0037] According to other examples, the encoded imaging dataset can comprise the activation of the bottleneck of the autoencoder neural network. In aspects of these examples, the encoded imaging dataset can consist of the activation of the bottleneck of the autoencoder neural network. This contributes to lowering the amount of data to be stored on storage media. The bottleneck is the lowest dimensional hidden layer of the autoencoder where the input data has undergone the most compression and only the relevant data required for the decoder unit to accurately reconstruct the original input is stored. The bottleneck layer can have a higher number of nodes, but a lower spatial resolution compared to other layers of the encoder and decoder units and the number of nodes in the bottleneck layer also gives the dimension of the encoding of the input. The activation of the bottleneck therefore refers to the output of the innermost hidden layer of the autoencoder. Therefore, the activation of the bottleneck can be understood as datarelated to the imaging or reference dataset, respectively, which has undergone the highest amount of compression during the computer implemented method as a whole.

[0038] The machine-learning model can be trained. According to an example of an embodiment of the invention, the autoencoder neural network can be trained comprising the following steps: generating training data from imaging datasets of objects comprising integrated circuit patterns and iterating the following steps: selecting one or more hyperparameter values for one or more hyperparameters of an machine learning model capable of encoding and decoding a dataset according to a sampling strategy; training the machine learning model based on subsets of the generated machine learning model training data and the selected one or more hyperparameter values; evaluating the trained machine learning model by computing an associated objective function value of an objective function. Finally, one of the trained machine learning models is selected based on the associated objective function value.

[0039] The machine learning model can be trained by optimizing one or more, loss fiinction(s). In an example of an embodiment of the invention, training can comprise minimizing a first loss function, wherein the first loss function can represent the reconstruction quality of patterns of the imaging data not contained in the reference dataset. The machine learning model is therefore trained to specifically reconstruct imaging data occurring from anomalous patterns, which improves the defect detection accuracy of defect detection methods using the reconstruction of the imaging dataset. In an example of an embodiment of the invention, training can comprise minimizing a second loss function representing a defect detection error of the defect detection method.

[0040] Any loss function as described in this relation to embodiments of the present invention can comprise an Lploss function, such as an Li or L2 loss function or a combination of both functions, a structured similarity index function, hierarchical extensions and / or adversarial loss terms. Preferably, the loss function comprises an L2 loss function.

[0041] According to a variant of embodiments of the invention, the machine learning model can be trained, wherein training of the machine learning model comprises optimizing a third loss function representing the reconstruction quality of the reconstruction of the imaging dataset. This additional approach of training can become beneficial for later defect detection workflows.

[0042] Additionally, during encoding, irrelevant data, such as noise, can be intentionally reduced or removed to lower the overall amount of data comprised in a dataset. Lowering or removal of noise can be facilitated by separate noise reduction algorithms or by the machine learning model, which can be trained to further reduce noise levels by not reconstructing data related to noise. Training data related to noise reduction can, for example, comprise a fourth loss function representing the noise reduction quality of the output ofthe machine learning model compared to the unencoded image, and / or by comparing pairs of previously prepared similar datasets having high and low noise levels.

[0043] According to variants of embodiments of the present invention, the machine learning model can comprise at least one encoding sequence transformation module and at least one decoding sequence transformation module. This enables the possibility of the encoded imaging and / or reference datasets having a variable code length. Therefore, instead of a fixed code length (and consequently file size), encoded datasets, which do not contain data related to any anomalies can have an even smaller file size. Therefore, incorporating at least one encoding sequence transformation module and at least one decoding sequence transformation module into the machine learning model can contribute to an additional reduction of storage space demand. The at least one encoding sequence transformation module and at least one decoding sequence transformation module could, for example, be implemented as a seq2seq model.

[0044] According to an embodiment of the invention, one or more measurements of the detected defects are determined, in particular size, area, dimension, shape parameters, distance, radius, aspect ratio, type, number of defects, location, density, spatial distribution of defects, existence of defects, etc. This allows for carrying out metrology tasks and for rule-based quality assessment of the object. The object comprising integrated circuit patterns can be a photolithography mask, a reticle or a wafer.

[0045] According to another embodiment of the invention, the invention further relates to a method for manufacturing a computer chip and a computer chip manufactured received from this method. The method involves the steps of (a) providing at least one object (1) comprising integrated circuit patterns, wherein the at least one object (1) comprises at least one photolithography mask, executing the computer-implemented method for detecting defects as defined in any one of claims above, selecting from the at least one object (1) any objects (5) meeting at least one predetermined quality feature, wherein the at least one predetermined quality feature comprises a maximum number of defects per object, and using at least one object (5) in a photolithography process to manufacture the computer chip.

[0046] Within the context of this embodiment, “at least one object” can preferably comprise a multiplicity of objects. The predetermined quality feature comprising a maximum number of defects per object can be set predetermined depending on quality demands of the computer chips to be produced using the object. Said number can be predetermined to be zero or greater than zero. For clarity, objects (1) found meeting the at least one predetermined quality feature as defined above are referred to as objects (5), while objects not meeting the at least one predetermined quality feature as defined above are referred to as objects (6). Objects (6) can then optionally be subjected to a repair method or be discarded. Methods for repairing defective objects comprising integrated circuit patterns are known in the art. In case an object (6) was successfully repaired to meet the at least one predetermined quality feature as defined above, andtherefore counting as an object (5), it can optionally be used in the manufacturing step as defined above. The predetermined quality feature can additionally be defined that only so-called printing defects are counted. Printing defects are defects on a photolithography mask which actually lead to defective integrated circuits on a wafer after being subjected to a photolithography method. Determining which defect is printing and which is not can be accomplished by expert inspection.

[0047] Optionally, defects can only comprise defects which would adversely affect the functionality of the computer chip. Such defects can, for example, comprise so-called “printing defects” on masks, namely defects which would actually produce defects on wafers when used in a photolithography process. The skilled person will acknowledge that the term “using” within the context of this embodiment can depend on the type of object (1) comprising integrated circuit patterns, i.e., if said object comprises, for example, a mask, a reticle, or a wafer. Therefore, in case said object comprises a wafer, a computer chip can be manufactured from a wafer that was checked for the absence of defects by a method as described above.

[0048] Systems and methods for reducing the memory space demand of workflows have been described above in the context of quality assurance of objects comprising integrated circuit patterns. However, a skilled person will acknowledge that the systems and methods described above could also be implemented to reduce the memory space demand of workflows related to contour detection or critical dimension repeatability.

[0049] Brief Description of the Drawings

[0050] In the following, advantageous exemplary embodiments of the invention are described and schematically shown in the figures. Throughout the figures and the description, same reference numbers are used to describe same features or components, unless stated otherwise.

[0051] Fig. 1 shows an imaging dataset of an object comprising integrated circuit patterns in the form of a photolithography mask comprising a defect;

[0052] Fig. 2 shows a flowchart illustrating computer implemented methods according to the present invention, including optional steps

[0053] Fig. 3 shows a diagram illustrating an embodiment of the present invention using encoded imaging and reference datasets as a combined input for the decoder layers of an autoencoder neural network to obtain a reconstruction of the imaging dataset;

[0054] Fig. 4 shows a flowchart illustrating a defect detection method according to an embodiment of the present invention comprising the use of a defect segmenter;

[0055] Fig. 5 shows a flowchart illustrating a system according to an embodiment of the present invention.

[0056] Fig. 6 shows a preferred variant of an imaging device used for acquiring imaging datasetsFig. 1 shows an exemplified imaging dataset 10 of an object 1 comprising integrated circuit patterns in the form of a photolithography mask comprising a defect 2. Defect detection methods known in the art compare such imaging datasets with reference datasets, which requires the imaging dataset 10 to be stored on a data storage. Due to the high number of pixels, each imaging dataset 10 can have a size of several tens or even hundreds of GBs, inter alia depending on the resolution of the imaging device used for acquiring the imaging dataset, the number of images per dataset, the size of the ROI, or the bit depth. Furthermore, due to the high throughput speeds of photolithography processes, the acquired data needs to be stored very fast, e.g., at a speed of several GBs per second. However, the throughput speed can become limited by the reading / writing speed of the actual storage device and / or the amount of data to be stored. Therefore, large amounts of high-speed storage space are required. An example of commonly used highspeed storage is flash-based data storage. In case of defect detection workflows this can be implemented in form of one or more interconnected compound(s) of several flash-based drives accessed via high-speed network connections. Writing and / or reading large amounts of data at a speed of several GBs per second on / from a data storage is known to be energy consuming. It is also known in the art that an increase in reading and writing speed comes with an increase in heat generation during operation, which can potentially lead to fab or sub-fab heating which is very concerning given the temperature has to be constant within less than one degree Celsius. Therefore, effective cooling is required to avoid a loss of reading / writing speed due to thermo-throttling. Furthermore, the lifetime of such storage devices is limited by a certain number of re-writing cycles, and therefore by the total amount of data written on the device. Therefore, it is an object of the present invention to increase the throughput of defect detection methods despite the before-mentioned speed limitations. It is another object of the present invention to reduce the storage space required during a defect detection process.

[0057] An object 1 comprising integrated circuit patterns can refer, for example, to a photolithography mask, a reticle or a wafer. In a photolithography mask or reticle, the integrated circuit patterns can refer to mask structures used to generate semiconductor patterns in a wafer during the photolithography process. In a wafer the integrated circuit patterns can refer to semiconductor structures, which are imprinted on the wafer during the photolithography process.

[0058] Fig. 2 is a flowchart illustrating the steps of computer implemented methods 100, 120 according to embodiments of the present invention. These methods comprise an image obtaining step 101, in which an imaging dataset 10 of an object 1 is obtained through an image acquisition method 300. The image acquisition method 300 can be selected from the group comprising time-delayed integration, x-ray imaging, scanning electron microscopy, focused ion beam microscopy, atomic force microscopy, and aerial imaging. Additionally, a reference dataset 11 is loaded from a database (not shown) during a first loading step 106. The loaded reference dataset can be stored in a memory storage (not shown). Thefollowing alignment step 102 can be implemented in two independent ways. Either the imaging dataset 10 is aligned with the reference dataset 11 or vice versa. In the first case, the imaging dataset 10 needs to be interpolated, in the latter case, float values according to the shift of the reference dataset 11 need to be saved additionally (both not shown in Fig. 2). The imaging dataset 10 is then encoded in a first encoding step 103 and stored on a data storage device in a storing step 104. It is then reobtained in a reobtaining step 105 to be used as a first input for the following decoding step 110. The reference dataset can be unloaded after step 102 in an optional unloading step 107 and later be reloaded in an optional second loading step 108. In a mandatory second encoding step 109 the reference dataset is encoded to provide a second input for the following decoding step 110. Step 109 can be carried out with or without prior carrying out of steps 107 and 108. In decoding step 110 a reconstruction 50 of the imaging dataset 10 is obtained using the decoding subunit of the machine learning model. In embodiments related to a method for detecting defects on an object comprising integrated circuit patterns, the computer implemented method 120 also involves subjecting the reconstruction of the imaging dataset to a defect detection method 200 to obtain defect detections by means of a defect criterion in a defect detection step 121.

[0059] Fig. 3 is a diagram illustrating the steps of a computer implemented method 100 according to embodiments of the present invention for reducing the memory space demand of a process of detecting defects 2 an object 1 comprising integrated circuit patterns, wherein the method 100 comprises: (a) obtaining an imaging dataset of the object through an image acquisition method 300; (b) loading a reference dataset 11 of the object from a database (not shown); (c) aligning 15 the imaging dataset 10 with the reference dataset 11 ; (d) applying a machine learning model capable of encoding and decoding a dataset 20 (in particular, an encoding operation 21) to the imaging dataset 10 to obtain an encoded imaging dataset 30; (e) storing the encoded imaging dataset 30 on a data storage 40; (f) accessing the data storage 40 to retrieve the encoded imaging dataset 30; (g) applying the machine learning model 20 (in particular, an encoding operation 22) to the reference dataset 11 to obtain an encoded reference dataset 31 ; and (h) applying a decoding operation 23 of the machine learning model 20 to a combined input comprising the encoded imaging dataset 30 and the encoded reference dataset 31 to obtain a reconstruction 50 of the imaging dataset 10.

[0060] Fig. 4 shows a diagram of an exemplified defect detection method 200 to obtain defect detections 70 by means of a defect criterion during step (h) of method 120 utilizing a defect segmenter 210. The computer implemented method 120 for the detection of defects as described above comprises the steps as illustrated in Fig. 3 followed by a defect detection method as exemplified in Fig. 4. The defect segmenter 210 receives as an input the anomaly dataset 60. This can, for example, be obtained by comparing the reconstruction 50 of the imaging dataset 10 with the reference dataset 11. In case the reconstruction 50 contained imaging data related to anomalous patterns, the resulting anomaly dataset 60 would contain data indicating at least one anomaly representing a possible defect 3 on the object 1. The defect segmenter210 then produces a defect probability map 211 indicating a defect candidate 4 and the location of said defect candidate 212 on the object 1 by means of a probability gradient 213.

[0061] The defect segmenter 210 can be implemented in form of a trained neural network. Training data can inter alia comprise expert annotations on previously detected defects or data regarding previously correctly identified defects by the same or similar neural networks. Such annotations can, for example, comprise bounding boxes, pixel-level annotations or point annotations. Once training has been completed, the learnable parameters of the defect segmenter can optionally be frozen. This means that the learnable parameters will not be adapted anymore. In certain examples of this embodiment of the present invention, the defect detection quality of the defect segmenter can also be used as learnable parameters in the form of a loss function for the machine learning model 20 used in previous steps as outlined above. This could, for example, be established by comparing the defect detections based on using imaging datasets showing integrated circuit patterns comprising known defects and imaging datasets showing corresponding integrated circuit patterns comprising no defects as inputs for the computer implemented method 120 as described above. The use of a defect segmenter as indicated in Fig. 4 for illustrative purpose and therefore not mandatory. Other possible defect detection methods 200 useful for application during the defect detection step, if applicable, are known in the art. Examples of possible other defect detection methods can comprise semantic segmentation, object detection, and / or combinations thereof.

[0062] The defect detection method 200 can alternatively comprise a method of identifying defects, which is not shown in the drawings. The defect identification method can comprise one or more preliminary defect identification methods to obtain defect detections by means of a defect criterion, in particular to distinguish defects from false positives. The preliminary defect identification method yields a preliminary defect detection result. To identify defect candidates, a model design of the object 1 is obtained. The model design can contain the ideal integrated circuit patterns of the object without defects. A model design can be a CAD file of a photolithography mask or reticle or a CAD file of the wafer structures to be printed. However, the model design is not necessarily equivalent or equal to the reference dataset 11 mentioned above.

[0063] Fig. 5 schematically illustrates a system 300, which can be used for inspecting an object 1 for defects 2. The system 300 includes an imaging device 310 and a processing device 320. The imaging device 310 is coupled to the processing device 320, e.g., via cable or wireless. They can be located in the same room, in the same lab, in the same fab or in different buildings. The imaging device 310 is configured to acquire imaging datasets 20 of the object 1. The object 1 comprises integrated circuit patterns, e.g., transistors such as field effect transistors, memory cells, et cetera. An example implementation of the imaging device 310 would be a SEM, a Helium ion microscope (HIM), a cross-beam device including FIB and SEM or any charged particle imaging device. In a preferred embodiment, time-delayed integration (TDI) is usedfor scanning a photolithography mask. For example, a photolithography mask of 10 to 12 cm can be split into 200 to 250 lines covering 85.000 to 110.000 pixels in width, which are scanned consecutively. In another example, an aerial image measurement system is used for obtaining the imaging dataset, e.g., equipped with a staring array sensor or a line-scanning sensor or a time-delayed integration (TDI) sensor. An aerial image is the radiation intensity distribution at substrate level.

[0064] The imaging device 310 can provide an imaging dataset 10 to the processing device 320. The processing device 320 includes a processor 321, e.g., implemented as a CPU or GPU. The processor 321 can receive the imaging dataset 10 via an interface 322. The processor 321 can load program code from a memory 323. The processor 321 can execute the program code. Upon executing the program code, the processor 321 can perform techniques such as described herein, e.g., applying an encoding operation of a machine learning model to an imaging dataset 10 and / or a reference dataset 11 to obtain encoded imaging 30 and / or reference 31 datasets, writing the encoded imaging dataset 30 on a data storage 40, applying a defect detection method 200, training the machine learning model 20, optimizing hyperparameters, etc. For example, the processor 321 can perform the computer implemented method shown in Figs. 2, 3 or, where applicable, 4, respectively, upon loading program code from the memory 323. The processing device 320 can optionally comprise a user interface 324 and / or a database 325.

[0065] Fig. 6 represents a preferred variant of the imaging device 310, in which the imaging device 310 is part of a photomask inspection system. A photomask 315 is arranged in such a way that an EUV beam 312 emitted from an EUV light source 311 is directed to the photomask 315 via an illumination system 313. The illumination system 313 is used to form the EUV radiation into a beam that is used to illuminate an examination field on the surface of the photomask 315 with uniform intensity. The examination field, which is small in relation to the surface of the photomask 315, is shown in a non-scale representation in Fig. 1. For example, the illuminated area may have dimensions of 0.5 mm x 0.8 mm. A field diaphragm is arranged in the illumination system 313 to limit the illuminated area to the examination field on the surface of the photomask 315. An X-Y positioner 316 can be used to move the photomask in the X-Y plane to bring various examination fields within the range of the EUV beam path.

[0066] The EUV beam 312 reflected from the photomask 315 continues through a projection optics 314 to an EUV camera 318 equipped with an image sensor 319. The image sensor 319 can comprise a TDI sensor. The projection optics 314 images the examination field of the photomask 315 onto the image sensor 319 of the EUV camera 318. The EUV light source 311, the illumination system 313, the photomask 315, the projection optics 314, and the EUV camera 318 are arranged in a vacuum housing 317 in which a negative pressure is applied during operation of the inspection system.Reference number list

[0067] 1 Object

[0068] 2 Defect

[0069] 3 Possible defect

[0070] 4 Defect candidate

[0071] 10 Imaging dataset

[0072] 11 Reference dataset

[0073] 20 Machine learning model

[0074] 21 First encoding operation

[0075] 22 Second encoding operation 23 Decoding operation

[0076] 30 Encoded imaging dataset

[0077] 31 Encoded reference dataset 40 Data storage

[0078] 50 Reconstruction

[0079] 60 Anomaly dataset

[0080] 70 Defect detection

[0081] 100 Computer implemented method 101 Image obtaining step

[0082] 102 Alignment step

[0083] 103 First encoding step

[0084] 104 Storing step

[0085] 105 Reobtaining step

[0086] 106 First loading step

[0087] 107 Unloading step

[0088] 108 Second loading step

[0089] 109 Second encoding step

[0090] 110 Decoding step

[0091] 120 Computer implemented method 121 Defect detection step

[0092] 200 Defect detection method

[0093] 210 Defect segmenter

[0094] 211 Defect probability map

[0095] 212 Defect candidate

[0096] 213 Probability gradient

[0097] 300 System

[0098] 310 Imaging deviceProcessing device Processor

[0099] Interface

[0100] Memory

[0101] User interface Database

[0102] Image acquisition method

Claims

Claims1. A computer implemented method (100) for reducing the memory space demand of a process of detecting defects (2) on an object (1) comprising integrated circuit patterns, the method comprising: (a) Obtaining an imaging dataset (10) of the object (1) through an image acquisition method (330);(b) Loading a reference dataset (11) of the object from a database (325);(c) Aligning the imaging dataset (10) with the reference dataset (11);(d) Applying a first encoding operation (21) of a machine learning model capable of encoding and decoding a dataset (20) to the imaging dataset (10) to obtain an encoded imaging dataset (30);(e) Storing the encoded imaging dataset (30) on a data storage (40);(f) Accessing the data storage (40) to retrieve the encoded imaging dataset (30);(g) Applying a second encoding operation (22) of the machine learning model (20) to the reference dataset (20) to obtain an encoded reference dataset (31); and(h) Applying a decoding operation (23) of the machine learning model to a combined input comprising the encoded imaging dataset (30) and the encoded reference dataset (31) to obtain a reconstruction (50) of the imaging dataset (10).

2. A computer implemented method (120) for detecting defects (2) on an object (1) comprising integrated circuit patterns, the method comprising:(a) Obtaining an imaging dataset (10) of the object (1) through an image acquisition method (330);(b) Loading a reference dataset (11) of the object from a database (325);(c) Aligning the imaging dataset (10) with the reference dataset (11);(d) Applying a first encoding operation (21) of a machine learning model capable of encoding and decoding a dataset (20) to the imaging dataset (10) to obtain an encoded imaging dataset (30);(e) Storing the encoded imaging dataset (30) on a data storage (40);(f) Accessing the data storage (40) to retrieve the encoded imaging dataset (30);(g) Applying a second encoding operation (22) of the machine learning model (20) to the reference dataset (11) to obtain an encoded reference dataset (31);(h) Applying a decoding operation (23) of the machine learning model (20) to a combined input comprising the encoded imaging dataset (30) and the encoded reference dataset (31) to obtain a reconstruction (50) of the imaging dataset (10); and(i) Subjecting the reconstruction (50) of the imaging dataset (10) to a defect detection method (200) to obtain defect detections (201) by means of a defect criterion.

3. The method according to any one of claims 1 or 2, wherein the machine learning model is trained by minimizing a first loss function, wherein the first loss function represents the reconstruction quality of patterns of the imaging data not contained in the reference dataset.

4. The method according to claim 3, as far as claim 3 depends on claim 2, wherein the machine learning model is trained by minimizing a second loss function, wherein the second loss function represents a detection error of the defect detection method (200).

5. The method according to any one of claims 2 to 4, wherein the first loss function comprises a loss function obtained by comparing a first defect detection result and a second defect detection result, wherein the first defect detection result is received by subjecting the imaging dataset (10) to the defect detection method (200), and wherein the second defect detection result is received by subjecting the reconstruction (50) of the imaging dataset (10) to the defect detection method (200).

6. The method according to any one of claims 2 to 4, wherein the first loss function comprises a loss function obtained by comparing a result of the defect detection method (200) to an imaging dataset comprising expert annotations.

7. The method according to any one of claims 2 to 5, further comprising determining one or more measurements of the detected defects (102), selected from dimension, shape parameters, distance, radius, aspect ratio, type, number of defects, location, density, spatial distribution of defects, existence of defects, and combinations thereof.

8. The method according to any one of the preceding claims, wherein the machine learning model capable of encoding and decoding a dataset comprises a diffusion model, an autoencoder neural network, or a variational autoencoder.

9. The method according to claim 7, wherein the machine learning model capable of encoding and decoding a dataset comprises an autoencoder neural network, and wherein steps (d) and (g) comprise applying encoder layers (31) of the autoencoder (30) neural network, and wherein during step (h) the decoding operation comprises applying decoder layers of the autoencoder neural network.

10. The method according to claim 8, wherein the encoded imaging dataset (21) comprises the activation of the bottleneck of the autoencoder neural network.

11. The method according to claim 9, wherein the encoded imaging dataset (21) consists of the activation of the bottleneck of the autoencoder neural network.

12. The method according to any one of the preceding claims, wherein the reference dataset comprises a design dataset, or an imaging dataset of an identical object having no defects.

13. The method according to any one of the preceding claims, wherein the reference dataset is unloaded after step (c) and loaded again between steps (f) and (g).

14. The method according to any one of the preceding claims, wherein the machine learning model is trained, and wherein training of the machine learning model comprises minimizing a third loss function, wherein the third loss function represents the reconstruction quality of the reconstruction of the imaging dataset.

15. The method according to any one of claims 3 to 13, wherein the loss function comprises an Lploss function, such as an Li or L2 loss function or a combination of both functions, a structured similarity index function, hierarchical extensions and / or adversarial loss terms.

16. The method according to claim 14, wherein the loss function comprises an L2 loss function.

17. The method according to any one of the preceding claims, wherein the machine learning model capable of encoding and decoding a dataset comprises an encoding sequence transformation module and a decoding sequence transformation module.

18. The method according to any one of the preceding claims, wherein the image acquisition method is selected from the group comprising time-delayed integration, x-ray imaging, scanning electron microscopy, focused ion beam microscopy, atomic force microscopy, aerial imaging, image simulation methods and combinations thereof.

19. The method according to any one of the preceding claims, wherein the object (1) comprising integrated circuit patterns is a photolithography mask, a reticle, or a wafer.

20. A computer-readable medium, having stored thereon a computer program executable by a computing device, the computer program comprising code for executing a method of any one of claims 1 to 18.

21. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out a method of any one of claims 1 to 18.2222. A system (300) for inspecting an object (1) comprising integrated circuit patterns to detect defects (2), the system comprising: an imaging device (310) adapted to provide an imaging dataset (20) of an object (1) comprising integrated circuit patterns; one or more processing devices (321); one or more machine-readable hardware storage devices (322) comprising instructions that are executable by the one or more processing devices (321) to perform operations comprising a method of any one of claims 1 to 18.

23. A method of manufacturing a computer chip, the method comprising the steps of:(a) Providing at least one object (1) comprising integrated circuit patterns, wherein the at least one object (1) comprises at least one photolithography mask;(b) Executing the method as defined in any one of claims 2 to 18;(c) Selecting from the at least one object (1) any objects (5) meeting at least one predetermined quality feature, wherein the at least one predetermined quality feature comprises a maximum number of defects per object;(d) Using at least one object (5) in a photolithography process to manufacture the computer chip.

24. A computer chip received from the method according to claim 23.

Images