Method for obtaining training data for training a model of a semiconductor manufacturing process

By generating a training data set of synthetic measurement data, optimizing the lithography process control strategy, the problem of low efficiency of lithography process monitoring and control in semiconductor manufacturing is solved, and the control accuracy and productivity are improved, and the cost is reduced.

CN113168108BActive Publication Date: 2024-11-19ASML NETHERLANDS BV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201980074980.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2018-11-14
Filing Date
2019-09-26
Publication Date
2024-11-19
Estimated Expiration
2039-09-26

AI Technical Summary

Technical Problem

During semiconductor manufacturing, prior art is difficult to effectively monitor and control lithography processes, resulting in the impact of the throughput, productivity and profitability of the manufacturing process, and the lack of tools to provide detailed process control options to optimize measurement and purchase costs. .

Method used

By generating a training data set of synthetic measurement data, used to train the model to optimize the lithography process control strategy, including obtaining behavioral characteristic data and generating synthetic measurement data from the data to simulate changes and impacts in the manufacturing process, and then selecting Best control strategy.

Benefits of technology

Improves control accuracy and efficiency of lithography processes, reduces measurement costs, and provides detailed process control options to optimize productivity and throughput, reducing defect risk.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113168108B_ABST
    Figure CN113168108B_ABST
Patent Text Reader

Abstract

A method for obtaining a training data set including synthetic measurement data is disclosed, the training data set being configured for use in training a model related to a manufacturing process for manufacturing an integrated circuit. The method comprises: obtaining behavioral characteristic data describing the behavior of process parameters resulting from the manufacturing process and / or associated tools or influences. Additionally or alternatively, measurement data performed on structures formed by the manufacturing process and / or similar manufacturing processes may be obtained. Using the behavioral characteristic data and / or the measurement data, synthetic measurement data are determined, the synthetic measurement data describing the effects of changes in the manufacturing process and / or associated tools or influences on the process parameters. The model is trained using the training data set including the synthetic measurement data.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of European patent application 18206285.1 filed on November 14, 2018, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present invention relates to methods and apparatus for applying a pattern to a substrate and / or measuring the pattern during a lithographic process. Background Art

[0004] A lithographic apparatus is a machine configured to apply a desired pattern to a substrate (usually to a target portion of the substrate). For example, a lithographic apparatus can be used in the manufacture of integrated circuits (ICs). In this example, a pattern forming device (which may alternatively be referred to as a mask or reticle) can be used to generate a circuit pattern to be formed on an individual layer of the IC. This pattern can be transferred to a target portion (e.g., a portion including one or more bare chips) on a substrate (e.g., a silicon wafer). The transfer of the pattern is usually performed via imaging on a layer of radiation-sensitive material (resist) disposed on the substrate. In general, a single substrate contains a network of adjacent target portions that are patterned continuously. Known lithographic apparatuses include so-called steppers, in which the target portion is irradiated by exposing the entire pattern to each target portion at once; and so-called scanners, in which the pattern is scanned by a radiation beam in a given direction ("scanning" direction), while the substrate is synchronously scanned in a direction parallel or antiparallel to the direction, to irradiate each target portion. It is also possible to transfer the pattern from the pattern forming device to the substrate by imprinting the pattern onto the substrate.

[0005] In order to monitor the lithographic process, parameters of the patterned substrate are measured. Parameters may include, for example, the overlay error between successive layers formed in or on the patterned substrate and the critical line width (CD) of the developed photoresist. The measurement may be performed on a production substrate and / or a dedicated measurement target. There are various techniques for measuring the microstructures formed during the lithographic process, including the use of scanning electron microscopes and various dedicated tools. A fast, non-invasive form of a dedicated inspection tool is a scatterometer, in which a radiation beam is directed onto a target on the substrate surface and the properties of the scattered or reflected beam are measured. Two main types of scatterometers are known. Spectral scatterometers direct a broadband radiation beam onto a substrate and measure the spectrum (intensity as a function of wavelength) of the radiation scattered into a specific narrow angular range. Angular resolved scatterometers use a monochromatic radiation beam and measure the intensity of the scattered radiation as a function of angle.

[0006] Examples of known scatterometers include angle-resolved scatterometers of the type described in US2006033921A1 and US2010201963A1. The target used by such a scatterometer is a relatively large (e.g., 40 μm×40 μm) grating, and the spot generated by the measurement beam is smaller than the grating (i.e., the grating is underfilled). In addition to measuring feature shapes by reconstruction, such a device can also be used to measure diffraction-based overlay, as described in the published patent application US2006066855A1. Diffraction-based overlay metrology using dark field imaging of diffraction orders enables overlay measurements of smaller targets. Examples of dark field imaging metrology can be found in international patent applications WO 2009 / 078708 and WO 2009 / 106279, the entire contents of which are hereby incorporated by reference. Other developments of the technology are described in published patent publications US20110027704A, US20110043791A, US2011102753A1, US20120044470A, US20120123581A, US20130258310A, US20130271740A and WO2013178422A1. These targets may be smaller than the illumination spot and may be surrounded by product structures on the wafer. Multiple gratings may be measured in one image using a composite grating target. The contents of all of these applications are also incorporated herein by reference.

[0007] When performing a semiconductor manufacturing process (such as applying a pattern on a substrate or measuring such a pattern), process control methods are used to monitor and control the process. Such process control techniques are usually performed based on a control strategy to obtain corrections to the semiconductor manufacturing process based on metrology efforts. By increasing the metrology effort, better corrections and controls can be achieved in principle (within limits). However, metrology takes time and therefore affects the throughput, productivity and profitability of the manufacturing process.

[0008] It would be desirable to provide more training data for training models used to make recommendations for manufacturing processes. Summary of the invention

[0009] In a first aspect of the present invention, a method for obtaining a training data set including synthetic measurement data is provided, wherein the training data set is configured for training a model related to a manufacturing process for manufacturing integrated circuits, the method comprising: obtaining behavioral characteristic data, the behavioral characteristic data describing the behavior of process parameters generated by the manufacturing process and / or related tools or influences; and determining the synthetic measurement data based on the behavioral characteristic data, the synthetic measurement data describing the influence of changes in the manufacturing process and / or the influence of related tools on the process parameters; and training the model using the training data set, the training data set including the synthetic measurement data.

[0010] In a second aspect of the invention there is provided a computer program comprising program instructions operable, when run on a suitable apparatus, to perform the method of the first aspect.

[0011] Other aspects, features and advantages of the present invention and the structure and operation of various embodiments of the present invention are described in more detail below with reference to the accompanying drawings. It should be noted that the present invention is not limited to the specific embodiments described herein. Such embodiments are proposed herein for illustrative purposes only. Based on the teachings contained herein, other embodiments will be apparent to those skilled in the relevant art. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Embodiments of the present invention will now be described by way of example with reference to the accompanying drawings, in which

[0013] Figure 1 depicts a lithographic apparatus and other apparatus forming a semiconductor device production facility;

[0014] Figure 2 is a flow chart describing an automated method for selecting an optimal control strategy; and

[0015] Figure 3 It is a description of generation for training. Figure 2 Flowchart of a method for synthesizing training data for a model. DETAILED DESCRIPTION

[0016] Before describing embodiments of the present invention in detail, it is instructive to present an example environment in which embodiments of the present invention may be implemented.

[0017] Figure 1 A lithographic apparatus LA is shown at 100 as part of an industrial production facility implementing a large-scale lithographic manufacturing process. In this example, the manufacturing process is suitable for manufacturing semiconductor products (integrated circuits) on substrates such as semiconductor wafers. The skilled person will appreciate that in variations of the process, a wide variety of products can be manufactured by processing different types of substrates. The production of semiconductor products is used purely as an example, which is of great commercial significance today.

[0018] In a lithographic apparatus (or simply "lithographic tool" 100), a measurement station MEA is shown at 102 and an exposure station EXP is shown at 104. A control unit LACU is shown at 106. In this example, each substrate visits a measurement station and an exposure station to have a pattern applied. For example, in an optical lithographic apparatus, a projection system is used to transfer a product pattern from a patterning device MA to a substrate using conditioned radiation and a projection system. This is done by forming an image of the pattern in a layer of radiation sensitive resist material.

[0019] The term "projection system" as used herein should be broadly interpreted to cover any type of projection system, including refractive, reflective, catadioptric, magnetic, electromagnetic and electrostatic optical systems, or any combination thereof, depending on the exposure radiation being used or other factors (such as the use of immersion liquid or the use of vacuum). The pattern forming MA device can be a mask or a reticle that imparts a pattern to a radiation beam transmitted or reflected by the pattern forming device. Well-known operating modes include stepping mode and scanning mode. It is well known that the projection system can cooperate with the support and positioning system for the substrate and the pattern forming device in a variety of ways to apply the desired pattern to many target portions across the substrate. A programmable pattern forming device can be used instead of a reticle with a fixed pattern. For example, the radiation can include electromagnetic radiation in the deep ultraviolet (DUV) or extreme ultraviolet (EUV) band. The present disclosure may also be applicable to other types of lithography processes, such as imprint lithography and direct write lithography, for example, by electron beam.

[0020] The lithographic apparatus control unit LACU controls all movements and measurements of the various actuators and sensors to receive the substrate W and mask MA and perform patterning operations. The LACU also includes signal processing and data processing capabilities to implement the desired calculations related to the operation of the apparatus. In practice, the control unit LACU will be implemented as a system with many sub-units, each of which handles real-time data acquisition, processing and control of a subsystem or component within the apparatus.

[0021] Before the pattern is applied to the substrate at the exposure station EXP, the substrate is processed at the measurement station MEA so that various preparatory steps can be performed. The preparatory steps may include: using a level sensor to map the surface height of the substrate; and using an alignment sensor to measure the position of alignment marks on the substrate. The alignment marks are nominally arranged in a regular grid pattern. However, due to the lack of precision when the marks are produced and due to the deformation of the substrate during its entire processing, the marks deviate from the ideal grid. Therefore, in addition to measuring the position and orientation of the substrate, if the device is to print product features at the correct position with very high accuracy, the alignment sensor must actually measure the position of many marks across the substrate area in detail. The device can be a so-called dual-stage type, which has two substrate stages, each substrate stage having a positioning system controlled by a control unit LACU. While one substrate on one substrate stage is being exposed at the exposure station EXP, another substrate can be loaded onto the other substrate stage at the measurement station MEA so that various preparatory steps can be performed. Therefore, measuring the alignment marks is very time-consuming, and providing two substrate stages enables the throughput of the device to be greatly improved. If the position sensor IF is not able to measure the position of the substrate table when located at the measurement station and the exposure station, a second position sensor may be provided to enable the position of the substrate table to be tracked at both stations. The lithographic apparatus LA may, for example, be of a so-called dual-stage type having two substrate tables and two stations (an exposure station and a measurement station) between which the substrate table can be exchanged.

[0022] Within a production facility, the apparatus 100 forms part of a "lithography cell" or "lithography cluster" which also contains a coating device 108 for applying photoresist and other coatings to substrates W for patterning by the apparatus 100. A baking device 110 and a developing device 112 are provided at the output side of the apparatus 100 for developing the exposed pattern into a physical resist pattern. Between all these devices, a substrate handling system is responsible for supporting the substrates and transferring them from one device to another. These devices, which are generally collectively referred to as tracks, are under the control of a track control unit, which itself is controlled by a monitoring system SCS, which also controls the lithography apparatus via the lithography apparatus control unit LACU. Thus, the different devices can be operated to maximize throughput and processing efficiency. The monitoring system SCS receives recipe information R which provides a definition of the steps performed to create each patterned substrate in great detail.

[0023] Once the pattern is applied and developed in the lithography unit, the patterned substrate 120 is transferred to other processing devices, such as those illustrated as 122, 124, and 126. In a typical manufacturing facility, many various processing steps are performed by various devices. By way of example only, the device 122 in this embodiment is an etching station, and the device 124 performs a post-etching annealing step. Other physical and / or chemical processing steps are applicable to other devices 126, etc. Making a real device may require various types of operations, such as depositing materials, modifying surface material features (oxidation, doping, ion implantation, etc.), chemical mechanical polishing (CMP), etc. In practice, the device 126 may represent a series of different processing steps performed in one or more devices. As another example, devices and processing steps may be provided to achieve self-aligned multiple patterning to produce multiple smaller features based on the precursor pattern laid by the lithography device.

[0024] As is well known, the manufacture of semiconductor devices includes many repetitions of such processing to build up device structures with appropriate materials and patterns layer by layer on the substrate. Thus, the substrates 130 arriving at the lithography cluster may be newly prepared substrates, or they may be substrates that have all been previously processed in the cluster or in another device. Likewise, depending on the desired processing, the substrates 132 on the way out of the device 126 may be returned to the same lithography cluster for subsequent patterning operations, they may be destined for patterning operations in a different cluster, or they may be finished products to be sent for dicing and packaging.

[0025] Each layer of the product structure requires a different set of process steps, and the type of apparatus 126 used at each layer can be completely different. Further, even in the case where the processing steps to be applied by the apparatus 126 are nominally the same, in a large facility there may be several supposedly identical machines working in parallel to perform steps 126 on different substrates. Small differences in the settings or failures between these machines may mean that they affect different substrates in different ways. Even steps that are relatively common to each layer, such as etching (apparatus 122), can be implemented by several etching apparatuses that are nominally the same but working in parallel to maximize throughput. Moreover, in practice, different layers require different etching processes (e.g., chemical etching, plasma etching), as well as special requirements, such as, for example, anisotropic etching, depending on the details of the material to be etched.

[0026] As just mentioned, the preceding process and / or the subsequent process may be performed in other lithography apparatuses, and may even be performed in different types of lithography apparatuses. For example, some layers that are very demanding on parameters such as resolution and overlay during device fabrication may be performed in more advanced lithography tools compared to other layers that are less demanding. Thus, some layers may be exposed in an immersion lithography tool, while other layers may be exposed in a 'dry' tool. Some layers may be exposed in a tool operating at a DUV wavelength, while other layers are exposed using EUV wavelength radiation.

[0027] In order to correctly and consistently expose the substrate exposed by the lithography apparatus, it is desirable to inspect the exposed substrate to measure characteristics such as overlay errors between subsequent layers, line thickness, critical dimensions (CD), etc. Thus, the manufacturing facility in which the lithography unit LC is located also includes a measurement system that receives some or all of the substrates W that have been processed in the lithography unit. The measurement results are provided directly or indirectly to the monitoring system SCS. If an error is detected, the exposure of subsequent substrates can be adjusted, especially when the measurement can be completed in a short time and quickly so that other substrates of the same batch are still to be exposed. In addition, the already exposed substrates can be stripped and reprocessed to increase yield, or discarded to avoid performing other processing on substrates that are known to be defective. In the case where only some target portions of the substrate are defective, further exposure can only be performed on those good target portions.

[0028] Figure 1 Also shown is a metrology device 140, which is provided for measuring parameters of the product at a desired stage in the manufacturing process. A common example of a metrology station in a modern lithography production facility is a scatterometer, such as a dark field scatterometer, an angle resolved scatterometer or a spectral scatterometer, and it can be used to measure the characteristics of the developed substrate 120 before etching in the device 122. For example, using the metrology device 140 it can be determined that important performance parameters such as overlay or critical dimension (CD) do not meet specified accuracy requirements in the developed resist. Prior to the etching step, there is an opportunity to strip the developed resist and reprocess the substrate 120 through the lithography cluster. The metrology results 142 from the device 140 can be used in the lithography cluster by the monitoring system SCS and / or the control unit LACU 106 to maintain accurate execution of the patterning operation, so that small adjustments can be made over time, thereby minimizing the risk of making a product that is out of specification and requires reprocessing.

[0029] Additionally, metrology devices 140 and / or other metrology devices (not shown) may be applied to measure characteristics of processed substrates 132, 134 and incoming substrate 130. Metrology devices may be used on processed substrates to determine important parameters such as overlay or CD.

[0030] During or before any actual imaging, the process parameters may have perturbations that cause them to deviate from specification (e.g., outside the process window (i.e., process parameter space) under which a compliant pattern would be produced), which may result in defects. For example, the focus may change due to the topography of the substrate to be exposed, drift in the substrate stage, deformation of the projection optics, etc.; the dose may change due to drift in source intensity, dwell time, etc. Various techniques can be used to identify the process parameter that is disturbed and to correct it. For example, if the focus is disturbed, for example because a region of the substrate is being exposed that is slightly higher than the rest of the substrate, the substrate stage can be moved or tilted to compensate for the perturbation.

[0031] Control of a manufacturing (e.g., lithography) process is typically based on feedback or feedforward measurements, which are then modeled using, for example, an inter-field (cross-substrate fingerprint) model and / or an intra-field (cross-field fingerprint) model. Depending on the circumstances, the modeling can also be extended to include an intra-die model (cross-die model). Within a die, there may be separate functional areas, such as memory areas, logic areas, contact areas, etc. Each different functional area or different functional area types may have a different process window, each process window having a different process window center. For example, different functional area types may have different heights and therefore different optimal focus settings. In addition, different functional area types may have different structural complexities and therefore different focus tolerances (focus process windows) around each optimal focus.

[0032] Other manufacturing processes may also be controlled in this manner using suitable models to implement corrections. For example, these corrections may include corrections to the etching process, corrections to the reticle, corrections to the rail system, corrections to the transient models associated with one or more devices used to perform the semiconductor manufacturing process.

[0033] Control of the lithographic apparatus can be achieved by modeling correction profiles (e.g., control profiles) for relevant parameters (or jointly optimized for more than one parameter). The modeled correction profile for each parameter is fed into the lithographic apparatus, which actuates the desired correction profile to control the lithographic process (exposure). The control can be based on a feed-forward model (e.g., based on data measured within the lithographic apparatus before exposure). The scanner itself has self-corrections that need to be actuated by the scanner during exposure. These self-corrections include, for example, feed-forward models such as reticle heating and wafer heating, machine calibrations such as wafer stage shape and layout related corrections.

[0034] Focus control is an example of a primary feedforward control loop based on a large amount of leveling data collected for each substrate, which is used to determine corrections to the exposure on that substrate, which corrections correct for the surface topography. Other corrections are based on feedback control loops. In addition to the primary feedforward control just mentioned, focus control can also have a feedback element based on measurements of focus from the exposed structure. Overlay control is typically based on a feedback loop; based on measurements of overlay from processed substrates. Dose control has no feedforward control other than the average dose, and is typically controlled in a feedback loop from post-exposure (e.g., post-etch) measurements via a correction profile determined per field (e.g., in the scan direction and the slit direction, respectively).

[0035] All of these correction sources are input to a lithographic apparatus that combines all of the corrections for each exposure and actuates them to optimize overlay, focus, dose, and imaging performance. There are several methods for the lithographic process to actuate correction profiles, for example, for controlling focus / dose and / or overlay. Algorithms, which essentially behave as filters, transform the corrections into set points for the stages and lenses / mirrors. The set points are defined as time-dependent trajectories that, for example, define the tilt of the reticle stage and / or wafer stage relative to each other during exposure. By moving accordingly, the actuators image the fingerprint onto the substrate. Such methods and others are obvious to those skilled in the art and will not be discussed further.

[0036] All of the control / correction methods described above have associated costs, specifically in terms of measurement effort. The degree of control or correction of a typical process is a balance between the required quality and the measurement effort. Generally, improving quality requires better correction, which in turn requires increased measurement effort. However, in order for a manufacturing process to be commercially acceptable, certain productivity levels or throughput targets need to be met, and increasing measurement effort will tend to reduce productivity / throughput. Currently, there are no tools that provide users with a detailed understanding of the available options for process control (e.g., in terms of yield, overlay or other parameters). Specifically, there is currently no tool to identify which control option is most appropriate from several possible control options based on the expected improvement associated with the control option, and taking into account aspects such as measurement cost, purchase cost, etc. Guidance for determining the control strategy that best suits a specific process requirement is an important step during the establishment of a semiconductor manufacturing process.

[0037] As such, it is proposed to evaluate different process control strategies and / or identify a preferred process control strategy based on:

[0038] Process data: This may include historical data and / or design data that characterizes a semiconductor manufacturing process (typically used to apply a series of layers to a substrate such as a wafer). Examples of process data may include one or more of the following:

[0039] - Layout data (eg, design data or mask data) that describes the layout of a particular layer. This may include the layout within the die (eg, the location and size of different functional areas within the die).

[0040] - Process specification metrics, such as process windows, which define minimum quality standards, such as maximum / minimum allowable values ​​(specification limits) of any process-related parameter (e.g., focus / dose / overlay / edge placement / throughput); process windows may be defined per field, die, substrate region, and / or functional area, as appropriate. Thus, critical structures or functional areas within a die may have more stringent process windows associated with them than other structures. Process specification metrics may also include target settings for any parameter (e.g., optimal focus setting). Likewise, these metrics may be set per any of the described regions / functional areas.

[0041] - Process context, such as: which available lithography apparatuses, etching apparatuses, deposition apparatuses and chambers of the mentioned apparatuses are used, and / or any settings of these apparatuses, control option settings, sensor readings, product definitions, context granularity.

[0042] The candidate process control strategies to be evaluated and any associated parameter settings required for their implementation; these can be entered or determined from the process data.

[0043] Quality metric prediction data. For example, this may include knowledge of characteristics of the candidate control strategy and / or any other control strategy suitable for predicting quality metric data (e.g., a reference strategy), such as how the control strategy affects performance when applied to the process represented by the process data. This may be based on:

[0044] o Historical quality metric data relevant to the process (e.g., based on measurements or previous production determinations), and / or

[0045] oModeled / simulated quality metric data related to the process.

[0046] Relevant cost metric data for each candidate control strategy and / or any other control strategy (e.g., reference strategy). Example cost metrics include:

[0047] - Required metrology effort. This can include measures of the density and / or frequency of measurements that should be performed to achieve a certain control strategy; for example, the sampling scheme used. The required metrology type can also be defined. The required metrology type can refer to measurement techniques or measurement parameters. Parameters can include one or more of the following: overlay metrology, focus metrology, CD (critical dimension) / SWA (sidewall angle) metrology. Metrology types may include:

[0048] oScattering-based measurements, such as:

[0049] Overlay / focusing technology based on dark field diffraction (asymmetric technology),

[0050] reconstruction-based techniques (e.g. from pupil images),

[0051] o Scanning electron microscopy

[0052] oInline vs. offline measurement (e.g., the need for inline measurement may be attributed to higher costs)

[0053] o Lithography device (scanner) measurement:

[0054] Alignment measurement, and / or

[0055] Leveling measurement

[0056] - Other indirect expenses required to implement each candidate control strategy; for example, the required hardware and software and any related costs associated therewith (e.g., software licensing costs, hardware costs).

[0057] Candidate process control strategies may be related to the setting and / or correction of any parameters related to the manufacturing process. These parameters may include (variable) parameters of any device involved (directly or indirectly) in the manufacturing process, such as lithography device settings (e.g., focus settings, dose settings, stage positioning set points), etching device settings, deposition device settings, tracking device settings, reticle manufacturing and / or reticle lithography settings, etc. Candidate process control strategies may also be related to characteristics of the control interface.

[0058] Candidate process control strategies can be related to the models used, such as the type of modeling strategy, for example, the order (model degree) of any modeling strategy (e.g., first order / higher order), whether the modeling is by stack, within-stack, between-field, within-field, between-die, within-die, and any other relevant model details used to simulate any manufacturing process in the manufacturing process (e.g., for determining corrections).

[0059] Figure 2is a flow chart describing a purely exemplary automated method for selecting an optimal control strategy. The method is described in more detail in European patent application EP18197856.0, which is incorporated herein by reference. Based on an evaluation of simulated product performance (associated with a certain input data set) and required measurement effort (e.g., throughput impact), the method selects and customizes available control strategies and evaluates the quality of each control strategy. This is done using artificial intelligence (AI) techniques such as deep learning techniques (e.g., using neural networks) or Bayesian network techniques. As a result, the control strategy selection method described above can be made more practical (especially for evaluating a large number of control strategies) and user-independent (e.g., user-specific recommendation results).

[0060] In EP18197856.0, the aforementioned process data, cost metric data, and quality metric data are used to train an AI network (e.g., a neural network or a Bayesian network) to associate the process data with the quality metric data taking into account the cost metric data. The quality metric data may be simulated or measured, and the quality metric data may be related to (e.g., including values ​​of) any one or more suitable quality metrics (e.g., overlay, edge placement error (EPE), critical dimension (CD), CD uniformity (CDU), or yield).

[0061] refer to Figure 2 , training data 200 (which may be a subset of validation data 240) is fed into a neural network 205 (or other suitable AI network). The training data includes process data (which may include any of the examples already provided) and associated quality metric data. For example, the training data may include control strategies and result quality metric values ​​associated with each control strategy. The training data may also include cost metric data, such as the associated cost of each control strategy. However, this is not essential, and the cost metric may be considered by simulation (e.g., error function 215), for example, as a boundary condition or constraint. The neural network 205 may start with a "blank" untrained network, or may otherwise have been trained in a calibration phase based on other training data (e.g., simulation data). Over time, the neural network 205 will learn which control strategies are better candidates for evaluation and are more likely to result in acceptable performance (based on the quality metric) and acceptable costs (based on the cost metric).

[0062] The simulation phase begins by correlating process data with quality metric data. Such methods may be based on modeling of a Bayesian network or a machine learning method (e.g., deep learning) that correlates process data with quality metric data. The simulation phase may begin with the neural network 205 outputting a candidate control strategy 210. The simulation step 220 includes evaluating an error function 215 based on the candidate control strategy 210. The error function 215 may model the error (residual) of the quality metric data based on the candidate control strategy 210. The error may, for example, be the average (simulated) quality metric value of the training data. As already mentioned, the cost metric data may be used to perform boundary condition adjustments on the error function 215. At step 225, the error is evaluated (e.g., compared to a threshold value; for example, the number of non-yielding dies or the error is less than or equal to 0) to determine whether the candidate control strategy 210 is acceptable. If the candidate control strategy 210 is not acceptable, this is fed back to the neural network 205, and the candidate control strategy 210 is further evaluated. When the error is assessed to be acceptable, the optimized control strategy 235 is output.

[0063] The control strategy optimization method may include a validation step 245 using a larger data set (i.e., validation data 240) to validate the optimized control strategy 235 against the validation data. The validation step 245 may simply repeat the simulation step 220 (i.e., evaluating the error function 215) of the optimized control strategy 235 using a large amount of validation data 240. Optionally, the validation step may also output a predicted quality metric value 250 (e.g., predicted overlay or EPE) based on the optimized control strategy 235.

[0064] Figure 2 One problem with the described approach is that, in order to properly train the AI ​​network, a large training data set is required. Sufficient training data can be difficult to obtain for various reasons (e.g., the measurement effort required and / or confidentiality issues). In particular, the training data sets available for training the model are often insufficiently representative of actual processing factors, such as offsets, drifts, fingerprints, and / or noise of one or more processing parameters.

[0065] Thus, a method for modifying a training data set configured for model training is now described. The method may include: obtaining a first data set including contextual and / or metric data associated with a semiconductor manufacturing process; and modifying the first data set by introducing variability based on characteristics of the semiconductor manufacturing process to obtain the training data set. Alternatively, the first data set is not used, and the training data includes fully synthetic data based on characteristics of the semiconductor manufacturing process. In this way, a training data set is obtained, which is a synthetic data set or a hybrid (semi-synthetic) data set.

[0066] This approach can be used to generate enough data needed to properly train a system from a smaller data set. Synthetic data can be generated from initial measurement data based on certain behavioral characteristics that are known and / or expected to occur. The training data set can then be a hybrid data set that includes both the synthetic data and the initial measurement data from which the synthetic data was generated. It is also within the scope of the present disclosure to generate fully synthetic data (i.e., generated only from known behavioral characteristics) without having any actual measurement data available.

[0067] The behavioral characteristics can be based on engineering knowledge (known to be possible and appropriate responses to known known situations) and existing customer data sets. In this way, the system is able to make preliminary control recommendations based on the limited available data sets. The behavioral characteristics can include, for example, one or more of the following:

[0068] Spatial fingerprints (including inter-field fingerprints, intra-field fingerprints, and / or spatial parameter distributions of other fingerprints), which can be imposed by any tool or influence (e.g., fingerprints generated by processing, clamping, lens aberrations, etching, deposition, focus / dose, reticle heating or other heating influences, etc.);

[0069] Temporal noise / drift of any parameter;

[0070] Context or context granularity, e.g., the etch chamber, lithography stage, and / or deposition station or other processing history to which the substrate has been subjected, and possible combinations of these;

[0071] Relationships between different types of data and / or similar processes;

[0072] Predicted events to which the system should respond (e.g., noise events and / or shifts in metric data), such as by proactively implementing a response change or deciding to ignore the event based on desired training results;

[0073] • Boundary conditions, such as performance specifications, measurement budget, available control options, and any control settings, however, any boundary conditions may alternatively be imposed by the simulation (eg, error function 215).

[0074] Using one or both of the known behavioral characteristics and any actual measurement results, synthetic data can be generated. The synthetic data can be, for example, based on extrapolation and / or interpolation of the behavioral characteristics over time (e.g., with additional substrates or stacks). For example, the synthetic data can describe how a parameter changes over time during an unmeasured period of time (e.g., for a modeled substrate or stack). By way of specific example, if measurement data for a parameter is only available for a few stacks (or no stacks are available at all), but the behavior of the parameter is known, for example, as a function of time, number of substrates, or number of stacks, then the parameter value can be interpolated for stacks between the few stacks and / or extrapolated for more stacks (e.g., by regression or similar modeling techniques) (e.g., extrapolated much further to additional stacks between the measured stacks and / or future stacks). It is also possible to perform regression based only on the available measurement data, where its behavioral characteristics are inferred therefrom. In this way, for example, synthetic measurement data describing one or more parameters on a large number of stacks can be generated based on real measurement data from a small number of stacks (or no stacks at all).

[0075] Other methods of generating synthetic data are possible, for example, by generating one type of data from another type of data. For example, if the relationship between the measured parameters after development (before etching) and after etching is known, then the post-etch parameter values ​​can be generated from the post-development parameter values, and vice versa. In addition, if the effects of noise, any other effects (e.g., heating or lens aberrations), or any interfering events are known, the effect of changes in any of these effects / events on the parameters can be inferred / estimated in order to generate synthetic data. Thus, for example, synthetic metrology data can be generated for different noise levels, as well as synthetic metrology data describing parameters that are subject to different levels of reticle heating or different lens aberration effects relative to the actual measurements performed.

[0076] Additional data can also be generated spatially based on sparse measurements. For example, where metrology data for only a few points on a substrate is available, additional metrology data for other locations on the substrate can be calculated.

[0077] Different scenarios may also be considered. For example, where the process fingerprints for each lithography stage, etch chamber, and / or any other processing station are known, then a combined fingerprint (and its impact) for any combination of these fingerprints may be generated based on measurements of one or a few stacks related to a much smaller subset of possible combinations, or not based on actual measurements at all.

[0078] Another way to generate synthetic measurement data can be based on characteristics (fingerprints / drifts) associated with other applications that are different from the application of interest, but due to commonalities between the application of interest and the other applications, commonalities in characteristic behavior can be assumed. For example, synthetic data for a particular process or layer can be derived from metric data for another process or layer due to known commonalities between the processes or layers (e.g., shared processing stations / contexts).

[0079] Synthetic measurement data introduces variability / behavior that can be used to train a model (eg, a neural network) faster, such that the model is trained in advance for actual occurring process variations that are representative of the process of interest.

[0080] Once an AI system is trained and in use, it can continue to learn while receiving real-world measurement data (e.g., data generated during production). It will then learn from other specific behaviors that may not have been anticipated in the training data. This additional measurement data will enable the system to better fine-tune its output when needed. This can be advantageous because synthetic data may not cover all behaviors that occur under actual production conditions.

[0081] Figure 3 300 and (optionally) measurement data 305 are used to generate synthetic data 310 to obtain training data 315 (training data 315 may also include actual measurement data 300). The generation step 310 may use any of the methods for generating synthetic training data disclosed herein. The model training step 320 may use the training data 315 to train a model such as a neural network to obtain a trained model 325.

[0082] At step 330, the trained model is used to provide control recommendations 335, as already described. This may include a recommended control strategy (or several ranked strategies) for a specific use case (e.g., a specific product and layer combination). This may include: making recommendations on which control components (e.g., specific software and / or hardware tools) and / or settings for each control component may be preferred. This step may also include: predicting the results of the lithography process and making recommendations therefor. At step 340, feedback data 345 may be collected. The feedback data may be specific to the control use case (e.g., based on a specific product and layer combination) and include measurement information from the substrate that is subject to the control strategy determined by the trained model and the corresponding control strategy selected. The feedback data 345 may include any other relevant information (e.g., about and / or describing the structure and / or stack of the process being controlled / recommended). The feedback data 345 may be fed back to the model 325, which may evaluate and learn to improve itself (e.g., determine the effectiveness of a specific control strategy based on corresponding measurements). Other embodiments of the present invention are disclosed in the following numbered list of clauses:

[0083] 1. A method for obtaining a training data set comprising synthetic metrology data, the training data set being configured for use in training a model related to a manufacturing process for manufacturing an integrated circuit, the method comprising:

[0084] obtaining behavioral characteristic data describing the behavior of process parameters resulting from the manufacturing process and / or associated tools or effects, and / or obtaining metrology data performed on structures formed by the manufacturing process and / or similar manufacturing processes;

[0085] Determining the synthetic measurement data based on the behavioral characteristic data and / or the measurement data, the synthetic measurement data describing the influence of changes in the manufacturing process and / or related tools or influences on the process parameters; and

[0086] The model is trained using the training data set, wherein the training data set includes the synthetic measurement data.

[0087] 2. A method according to clause 1, wherein the model is based on a neural network.

[0088] 3. A method according to claim 1 or 2, wherein the behavioral characteristic data includes data related to one or more of the following: spatial parameter distribution imposed by any tool or influence, temporal noise / drift of any parameter, context or context granularity, relationships between different data types and / or similar processes, and predicted events to which the system should respond.

[0089] 4. A method according to any preceding clause, wherein the synthetic measurement data is determined from both the behavioural characteristic data and the measurement data.

[0090] 5. The method of clause 4, wherein the training data set comprises both the synthetic measurement data and the measurement data.

[0091] 6. A method according to any preceding clause, wherein determining the synthetic measurement data comprises: interpolating and / or extrapolating the behavioral characteristics over time to obtain an extrapolated process parameter value.

[0092] 7. A method according to any preceding clause, wherein determining the composite measurement data comprises: converting measurement data of one type into measurement data of another type.

[0093] 8. A method according to any of the preceding clauses, wherein the synthetic measurement data is determined based on behavioral characteristic data and / or measurement data associated with one or more applications different from the application of interest, wherein it is assumed that this is due to commonalities in characteristic behavior between the one or more applications and the application of interest.

[0094] 9. A method according to any preceding clause, wherein the composite metrology data is generated for different combinations of processing stations of a manufacturing process.

[0095] 10. A method according to any preceding clause, wherein during use of the model to make recommendations on aspects of a manufacturing process, a metrology equipment related to the manufacturing process, and corresponding recommendations from the model are fed back as feedback data to further train the model.

[0096] 11. The method of clause 10, wherein the feedback data further comprises data relating to an actual structure or stack-up formed during the manufacturing process.

[0097] 12. A method according to any preceding clause, wherein the model is configured to determine a preferred control strategy for controlling a manufacturing process.

[0098] 13. A method according to any preceding clause, further providing a control scheme associated with a preferred control strategy.

[0099] 14. A method according to any preceding clause, wherein the model is configured to evaluate a plurality of candidate control strategies based on process data relating to the manufacturing process.

[0100] 15. The method of clause 14, wherein the step of evaluating is performed by minimizing an error function of the quality metric.

[0101] 16. The method of clause 14 or 15, wherein the model is configured to evaluate a plurality of candidate control strategies based on cost metric data, the cost metric data comprising associated cost metrics for the candidate control strategies.

[0102] 17. A method according to any preceding clause, further providing a control scheme associated with a preferred control strategy.

[0103] 18. A computer program comprising program instructions operable, when run on suitable apparatus, to perform a method according to any preceding clause.

[0104] 19. A non-transitory computer program carrier comprising a computer program according to clause 18.

[0105] Although a patterning device in the form of a physical reticle has been described, the term "patterning device" in this application also includes a data product conveying a pattern in digital form, for example to be used in conjunction with a programmable patterning device.

[0106] Although the above may have made specific reference to the use of embodiments of the invention in the context of optical lithography, it will be appreciated that the invention may be used in other applications, for example, imprint lithography, and is not limited to optical lithography where the context permits. In imprint lithography, the topography in the patterning device defines the pattern created on the substrate. The topography of the patterning device may be pressed into a layer of resist supplied to the substrate, upon which the resist is cured by applying electromagnetic radiation, heat, pressure, or a combination thereof. After the resist is cured, the patterning device is removed from the resist, leaving a pattern in the resist.

[0107] The terms "radiation" and "beam" as used with respect to a lithographic apparatus encompass all types of electromagnetic radiation, including ultraviolet (UV) radiation (e.g., having a wavelength of approximately 365 nm, 355 nm, 248 nm, 193 nm, 157 nm, or 126 nm) and extreme ultraviolet (EUV) radiation (e.g., having a wavelength in the range of 5 nm to 20 nm), as well as particle beams (e.g., ion beams or electron beams).

[0108] The term "lens," where the context allows, may refer to any one or combination of various types of optical components, including refractive, reflective, magnetic, electromagnetic, and electrostatic optical components.

[0109] The foregoing description of specific embodiments will so fully reveal the general nature of the invention that others can easily modify and / or adjust such specific embodiments for various applications without undue experimentation by applying the knowledge of those skilled in the art without departing from the general concept of the invention. Therefore, based on the teachings and guidance set forth herein, such adjustments and modifications are intended to fall within the meaning and range of equivalents of the disclosed embodiments. It should be understood that the purpose of the terms or expressions herein is to illustrate by way of example and not by way of limitation, so that the terms or expressions of this specification will be interpreted by those skilled in the art based on the teachings and guidance.

[0110] The breadth and scope of the present invention should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.

Claims

1. A method for obtaining a training data set comprising synthetic metrology data, the training data set being configured for use in training a model related to a manufacturing process for manufacturing an integrated circuit, the method comprising: obtaining behavioral characteristic data describing behavior of process parameters resulting from the manufacturing process and / or associated tools or influences; Determining the synthetic measurement data based on the behavioral characteristic data, the synthetic measurement data describing the influence of changes in the manufacturing process and / or related tools or influences on the process parameters; as well as The model is trained using the training data set, wherein the training data set includes the synthetic measurement data.

2. The method of claim 1, wherein the model is based on a neural network.

3. A method according to claim 1, wherein the behavioral characteristic data includes data related to one or more of the following: spatial parameter distribution imposed by any tools or influences related to the manufacturing process, temporal noise / drift of any parameters related to the manufacturing process, scenarios or scenario granularity, relationships between different data types and / or similar processes, and predicted events to which the system should respond.

4. The method of claim 1 , wherein determining the synthetic measurement data comprises: The behavioral characteristic is interpolated and / or extrapolated over time to obtain an extrapolated process parameter value.

5. The method of claim 1, wherein determining the synthetic measurement data is based on behavioral characteristic data associated with one or more applications different from the application of interest, wherein it is assumed that this is caused by commonality of characteristic behaviors between the one or more applications and the application of interest. The method of claim 1 , wherein the composite metrology data is generated for different combinations of processing stations of the manufacturing process.

7. The method according to claim 1, wherein during the use of the model to make suggestions on the manufacturing process, aspects of the measurement equipment related to the manufacturing process, and corresponding suggestions from the model are fed back as feedback data to further train the model.

8. The method of claim 1, wherein the model is configured to determine a preferred control strategy for controlling the manufacturing process.

9. The method of claim 8, further providing a control scheme associated with the preferred control strategy.

10. The method of claim 1, wherein the model is configured to evaluate a plurality of candidate control strategies based on process data related to the manufacturing process.

11. The method of claim 10, wherein the step of evaluating is performed by minimizing an error function of the quality metric. 12 . The method of claim 10 , wherein the model is configured to evaluate the plurality of candidate control strategies based on cost metric data, the cost metric data comprising associated cost metrics for the candidate control strategies.

13. A computer program product comprising program instructions operable to obtain a training data set comprising synthetic metrology data, the training data set being configured for use in training a model related to a manufacturing process for manufacturing an integrated circuit, the instructions being configured to: obtaining behavioral characteristic data describing behavior of process parameters resulting from the manufacturing process and / or associated tools or influences; The synthetic measurement data is determined based on the behavioral characteristic data, and the synthetic measurement data describes the influence of changes in the manufacturing process and / or related tools or influences on the process parameters.

14. The computer program product of claim 13, further comprising: Instructions configured to train the model using the training data set including the synthetic measurement data.

15. The computer program product of claim 13, the model being based on a neural network.

16. A computer program product according to claim 13, wherein the behavioral characteristic data includes data related to one or more of the following: spatial parameter distribution imposed by any tools or influences related to the manufacturing process, temporal noise / drift of any parameters related to the manufacturing process, scenarios, or scenario granularity, relationships between different data types and / or similar processes, and predicted events.

17. A computer program product according to claim 13, wherein the instructions are further configured to determine the synthetic measurement data based on behavioral characteristic data associated with one or more applications different from the application of interest, wherein it is assumed that this is caused by commonality of characteristic behaviors between the one or more applications and the application of interest.

18. The computer program product of claim 13, wherein the model is configured to determine a preferred control strategy for controlling the manufacturing process.

19. The computer program product of claim 18, further comprising instructions for providing a control scheme associated with the preferred control strategy.

20. The computer program product according to claim 13, wherein the computer program product is a non-transitory computer program carrier.