Determine corrections to the process

Through the combination of sparse and dense data adjustment model, the poor correction problem caused by sparse data sampling in lithography technology is solved. Recurrent neural network is used to predict higher-order parameters, which realizes more efficient lithography process control and improves pattern reproduction Accuracy and productivity.

CN112969967BActive Publication Date: 2024-12-03ASML NETHERLANDS BV
View PDF 11 Cites 0 Cited by

Patent Information

Application Number
CN201980073186.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-09-25
Filing Date
2019-10-09
Publication Date
2024-12-03
Estimated Expiration
2039-10-09

AI Technical Summary

Technical Problem

In existing lithography technology, sparse data sampling leads to inaccurate intensive measurement results, and the uncaptured fingerprints in batch control cannot be fully captured, resulting in poor correction of each exposure, and the behavior of distributed sampling is unstable, affecting pattern reproduction. quality.

Method used

By obtaining sparse and dense data, determining sparse to dense mismatch, adjusting the model to reduce mismatch, applying the adjusted model for correction, combining recursive neural network to predict higher-order parameters, reducing the frequency of dense measurements, and improving the control accuracy of exposure process.

Benefits of technology

It improves the correction accuracy and stability of the lithography process, reduces the intensive measurement frequency, and enhances the accuracy and production efficiency of pattern reproduction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112969967B_ABST
    Figure CN112969967B_ABST
Patent Text Reader

Abstract

A method for configuring a semiconductor manufacturing process, the method comprising: obtaining a first value of a first parameter based on measurements associated with a first operation of a process step in the semiconductor manufacturing process and a first sampling scheme; determining a predicted value of the first parameter based on the first value using a recurrent neural network; and using the predicted value of the first parameter to configure a subsequent operation of the process step in the semiconductor manufacturing process.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to EP application 18204882.7 filed on November 7, 2018, EP application 19150953.8 filed on January 9, 2019, EP application 19173992.9 filed on May 13, 2019, and EP application 19199505.9 filed on September 25, 2019, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present invention relates to a method of determining a correction to a process, a semiconductor manufacturing process, a lithography apparatus, a lithography unit and a related computer program product. Background Art

[0004] A lithographic apparatus is a machine configured to apply a desired pattern to a substrate. A lithographic apparatus can be used, for example, to manufacture integrated circuits (ICs). A lithographic apparatus can, for example, project a pattern (also often referred to as a "design layout" or "design") at a pattern forming device (e.g., a mask) onto a layer of radiation-sensitive material (resist) disposed on a substrate (e.g., a wafer).

[0005] In order to project a pattern onto a substrate, a lithographic apparatus may use electromagnetic radiation. The wavelength of the radiation determines the minimum size of a feature that can be formed on a substrate. Typical wavelengths currently used are 365nm (i-line), 248nm, 193nm, and 13.5nm. Compared to a lithographic apparatus using radiation with a wavelength of, for example, 193nm, a lithographic apparatus using extreme ultraviolet (EUV) radiation with a wavelength in the range of 4nm to 20nm (e.g., 6.7nm or 13.5nm) can be used to form smaller features on a substrate.

[0006] Low-k 1 Photolithography can be used to process features whose size is smaller than the classical resolution limit of the photolithography apparatus. In this process, the resolution formula can be expressed as CD = k 1 × λ / NA, where λ is the wavelength of the radiation employed, NA is the numerical aperture of the projection optics in the lithographic apparatus, CD is the “critical dimension” (usually the smallest feature size printed, but in this case the half pitch), and k 1 is the empirical resolution factor. Generally speaking, k 1The smaller the k, the more difficult it is to reproduce on a substrate a pattern that is similar in shape and size to what the circuit designer planned to achieve specific electrical functionality and performance. To overcome these challenges, complex fine-tuning steps may be applied to the lithographic projection apparatus and / or the design layout. These include, for example, but are not limited to, optimization of the NA, customized illumination schemes, use of phase-shifted patterning equipment, various optimizations of the design layout such as optical proximity correction (OPC, sometimes also referred to as "optical and process correction") in the design layout, or other methods generally defined as "resolution enhancement techniques" (RET). Alternatively, a tight control loop for controlling the stability of the lithographic apparatus can be used to improve the pattern resolution at low k. 1 The following reappearance.

[0007] International patent application WO 2015049087, which is incorporated herein by reference in its entirety, discloses a method of obtaining diagnostic information related to an industrial process. At a stage during the execution of a lithography process, alignment data or other measurements are taken to obtain object data representing positional deviations or other parameters measured at points spatially distributed across each wafer. Overlay and alignment residuals typically show patterns across the wafer, which are referred to as fingerprints.

[0008] In semiconductor manufacturing, the critical dimension (CD) performance parameter fingerprint can be corrected using a simple control loop. Typically, a feedback mechanism uses the scanner (a lithography device) as an actuator to control the average dose per wafer. Similarly, for the overlay performance parameter overlay, the fingerprint induced by the process tool can be corrected by adjusting the scanner actuator.

[0009] Sparse after-development inspection (ADI) measurements are used as input to a global model for controlling the scanner (usually run-to-run). Dense ADI measurements, measured less frequently, are used for modeling per exposure. By modeling at a higher spatial density using dense data, per exposure modeling can be performed for fields with large residuals. Corrections requiring such denser metrology sampling are made infrequently without adversely affecting throughput.

[0010] It is a problem that model parameters based on sparse ADI data often do not accurately represent the densely measured parameter values. This may be due to the crosstalk between the model parameters and the uncaptured parts of the fingerprint. Furthermore, the size of the model may be too large for such a sparse data set. This introduces the following problem: the per-field model cannot fully capture the uncaptured fingerprint during batch control. Another problem is that the sparse to dense behavior of distributed sampling is unstable, where different wafers (and different groups) have different samplings, so the layout of superimposing many wafers effectively leads to dense measurement results. There are large residuals between the modeled sparse data and the densely measured parameter values. This will lead to a poor fingerprint description, resulting in suboptimal correction for each exposure.

[0011] For alignment control, only a few alignment marks can be measured during exposure without affecting throughput is another problem. Higher-order alignment control requires denser alignment layout and affects throughput. Figure 5 As shown, a solution to this problem is to measure denser alignment marks in an offline tool (Takehisa Yahiro et al., Feed-forward alignment correction for advanced overlay process control usinga standalone alignment station "Litho Booster", Proc. SPIE 10585, Metrology, Inspection and Process Control of Microlithography XXXII) and feed forward this high-order correction during exposure, during which the low-order corrections are still calculated.

[0012] For overlay control, in practice, a dense overlay measurement (called high-order parameter update) is performed only once in several packets to update the high-order correction. The high-order parameters used to determine the scanner control recipe do not change between high-order parameter update measurements. Summary of the invention

[0013] It would be desirable to provide a method of determining corrections to a process that addresses one or more of the problems or limitations discussed above.

[0014] Embodiments of the invention are disclosed in the claims and the detailed description.

[0015] In a first aspect of the invention there is provided a method for determining a correction to a process, the method comprising:

[0016] - obtaining first sparse data measured using a sparse sampling layout, the first sparse data representing measured values ​​of a parameter across one or more substrates passing through the process;

[0017] - obtaining dense data measured using a dense sampling layout, the dense data representing measured values ​​of a parameter across one or more substrates passing through the process, the dense sampling layout being spatially denser than a sparse sampling layout;

[0018] -Apply the model to sparse and dense data to determine sparse to dense mismatch;

[0019] - obtaining second sparse data measured using the sparse sampling layout, the second sparse data representing measured values ​​of the parameter across the substrate passing through the process;

[0020] -Adjust the model based on sparse to dense mismatch;

[0021] - applying the adjusted model to the second sparse data to determine a sparse model result; and

[0022] -Determine corrections to the process based on the sparse model results.

[0023] In a second aspect of the invention there is provided a semiconductor manufacturing process comprising a method for determining a correction to the process according to the method of the first aspect.

[0024] In a third aspect of the present invention, there is provided a lithographic apparatus, the lithographic apparatus comprising:

[0025] - an illumination system configured to provide a projection radiation beam;

[0026] - a support structure configured to support a patterning device configured to pattern the projection beam according to a desired pattern;

[0027] - a substrate table configured to hold a substrate;

[0028] - a projection system configured to project the patterned light beam onto a target portion of the substrate; and

[0029] - A processing unit configured to determine a correction to the process according to the method of the first aspect.

[0030] In a fourth aspect of the present invention, there is provided a lithography cell comprising the lithography apparatus of the third aspect.

[0031] In a fifth aspect of the present invention, there is provided a computer program product comprising machine-readable instructions for causing a general-purpose data processing apparatus to execute the steps of the method of the first aspect.

[0032] It would also be desirable to provide a method of modeling parameter data without requiring intensive measurement of the parameter data across a substrate at a high temporal sampling frequency.

[0033] In a sixth aspect of the present invention, a method for modeling parameter data is provided, the method comprising: obtaining a set of values ​​of a parameter associated with a plurality of positions on a substrate; obtaining a first value matrix based on an evaluation of one or more basis functions at the plurality of positions; obtaining a second value matrix based on training an adjustable number matrix to a set of values ​​of previously obtained parameters associated with a previous substrate; determining coefficients of a model using the first value matrix and the second value matrix and the obtained set of values; and providing modeled values ​​of the parameters using the coefficients and the second value matrix.

[0034] It would also be desirable to provide a method of predicting parameters associated with a semiconductor manufacturing process that reduces the frequency of providing intensive measurement data while retaining the ability to accurately control the process.

[0035] In a seventh aspect of the present invention, a method for configuring a semiconductor manufacturing process is provided, the method comprising: obtaining a first value of a first parameter based on measurements associated with a first operation of a process step in the semiconductor manufacturing process and a first sampling scheme; determining a predicted value of the first parameter based on the first value using a recursive neural network; and using the predicted value of the first parameter to configure subsequent operations of the process step in the semiconductor manufacturing process.

[0036] Preferably, the method further comprises determining a control scheme for a subsequent operation of a process step in the semiconductor manufacturing process using a predicted value of a first parameter determined based on a first value obtained from a measurement at a first operation of the process step in the semiconductor manufacturing process.

[0037] In an eighth aspect of the present invention, there is provided a semiconductor manufacturing process comprising a method for predicting a value of a parameter associated with the semiconductor manufacturing process according to the method of the seventh aspect.

[0038] In a ninth aspect of the present invention, there is provided a lithographic apparatus comprising:

[0039] - an illumination system configured to provide a projection radiation beam;

[0040] - a support structure configured to support a patterning device configured to pattern the projection beam according to a desired pattern;

[0041] - a substrate table configured to hold a substrate;

[0042] - a projection system configured to project the patterned light beam onto a target portion of the substrate; and

[0043] - a processing unit, configured to:

[0044] The method according to the seventh aspect predicts a value of a parameter associated with a semiconductor manufacturing process.

[0045] In a tenth aspect of the present invention, there is provided a computer program product comprising machine-readable instructions for causing a general-purpose data processing apparatus to execute the steps of the method of the seventh aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Embodiments of the invention will now be described, by way of example only, with reference to the accompanying schematic drawings, in which

[0047] - Figure 1 A schematic overview of a lithographic apparatus is depicted;

[0048] - Figure 2 A schematic overview of a lithography unit is depicted;

[0049] - Figure 3 Depicted is a schematic representation of overall lithography that shows the collaboration between three key technologies to optimize semiconductor manufacturing;

[0050] - Figure 4 The post-development overlay control and post-etch overlay control of the process are schematically depicted;

[0051] - Figure 5 is a flow chart of a method for determining a correction to a process according to one embodiment;

[0052] - Figure 6 is a flow chart of a method for determining a correction to a process using distributed dense sampling according to another embodiment;

[0053] - Figure 7 depicts results of an example comparing residual overlay performance of conventional methods and embodiments; and

[0054] - Figure 8 Depicted with Figure 7 The comparison is similar to that of , but for the results of the example comparison of distributed sampling;

[0055] - Fig. 9 A method for modeling measured parametric data is described;

[0056] - Fig.10 A method of modeling measured parameter data according to one embodiment is depicted;

[0057] - Fig.11 A schematic overview of overlay sampling and control of a semiconductor manufacturing process is depicted;

[0058] - Fig.12 A schematic overview of known alignment sampling and control of a semiconductor manufacturing process is shown;

[0059] - Fig.13 A schematic overview of a method for sampling and controlling a semiconductor manufacturing process according to an embodiment of the present invention is depicted;

[0060] - Fig.14 is with Fig.11 Figure 2. Overlay of higher-order parameter updates corresponding to the sampling and control schemes of ;

[0061] - Fig.15 is a graph of an overlay higher order parameter update without update according to one embodiment of the present invention;

[0062] - Fig.16 According to one embodiment of the present invention, the update Fig.13 A graph of overlaid higher-order parameter updates corresponding to the sampling and control scheme of ; and

[0063] - Fig.17 Plotting the results without high-order overlay parameter correction and with Fig.14 The corresponding higher-order parameter updates and Fig.15 and Fig.16 The corresponding overlay parameters predict the measured overlay wafer map. DETAILED DESCRIPTION

[0064] In this document, the terms "radiation" and "beam" are used to cover all types of electromagnetic radiation, including UV radiation (e.g., where the wavelength is 365nm, 248nm, 193nm, 157nm or 126nm) and EUV (extreme ultraviolet radiation, e.g., where the wavelength is in the range of about 5nm to 100nm).

[0065] The terms "reticle", "mask" or "patterning device" as used herein may be broadly interpreted as referring to a general patterning device that can be used to impart an incident radiation beam with a patterned cross-section that corresponds to the pattern to be produced in a target portion of the substrate. In this context, the term "light valve" may also be used. In addition to classical masks (transmissive or reflective, binary, phase-shifting, hybrid, etc.), other examples of such patterning devices include programmable mirror arrays and programmable LCD arrays.

[0066] Figure 1A lithographic apparatus LA is schematically depicted. The lithographic apparatus LA comprises an illumination system (also referred to as an illuminator) IL configured to condition a radiation beam B (e.g., UV radiation, DUV radiation, or EUV radiation); a mask support (e.g., a mask stage) MT configured to support a pattern forming device (e.g., a mask) MA and connected to a first positioner PM configured to precisely position the pattern forming device MA according to certain parameters; a substrate support (e.g., a wafer stage) WT configured to hold a substrate (e.g., a resist-coated wafer) W and connected to a second positioner PW configured to precisely position the substrate support according to certain parameters; and a projection system (e.g., a refractive projection lens system) PS configured to project a pattern imparted to the radiation beam B by the pattern forming device MA onto a target portion C (e.g., comprising one or more dies) of the substrate W.

[0067] In operation, the illumination system IL receives a radiation beam from a radiation source SO (e.g., via a beam delivery system BD). The illumination system IL may include various types of optical components, such as refractive, reflective, magnetic, electromagnetic, electrostatic and / or other types of optical components, or any combination thereof, for directing, shaping and / or controlling the radiation. The illuminator IL may be used to condition the radiation beam B so that it has a desired spatial and angular intensity distribution in its cross-section at the plane of the patterning device MA.

[0068] The term "projection system" PS as used herein should be broadly interpreted as covering various types of projection systems, including refractive, reflective, catadioptric, anamorphic, magnetic, electromagnetic and / or electrostatic optical systems, or any combination thereof, depending on the exposure radiation being used and / or other factors such as the use of an immersion liquid or the use of a vacuum. Any use of the term "projection lens" herein may be considered synonymous with the more general term "projection system" PS.

[0069] The lithographic apparatus LA may be of a type in which at least a portion of the substrate may be covered by a liquid having a relatively high refractive index (e.g. water) to fill the space between the projection system PS and the substrate W, which is also known as immersion lithography. More information on immersion techniques is given in US6952253, incorporated herein by reference.

[0070] The lithographic apparatus LA may also be of a type having two or more substrate supports WT (also referred to as a "dual stage"). In such a "multi-stage" machine, the substrate supports WT may be used in parallel, and / or preparatory steps for subsequent exposure of a substrate W may be performed on a substrate W located on one of the substrate supports WT while another substrate W on another substrate support WT is being used to expose a pattern on another substrate W.

[0071] In addition to the substrate support WT, the lithographic apparatus LA may comprise a measurement platform. The measurement platform is arranged to hold sensors and / or cleaning equipment. The sensors may be arranged to measure characteristics of the projection system PS or characteristics of the radiation beam B. The measurement platform may hold a plurality of sensors. The cleaning equipment may be arranged to clean part of the lithographic apparatus, for example part of the projection system PS or part of a system for providing immersion liquid. The measurement platform may be moved under the projection system PS when the substrate support WT is away from the projection system PS.

[0072] In operation, a radiation beam B is incident on a patterning device (e.g. a mask MA), which is held on a mask support MT, and is patterned by a pattern (design layout) present on the patterning device MA. After passing through the patterning device MA, the radiation beam B passes through a projection system PS, which focuses the beam onto a target portion C of the substrate W. With the aid of a second positioner PW and a position measurement system IF, the substrate support WT can be precisely moved, for example, in order to position different target portions C at focus and alignment positions in the path of the radiation beam B. Likewise, the first positioner PM and possibly another position sensor (which is located in the center of the substrate support WT) are provided. Figure 1 The pattern forming device MA (not explicitly shown) can be used to accurately position the pattern forming device MA relative to the path of the radiation beam B. The pattern forming device MA and the substrate W can be aligned using the mask alignment marks M1, M2 and the substrate alignment marks P1, P2. Although the substrate alignment marks P1, P2 occupy dedicated target portions as shown, they can be located in the space between target portions. When the substrate alignment marks P1, P2 are located between target portions C, they are called scribe alignment marks.

[0073] like Figure 2As shown, the lithography apparatus LA may form part of a lithography cell LC, which is sometimes referred to as a lithocell or (lithography) cluster. The lithography cell LC may also include devices for performing pre-exposure processes and post-exposure processes on a substrate. Traditionally, these devices include a spin coater SC for depositing a resist layer, a developer DE for developing the exposed resist, a chill plate CH and / or a baking plate BK, for example, for regulating the temperature of the substrate W (for example, for regulating the solvent in the resist layer). The substrate handler or robot RO picks up the substrate W from the input / output ports I / O1, I / O2, moves them between different process devices, and then transfers the substrate W to the feed table LB of the lithography apparatus LA. These devices in the lithography cell (often collectively referred to as tracks) are usually under the control of a track control unit TCU, which itself is controlled by a supervisory control system SCS, which also controls the lithography apparatus LA, for example, via the lithography control unit LACU.

[0074] In order to correctly and consistently expose the substrate W exposed by the lithography apparatus LA, it is desirable to inspect the substrate to measure characteristics of the patterned structure, such as overlay errors between subsequent layers, line thickness, critical dimensions (CD), etc. To this end, an inspection tool (not shown) may be included in the lithography cell LC. If errors are detected, for example, adjustments may be made to the exposure of subsequent substrates or other process steps to be performed on the substrate W, especially if the inspection is still to be performed before other substrates W of the same batch or group are exposed or processed.

[0075] The inspection device (which may also be referred to as a metrology device) is used to determine characteristics of the substrate W, in particular, how characteristics of different substrates W vary or how characteristics associated with different layers of the same substrate W vary from layer to layer. Alternatively, the inspection device may be configured to identify defects on the substrate W and may, for example, be part of the lithography cell LC, or may be integrated into the lithography apparatus LA, or may even be a stand-alone device. The inspection device may measure characteristics on a latent image (the image in the resist layer after exposure), a semi-latent image (the image in the resist layer after a post-exposure bake step PEB), or a developed resist image (where exposed or unexposed parts of the resist have been removed), or even an etched image (after a pattern transfer step such as etching).

[0076] Typically, the patterning process in the lithography apparatus LA is one of the most critical steps in a process that requires high-precision sizing and placement of structures on the substrate W. In order to ensure this high precision, Figure 3As schematically depicted, the three systems can be combined into a so-called "holistic" control environment. One of these systems is the lithography apparatus LA, which is (virtually) connected to the metrology tool MT (second system) and the computer system CL (third system). The key to this "holistic" environment is to optimize the cooperation between these three systems to enhance the overall process window and provide a tight control loop to ensure that the patterning performed by the lithography apparatus LA remains within the process window. The process window defines the range of process parameters (e.g., dose, focus, and overlay) within which a particular manufacturing process produces a defined result (e.g., a functional semiconductor device) - typically within this process parameter range, changes in process parameters in the lithography process or patterning process are allowed.

[0077] The computer system CL can use (a portion of) the design layout to be patterned to predict which resolution enhancement techniques to use, and perform computational lithography simulations and calculations to determine which mask layouts and lithography apparatus settings achieve the maximum overall process window (in Figure 3 Typically, the resolution enhancement technique is arranged to match the patterning possibilities of the lithographic apparatus LA. The computer system CL may also be used to detect where within the process window the lithographic apparatus LA is currently operating (e.g. using input from the metrology tool MT) to predict whether defects may exist due to, for example, suboptimal processing (e.g., in the process window in which the lithographic apparatus LA is currently operating). Figure 3 , depicted by an arrow pointing to “0” in the second scale SC2).

[0078] The metrology tool MT may provide input to the computer system CL to enable accurate simulations and predictions, and may provide feedback to the lithographic apparatus LA to identify, for example, possible drift of the lithographic apparatus LA in a calibrated state (e.g., Figure 3 , depicted by multiple arrows in the third scale SC3).

[0079] Process Control Using Sparse-to-Dense Mismatch Correction

[0080] Figure 4 The post-development overlay control and post-etch overlay control of the process are schematically depicted. The feedback control loop operates outside the lithography apparatus. In the control loop design, the following parts play a role.

[0081] Horizontal arrows 402 represent the flow of substrates through the lithography process. Several arrows are stacked, which represent time t. Exposure (EXP) 404 is followed by post-development inspection (ADI) overlay measurements 406, 408. Etching (ETC) 410 is followed by dense and sparse post-etch inspection (AEI) overlay measurements 412 and ultra-dense post-etch inspection overlay measurements 414. Sparse measurements 406, 412 are used to limit metering time, where dense measurements 408, 412, 414 are performed less frequently because they require more metering time. Post-development 406 measurements using a sparse sampling layout are performed to produce sparse ADI data 416 (e.g., ≤200 points per wafer). Sparse data is modeled using a certain model (composed of different parameter sets; for example, radial tangential field parameters, hyperbolic or exponential edge models, or intra-field polynomial models) that describes the process fingerprint 428 in an adequate manner (fingerprint capture) without introducing too much noise. More parameters means better fingerprint description, but also more noise.

[0082] The sparse model results (process fingerprints) 428 are averaged over each group (e.g., using an exponentially weighted moving average) to reduce the impact of inter-group variations, and as such can be used directly or via a correction optimization step (OPT) 426 to provide a stable correction for the exposure 404 that can be applied to one or more next groups.

[0083] The sparse measurement layout is optimized to capture the data into the model (reducing the uncertainty of the normalized model) and has uniform spatial coverage. The uncertainty of the model is generally defined as the propagation of the measurement error relative to the modeling (e.g., fitting) error when the model is applied to the measurement. A more detailed explanation of the uncertainty of the model (more specifically, the normalized model uncertainty (nMU) (commonly referred to as the G optimality criterion)) is given in paragraph

[0170] of U.S. Patent Application Publication US2018 / 0067900, the entire contents of which are incorporated herein by reference.

[0084] Post-etch 412 measurements using either a sparse sampling layout or a dense sampling layout produce sparse AEI data 422. These post-etch measurement data 422 are used to derive global sparse model results 434. The global sparse model results 434 are used to apply a model offset (metrology to device, MTD) 432 to the post-development data via the ADI measurement steps 406, 408. This optional step can be performed so that control is based on reducing post-etch overlay.

[0085] Dense post-development 408 measurements or ultra-dense post-etch 414 measurements are performed. These measurements are typically not performed for every packet due to the additional metrology effort required. Dense data 418 (e.g., 2,000 points per wafer) or ultra-dense data 424 (e.g., 10,000 points per wafer) are also used to model fingerprints associated with individual exposure fields, for example, to implement per-exposure correction (CPE) for overlay fingerprints. To this end, higher-order modeling (typically, per-field models) is performed to provide dense model results 436 and ultra-dense model results 438. The assumption is that higher-order fingerprints may be more stable and, as such, do not need to be modeled for each stack.

[0086] To avoid double correction, the higher order model is determined based on the residuals of the lower order model. The content of the lower order model can be determined based on the dense data, however, this may result in a dense to sparse mismatch. That is, the regular lower model determination is based on sparse data, which may (only when the lower order model does not capture the fingerprint - which is usually the case) give different results than when modeling on the dense data. To correct for this, the dense data can first be downsampled to determine the content of the lower order model. Therefore, each exposure modeling is based on the residuals of the sparse subsampled data.

[0087] The higher order (per exposure modeling) ADI fingerprint 430 and the higher order (per exposure modeling) AEI fingerprint 438 are added (optionally also after averaging over groups) to the lower order fingerprints to provide a single set of corrections via an optimization step (OPT) 426 .

[0088] The post-etch measurement has the same density as the post-development measurement, or the post-etch measurement has a higher density than the post-development measurement, thus allowing for an even higher order fit.

[0089] Additionally or alternatively, dense data can be collected by means of distributed sampling. Different wafers (and different groups) have different sampling schemes, so a layout that overlays the sampling schemes of many wafers effectively results in dense measurement results. In this case, no separate dense measurements are required.

[0090] There is a sparse to dense mismatch problem. The performance of sparse measurements depends on how well the fingerprint can be described by a finite number of sampling points. If the model used does not fully describe the fingerprint (which is usually the case), the fingerprint capture will not be as good as when using a dense layout.

[0091] There is the problem of dense overlay to dense equipment measurement mismatch. Dense overlay data does not always represent the overlay between actual product structures on the wafer. The reason is that the design of the overlay target is inherently different from the product structure and therefore has a different response to the optical signal used for metrology. This contributes to the metrology-to-equipment offset. The inherent noise of the metrology equipment also plays a role in this. If the overlay data can be trained (such as using electrical measurements as a reference) to filter out this offset, then a 'cleaner' overlay measurement may be used to control the product overlay.

[0092] There is the problem of noise propagation in distributed sampling. This is a form of sparse to dense mismatch, but now it varies from wafer to wafer or from group to group (depending on the type of distribution).

[0093] Another problem is that the size of the model used on sparse data may be too large.

[0094] Typically, a per-field model is used for modeling each exposure (field), which will partially pick up sparse-to-dense mismatches that the per-field model is not designed for. A global model can be used, but the per-field model content is missed. Combining is expensive in terms of metrology requirements and will show increased noise performance.

[0095] Another problem is that per-exposure modeling does not capture the higher-order content of each pile or the sparse-to-dense mismatch.

[0096] In the case of distributed sampling, the dense-to-sparse processing in each exposure modeling is not optimal, except when the same wafer is measured with respect to the sparse measurement and when a specific sparse layout is used for mismatch processing. In fact, even in the case of distributed sampling, each exposure modeling update will not be performed for each group.

[0097] Reference below Figure 5 and 6 In one embodiment described, historical data is used to focus on the sparse to dense mismatch to solve the above problem. For example, given a model and a sampling layout, the following steps are taken:

[0098] Based on historical data, sparse overlay measurements are used as the trainable dataset, while dense overlay data are used as the reference dataset.

[0099] Based on the application of the model, an automated algorithm is used to assess which portion or portions of the model correction parameters are responsible for the significant sparse to dense mismatch. Based on this knowledge, a set of weighting factors is obtained that minimizes the sparse to dense mismatch for the historical data set. The adjusted model is applied to future data sets.

[0100] Figure 5is a flow chart of a method for determining a correction to a process according to one embodiment. The method has the following steps:

[0101] First (historical) sparse data 504 is obtained using sparse sampling layout measurements 502, the first (historical) sparse data 504 representing measured values ​​of parameters across substrates through the process. Examples of parameters are overlay, alignment, focus, and dose.

[0102] Historical dense data 512 is obtained that represents measured values ​​of the parameter across one or more substrates through the process using a dense sampling layout measured 510 that is spatially denser than a sparse sampling layout.

[0103] The model 508 is applied 506 to the sparse data and the dense data to determine the sparse to dense mismatch.

[0104] Second sparse data 522 is obtained, which represents the measured value of the parameter across the entire substrate through the process. The second sparse data 522 is measured 516 using a sparse sampling layout.

[0105] Based on the sparse to dense mismatch, the model 508 is adjusted 524. This can be done by evaluating the different contributions of the corresponding parts of the model to the sparse to dense mismatch. Then, a weighting 514 factor is determined to weight the corresponding parts of the model to reduce the sparse to dense mismatch or preferably minimize the sparse to dense mismatch. The model 508 is adjusted using the weighting factor 514. Determining the sparse to dense mismatch can include: training a matrix on the first sparse data and the dense data. In this case, the step of adjusting the model based on the sparse to dense mismatch then includes: using the matrix to modify the second sparse data.

[0106] The adjusted model is applied 524 to the second sparse data 522 to determine a sparse model result 526 .

[0107] The sparse model results 526 may be averaged 532 over several groups. Based on the sparse model results 526, corrections to the process are determined 542.

[0108] The correction is then applied to the process to control the process. For example, scanner settings can be adjusted based on the correction to control the photolithography process.

[0109] Post-etch inspection or electrical measurements of semiconductor device 536 may be used to determine a metrology-to-device (MTD) model offset 540, which is applied along with a determined correction 542. When the MTD between overlay and some other measurement is of interest, the other measurement may be used as reference dense data 512, and the sparse overlay data 504 itself may be trained to that reference dense data 512.

[0110] Figure 6 is a flow chart of a method for determining a correction to a process according to another embodiment using distributed dense sampling. Figure 5 The same features have the same reference numerals. The method comprises the following steps:

[0111] First (historical) sparse data 504 is obtained using the sparse sampling layout measurements 502, the first (historical) sparse data 504 representing measured values ​​of the parameter across substrates going through the process.

[0112] Historical dense data 512 is obtained that represents measured values ​​of the parameter across one or more substrates passing through the process using a dense sampling layout measured 510 that is spatially denser than a sparse sampling layout.

[0113] The model 508 is applied 506 to the sparse data and the dense data to determine the sparse to dense mismatch.

[0114] Second sparse data 622 representing measured values ​​of a parameter across a plurality of substrates passing through a process is obtained by downsampling 620 second dense data 618. The dense data 618 is measured 616 using a different sparse sampling layout distributed across a plurality of substrates.

[0115] The model 508 is adjusted 524 based on the sparse to dense mismatch. This can be done by evaluating the different contributions of the corresponding parts of the model to the sparse to dense mismatch. A weighting 514 factor is determined to weight the corresponding parts of the model to reduce the sparse to dense mismatch or preferably minimize the sparse to dense mismatch. The model 508 is adjusted using the weighting factor 514. Determining the sparse to dense mismatch can include: training a matrix on the first sparse data and the dense data. In this case, the step of adjusting the model based on the sparse to dense mismatch then includes: using the matrix to modify the second sparse data.

[0116] The adjusted model is applied 524 to the second sparse data 522 to determine sparse model results 526, 628. The sparse model results include a sparse model residual 628.

[0117] Model 508 is applied 630 to sparse model residuals 628 to determine dense model results 634, which are per-exposure modeling fingerprints. Dense model results 634 may be averaged 638 over several groups.

[0118] The sparse model results 526 may be averaged 532 over several groups. Based on the sparse model results 526 and the dense model results 634, corrections to the process are determined 542.

[0119] The correction is then applied to the process to control the process. For example, scanner settings can be adjusted based on the correction to control the photolithography process.

[0120] Post-etch inspection or electrical measurements of semiconductor device 536 may be used to determine a metrology-to-device (MTD) model offset 540, which is applied along with a determined correction 542. When the MTD between overlay and some other measurement is of interest, the other measurement may be used as reference dense data 512, and the sparse overlay data 504 itself may be trained to that reference dense data 512.

[0121] In an embodiment, a method is implemented to map model parameters associated with fitting sparse parameter data (e.g., sparse ADI overlay, alignment, focus, etc.) to updated model parameters that are better fitted to describe the dense distribution of values ​​of the parameter of interest across the wafer.

[0122] For example, a matrix can be trained on historical sparse and dense (ADI / AEI) data that maps sparse model parameters to modified sparse model parameters that are more suitable for representing densely measured parameter data.

[0123] Using modified model parameters produces smaller residuals between modeled sparse data and densely measured parameter values, resulting in better fingerprint descriptions (global model and per-field model), resulting in better correction for per-exposure correction (CPE).

[0124] Figure 7 The results of an example comparing the residual overlay performance of the dense layout of the conventional method and the embodiment are depicted. The sparse layout contains 200 points, where the dense layout is . Wafer maps 702 to 710 of overlay residuals are shown. Large residuals are depicted as longer vectors at each point on the wafer. Therefore, darker areas of the wafer map represent poorer residual overlay performance. The results of the decorrected 702 and corrected 704 using the conventional method (A) are shown. Figure 5 The decorrected 708 results and the corrected 710 results of the method (B) of the described embodiment. The box plot 712 compares the mean 3-sigma overlay (overlay residual) in the x-direction (OVX) and y-direction (OVY) in nm for the conventional method (A) and the embodiment (B). The embodiment (B) has a lower overlay residual than the conventional method (A). This illustrates the advantages of the embodiment.

[0125] Figure 8Results for an example of distributed sampling are depicted. The distributed layout uses 12 wafer cycles on 200 points in each different sparse layout to provide a dense layout (~2000 points). Figure 7 Wafer images 802 to 812 of overlay residuals are shown. The results of decorrection 802 and correction 804, 806 using conventional method (A) are shown. Figure 6 Results of the method (B) of the described embodiment after decorrection 808 and after correction 810, 812. Modeling is performed in two steps: on a global level 804, 810 and from the global residuals for the per-exposure correction (CPE) model (per-exposure modeling level) 806, 812.

[0126] Box plot 814 compares the mean 3-sigma overlay (overlay residual) in the x-direction (OVX) and y-direction (OVY) in nm for the conventional method (A) and the embodiment (B) at the global level (step 1). The embodiment (B) has a lower overlay residual than the conventional method (A).

[0127] Box plots 816 compare the mean 3-sigma overlay (overlay residual) in nm in the x-direction (OVX) and y-direction (OVY) for the conventional method (A) and embodiment (B) at the per-exposure modeling level (step 2). Embodiment (B) has lower overlay residuals than the conventional method (A). The noise suppression of embodiment (B) already helps at the global level 814, so the cumulative effect makes the per-exposure modeling correction even more noise free at the per-exposure modeling level 816.

[0128] Embodiments may be implemented in a semiconductor manufacturing process including a method for determining a correction to the process according to embodiments described herein.

[0129] Examples can be found in Figure 1 and Figure 2 The method is implemented in the lithographic apparatus or lithographic cell described.

[0130] The embodiments may be implemented in a computer program product comprising machine-readable instructions for causing a general-purpose data processing apparatus such as Figure 2 The LACU in the embodiment of the present invention performs the steps of the method described in this article.

[0131] Embodiments actively adjust the overlay model with respect to historical data, thereby suppressing noise from the overlay model. Information collected from historical data sets is used to learn more about the interaction of metrology data and model / sampling. Embodiments also provide two-step noise suppression when distributed sampling.

[0132] If we observe the training results and determine which parameters are becoming more important and which are becoming less relevant, we may understand which process tool is failing and then proactively improve the control process (such as increasing the frequency of modeling per exposure, etc.).

[0133] As mentioned above, typically a per-field model is used for per-exposure modeling, which in part will pick up sparse-to-dense mismatches for which the per-exposure modeling is not designed. Embodiments reduce the noise impact of the per-field model so that fewer measurement points can be used for per-exposure modeling updates.

[0134] In summary, embodiments provide improved process control resulting in process yield performance gains.

[0135] Further embodiments related to the various aspects of the invention described so far are disclosed in the following list of numbered clauses:

[0136] 1. A method of determining a correction to a process, the method comprising:

[0137] - obtaining first sparse data measured using a sparse sampling layout, the first sparse data representing measured values ​​of a parameter across one or more substrates passing through the process;

[0138] - obtaining dense data measured using a dense sampling layout, the dense data representing measured values ​​of a parameter across one or more substrates passing through the process, the dense sampling layout being spatially denser than a sparse sampling layout;

[0139] -Apply the model to sparse and dense data to determine sparse to dense mismatch;

[0140] - obtaining second sparse data measured using the sparse sampling layout, the second sparse data representing measured values ​​of the parameter across the substrate passing through the process;

[0141] -Adjust the model based on sparse to dense mismatch;

[0142] - applying the adjusted model to the second sparse data to determine a sparse model result; and

[0143] -Determine corrections to the process based on the sparse model results.

[0144] 2. The method of clause 1, wherein the step of adapting the model based on the sparse to dense mismatch comprises:

[0145] - Evaluate the different contributions of corresponding parts of the model to the sparse to dense mismatch;

[0146] - determining weighting factors for weighting corresponding parts of the model to reduce the sparse to dense mismatch; and

[0147] - Use weighting factors to adjust the model.

[0148] 3. The method according to clause 2, wherein the step of determining the weighting factor comprises:

[0149] - Determine a weighting factor that is used to weight corresponding parts of the model so that the sparse to dense mismatch is minimized.

[0150] 4. A method according to any preceding clause, wherein determining the sparse to dense mismatch comprises: training a matrix on first sparse data and dense data; and the step of adjusting the model based on the sparse to dense mismatch comprises: using the matrix to modify the second sparse data.

[0151] 5. A method according to any preceding clause, wherein

[0152] - the step of obtaining second sparse data comprises: obtaining second sparse data by downsampling second dense data, the second dense data representing measured values ​​of the parameter across a plurality of substrates passing through the process, the dense data being measured using different sparse sampling layouts distributed across the plurality of substrates;

[0153] -Sparse model results include sparse model residuals;

[0154] -The method further comprises: applying the model to the sparse model residuals to determine a dense model result; and

[0155] -The step of determining corrections to the process is also based on the intensive model results.

[0156] 6. A semiconductor manufacturing process comprising a method for determining a correction to the process according to any preceding clause.

[0157] 7. A lithography apparatus comprising:

[0158] - an illumination system configured to provide a projection radiation beam;

[0159] - a support structure configured to support a patterning device configured to pattern the projection beam according to a desired pattern;

[0160] - a substrate table configured to hold a substrate;

[0161] - a projection system configured to project the patterned beam onto a target portion of the substrate; and

[0162] - a processing unit configured to determine a correction to the process according to the method of any of clauses 1 to 5.

[0163] 8. A lithography cell comprising a lithography apparatus according to clause 7.

[0164] 9. A computer program product comprising machine-readable instructions for causing a general purpose data processing apparatus to perform the steps of the method according to any one of clauses 1 to 5.

[0165] Machine Learning-Based Enhancement of Sparsely Measured Parametric Data

[0166] In another aspect of the invention, a model is trained based on dense measurement data. A typical model includes model coefficients and basis functions, wherein the model coefficients are associated with the basis functions. The basis functions can be evaluated at coordinates on a measurement grid of interest and then multiplied with the model coefficients to obtain one or more modeled values ​​of a parameter of interest (e.g., overlay, etc.).

[0167] Traditionally, the use of the model is achieved by constructing a design matrix comprising a matrix DM of size N*M comprising: N values ​​for each of the M basis functions obtained by determining the value of each basis function 1..M at each coordinate 1..N. The coordinates are usually associated with the coordinates for which the measured values ​​of the parameters (overlay, alignment, CD, or any other performance or process parameter) are available. Multiplying the design matrix DM with a column vector F of fitting coefficients f1..fM results in a column vector MV with values ​​close to the original measured values ​​of the parameters of interest. This is done in Fig. 9 Mathematically, the task of determining the fitting coefficients based on the vector of available measurement data is then: DM*F=MV, which can be converted to: F=(DM)^-1*MV. Further reference is made to international patent application WO2013092106A1, in particular paragraphs

[0387] to

[0394] , which explain a method of defining a design matrix and determining a sampling scheme that is tailored to minimize the model uncertainty associated with the sampling scheme.

[0168] The basis functions utilized must represent the typical spatial behavior of the values ​​of the parameter of interest (e.g., across a substrate, a field on a substrate, or even a region (subfield) within a field). When, for example, the overlay fingerprint across a substrate is typically limited to a sixth-order polynomial order, the basis functions utilized do not have to be extended to a tenth-order polynomial, as this may result in overfitting behavior. However, the opposite may be the case; if the parameter values ​​are expected to exhibit high-order behavior, the basis functions need to be extended to a sufficiently high order to model the measured data with sufficient accuracy. In the latter case, it is necessary to obtain measured data with a sufficiently high density to prevent underfitting from occurring, which is usually regarded as non-physical fitting artifacts, such as high amplitude and high-order fluctuations in the modeled parameter values. Therefore, a disadvantage of the prior art modeling methods is that the selected model (e.g., the selected set of basis functions) needs to be compatible with the spatial variations that actually occur during a possibly long period of time. This may require frequent measurements of the parameters (values) in time and dense measurements in space in order to ensure that the modeled values ​​represent the actual parameter fingerprint behavior. However, in a high-volume manufacturing environment, there is little opportunity to meet this requirement, as this will have a negative impact on the throughput of the manufacturing process.

[0169] Therefore, there is a need for a modeling approach that ensures high-accuracy modeling of all actually occurring spatial components included within the measurement data of the parameter of interest, while being trained on dense measurement data but being deployed on frequently measured sparse measurement data.

[0170] It is proposed to extend the state of the art modeling methods by so called free-form parameters which are independent of the basis functions but which are trained outside the evaluation of the design matrix on dense measured data. Fig.10 The method of the present invention is depicted in . The free-form parameters are laid out as a free-form matrix FM of size N*L, where L is the number of free-form coefficients associated with the free-form matrix FM, and the values ​​of the free-form matrix are (dynamically) adjustable numbers. The fitting coefficient column vector F' now includes the fitting coefficients associated with the basis functions, and also includes L free-form fitting coefficients. The fitting coefficients associated with the basis functions can be determined based on the previously defined matrix equation (F = (DM^-1)*MV), and the free-form parameters can be tuned to improve the quality of the modeling parameter values ​​(for example, based on the similarity with the measured values ​​M). The free-form parameters and free-form coefficients can be determined, for example, by using a gradient descent method applied to the matrix equation F' = ((DM+FM)^-1)*MV. The main result is to obtain an optimized set of free-form parameters and free-form coefficients.

[0171] Free-form parameters can be obtained by machine learning methods using a data set of densely measured parameter values. In this case, the free-form parameters are considered to be features related to the observed parameter values ​​(measurements). The training aspect ensures that the model (especially the trained free-form fitting coefficients) accurately represents the higher-order behavior of the measured data. Typically, the lower-order behavior is accurately obtained by modeling frequently measured sparse data using lower-order basis functions and fitting coefficients associated with these basis functions. The high-order content of the modeled measurement data is then mainly provided by the trained free-form parameters and free-form fitting coefficients. Using the modeling method of the present invention, there is no need to frequently provide dense measurement data, because the model itself provides the higher-order content (prediction) without relying on some minimum required density of the data.

[0172] In one embodiment of the present invention, a method for modeling parameter data is provided, the method comprising: obtaining values ​​of parameters associated with multiple locations on a substrate; and providing a first value matrix based on evaluating one or more basis functions at the multiple locations and providing a second value matrix based on training an adjustable digital matrix to a previously obtained set of values ​​of the parameters associated with previous substrates; aggregating the first set of values ​​and the second set of values; determining coefficients of a model using the aggregated set and the obtained values ​​of the parameters; and using the coefficients and the second value matrix to provide modeled values ​​of the parameters.

[0173] In one embodiment, the parameter is overlay.

[0174] In one embodiment, the plurality of locations is associated with a sparse measurement scheme across one or more substrates, such as wafers processed during semiconductor manufacturing.

[0175] In one embodiment, the basis functions are polynomials.

[0176] In one embodiment, the basis functions are mutually orthogonal.

[0177] In one embodiment, the basis functions are at most sixth order across the substrate.

[0178] In one embodiment, the plurality of locations are associated with exposure fields on the wafer.

[0179] In one embodiment, the model parameters are free form model parameters.

[0180] In one embodiment, the trained relationship is obtained based on initial values ​​of the parameter across one or more densely sampled previous substrates.

[0181] In one embodiment, the trained relationship is based on using a machine learning method such as a neural network to establish a relationship between various model parameters and fingerprints of the parameters across one or more substrates.

[0182] In one embodiment, the fingerprint of the parameter comprises higher order components.

[0183] In one embodiment, the higher order components are related to sample point specific values ​​of the parameters.

[0184] Alternatively, the modeling of the measured data can be based entirely on free-form model parameters and free-form coefficients. In this case, the method for modeling parameter data according to an embodiment includes: obtaining a set of values ​​of a parameter associated with a plurality of positions on a substrate; and using the set of values ​​as an input to a machine learning model, the machine learning model being trained to map one or more values ​​of the parameter to a predicted value of the parameter to obtain a modeled value of the parameter. In this case, the values ​​of the parameters associated with the various positions on the substrate are considered to be features of the (modeled) values ​​of the parameters that can be mapped to these positions. For example, the machine learning model can be a neural network with weights and biases that is optimized to map the obtained values ​​to the modeled values. The training can be based on a matrix defining N*M free-form parameters and M free-form coefficients associated with the machine learning model, where N is the number of positions on the substrate, and based on optimizing the similarity between the N predicted values ​​of the parameter and the measured value of the parameter to determine the M free-form coefficients of the machine learning model. Using the determined M free-form coefficients and N*M free-form parameters, N modeled values ​​of the parameter can be determined.

[0185] Other embodiments of the invention are disclosed in the following numbered list of clauses:

[0186] 10. A method for modeling parameter data, the method comprising:

[0187] obtaining a set of values ​​for a parameter associated with a plurality of locations on a substrate;

[0188] obtaining a first value matrix based on evaluation of one or more basis functions at a plurality of locations;

[0189] obtaining a second matrix of values ​​based on training the adjustable digital matrix to a previously obtained set of values ​​for parameters associated with a previous substrate;

[0190] determining coefficients of the model using the first value matrix and the second value matrix and the obtained set of values; and

[0191] Use the coefficient and second value matrices to provide modeled values ​​for the parameters.

[0192] 11. The method of clause 10, wherein the parameter is overlay.

[0193] 12. A method according to clause 10 or 11, wherein the plurality of locations are associated with a sparse measurement scheme across one or more substrates passing through a semiconductor manufacturing process.

[0194] 13. A method according to any of clauses 10 to 12, wherein the basis functions are polynomials.

[0195] 14. A method according to any one of clauses 10 to 13, wherein the basis functions are mutually orthogonal.

[0196] 15. The method of clause 13, wherein the basis functions are at most 6th order across the substrate.

[0197] 16. A method according to any of clauses 10 to 15, wherein the plurality of positions relate to an exposure field on the substrate.

[0198] 17. A method according to any of clauses 10 to 16, wherein the adjustable number is a free form model parameter.

[0199] 18. A method according to any of clauses 10 to 17, wherein the set of previously obtained values ​​of the parameter is associated with a previously densely measured substrate.

[0200] 19. A method according to any of clauses 10 to 18, wherein the training is based on using a machine learning method such as a neural network to establish a relationship between the adjustable numbers and the fingerprint of the parameters across each individual one of the previous substrates.

[0201] 20. The method of clause 19, wherein the fingerprint of the parameter comprises higher order components.

[0202] 21. A method according to clause 20, wherein the higher order components are related to sample point specific values ​​of the parameters.

[0203] 22. A computer program product comprising machine-readable instructions for causing a general purpose data processing apparatus to perform a method according to any of clauses 10 to 21.

[0204] Recurrent Neural Networks for Predicting High-Order Fingerprints

[0205] Fig.11 A schematic diagram depicting overlay sampling and control of a semiconductor manufacturing process. Fig.11, shows a sequence of ten operations L1 to L10 of an exposure process step on ten wafer groups (or batches or wafers). The value of the high-order overlay parameter HO1 is obtained based on a measurement 1104 of the first group L1 using a spatially dense sampling scheme. The high-order overlay parameter HO1 is used to configure the semiconductor manufacturing process, for example, by determining a control scheme 1106 for subsequent exposures L2 to L6 for the next five groups. Then, an updated value of the high-order overlay parameter HO6 is obtained based on the earlier 1102 high-order overlay parameter HO1 and based on a measurement 1108 of the sixth group L6 using a spatially dense sampling scheme. In this example, the higher-order parameter update is repeated when exposing every fifth group.

[0206] At the same time, for each group of exposures, low-order corrections for each group are calculated based on sparse measurements. For example, when exposing group L1, low-order overlay parameters LO1 are obtained based on measurements 1110 using a sparse sampling scheme that is less dense in space and more frequent than a spatially dense sampling scheme. The low-order parameters LO1 are used to configure the semiconductor manufacturing process, for example, by determining a control scheme 1112 for a subsequent operation L2 of the exposure step.

[0207] Therefore, low-order corrections are computed for each group from sparse measurements, and high-order corrections are obtained from dense measurements once in several groups.

[0208] Fig.12 A schematic overview of alignment sampling and control of a semiconductor manufacturing process is depicted. Fig.12 , wafer groups L1 to L10 have an offline alignment mark measurement step 1202. The measurement 1204 is performed by an offline measurement tool 1206 that is optimized for offline measurements with a high spatial sampling density. For each wafer group L1 to L10, the measured high-order alignment parameter value 1208 is stored in HO1 to HO10. Each high-order alignment parameter value is then used to determine a control scheme 1212 for the operation of an exposure step 1214 on the corresponding wafer group L1 to L10.

[0209] At the same time, for each grouped exposure, low-order corrections for each group are calculated based on sparse measurements. For example, when group L1 is exposed 1214, low-order alignment parameters 1216 are obtained based on measurements using a sparse sampling scheme that is less dense in space than a spatially dense sampling scheme. It has the same frequency (per group) as the offline dense measurements 1204 of the high-order alignment parameters. The low-order parameters 1216 are used to determine the control scheme for operation L1 of the same exposure step.

[0210] The embodiment adopts a strategy of updating the overlay and alignment measurements between dense measurements using a long short-term memory (LSTM) prediction algorithm. This improves the performance of alignment and overlay control with minimal impact on throughput. A completely independent LSTM network prediction (no dense measurements required after training) is also possible, however if the learning is insufficient, it may diverge after a certain time.

[0211] Fig.13 A schematic overview of a method for sampling and controlling a semiconductor manufacturing process according to an embodiment of the present invention is depicted.

[0212] refer to Fig.13 , the update of high-order parameters is achieved by using an LSTM neural network to predict between groups / wafers. This provides improved high-order corrections for both alignment and overlay. Low-order corrections are measured for each wafer, while high-order corrections are predicted by LSTM between groups / wafers. The neural network is configured using the initial training (TRN) step 1302.

[0213] Fig.13 A method for predicting the value of a high-order parameter associated with a semiconductor manufacturing process is depicted. In this example, the semiconductor manufacturing process is a photolithography exposure process. The parameter is a three-order scanner exposure magnification parameter D3y in the y direction. The method includes obtaining a value 1308 of the high-order parameter based on measurements 1306 using a spatially dense sampling scheme. A recursive neural network (NN) 1304 is used to determine a predicted value 1312 of the high-order parameter based on the measured value 1308. The measured value 1308 of the high-order parameter can be directly used 1310 to determine a control scheme for processing the measured grouping.

[0214] A predicted value 1312 of a high-order parameter is determined based on a value 1308 of the high-order parameter obtained from a measurement 1306 at a first operation L1 of a process step in a semiconductor manufacturing process to determine a control scheme 1314 for a subsequent operation L2 of a process step in the semiconductor manufacturing process. The predicted value is fed back 1324 to a recurrent neural network (NN t=2 ). For exposure of groups L3-L5, the recurrent neural network (NN t=2..5 ) is self-updated to generate prediction values ​​for determining the control scheme for the successive groups. Subsequent values ​​626 of the high-order parameters are obtained based on the measurements 628 using the dense sampling scheme. The subsequent values ​​626 are used to update the recurrent neural network (NN t=6 ).

[0215] At the same time, values ​​1318 of low-order parameters are obtained based on measurements using a spatially sparse sampling scheme (e.g., values ​​1320 from grouped measurements at exposure step L2). The sparse sampling scheme is less spatially dense and more frequent than the high-order sampling scheme used for measurements 1306, 1328. The values ​​1320 of the low-order parameters are used to determine a control scheme 1322 for subsequent operation L2 of the exposure process step.

[0216] The value 1320 of the low-level parameter may be obtained based on measurements performed on the same substrate supported on the same substrate stage where the subsequent operation L2 of the process step is performed. A control scheme 1322 is determined using the value 1320 of the low-level parameter.

[0217] In this example, the semiconductor manufacturing process is a batch process for patterning substrates. The sampling scheme used to obtain high-order parameters measures every 5 batches (e.g. Fig.13 The second sampling plan measures once per batch.

[0218] In this example, the semiconductor manufacturing process is a process of patterning a substrate using exposure fields. The spatial density of the sampling scheme for obtaining high-order parameters is 200 to 300 measurement points per field, and the spatial density of the sampling scheme for obtaining low-order parameters is 2 to 3 measurement points per field.

[0219] As reference Figure 6 The described method of predicting the value of a parameter associated with a semiconductor manufacturing process can be implemented within a semiconductor manufacturing process. The method can be implemented in a processor having a processing unit such as Figure 2 It can be implemented in the LACU of the lithography device. Figure 2 Supervisory Control System SCS or Figure 3 is implemented in a processor in a computer system CL.

[0220] The present invention may also be embodied as a computer program product comprising computer readable instructions for causing a general purpose data processing apparatus to execute the Fig.13 The steps of the method described.

[0221] Fig.13 The advantage of this method is that Fig.11Compared to the method of , no additional measurements are required for overlay. For alignment, spatially dense measurements can be performed with only a few wafers per group, while all wafers can receive different control schemes determined based on different high-order parameters. Intermediate groups (for overlay) or wafers (for alignment) receive updated control schemes determined by the high-order parameters predicted by the LSTM. For overlay parameters and alignment parameters, no wafer stage (chuck) matching is required because low-level measurements and corresponding control scheme updates are performed on the same wafer stage.

[0222] Fig.14 is with Fig.11 Figure 2. Overlay of high-order parameter updates corresponding to the sampling and control schemes. Figures 14 to 16 In FIG. 1 , the horizontal axis is the wafer number WN. The vertical axis is the three-step scanner exposure magnification parameter D3y in the y direction. Figures 14 to 16 , the solid line graph 1402 is the raw values ​​of the high-order parameters from a dense measurement sampling.

[0223] refer to Fig.14 , the dashed line graph 1404 to the right of the vertical dotted line is a high-order parameter update, where the parameters are updated every 10 points (note that Fig.11 This updating is shown every 5 points).

[0224] Fig.15 is a diagram of an overlay higher order parameter update without update according to one embodiment of the present invention.

[0225] refer to Fig.15 , the dashed line graph 1504 to the right of the vertical dotted line is the predicted parameter value that has not been updated. Fig.13 Correspondingly, but there is no update in L6 or later. Therefore, it uses a free LSTM, which has no update after training.

[0226] Fig.16 is an update and Fig.13 Figure 2. Overlay of higher-order parameter updates corresponding to the sampling and control scheme.

[0227] refer to Fig.16 , the dashed line graph 1604 to the right of the vertical dotted line is the predicted parameter value, which is updated every 10 points (note that Fig.13 This update is shown every 5 points). The LSTM network was trained using 78% of the data to the left of the vertical dotted line. The LSTM network was validated on the remaining (22% - 35 wafers, dashed line 904) data. The root mean square error (RMSE) between the original D3y and the predicted D3y is Fig.16 is the best (RMSE = 2.12e-05), while Fig.15(RMSE = 2.83e-05) and Fig.14 (RMSE = 2.82e-05) is equivalent.

[0228] Fig.17 Depicts the absence of high-order overlay parameter correction and the Fig.14 The corresponding high-order parameter updates and Fig.15 and Fig.16 The corresponding overlay parameters predict the wafer map of the measured overlay.

[0229] In the wafer plot, the actual correction is applied on a dense (13*19 points per field) layout and the correction capability of each case is compared. The plot shows the mean plus 3 sigma overlay. Darker patterns indicate poor overlay performance. A baseline wafer map 1702 without high-order correction is shown. For conventional overlay, the wafer map 1704 consists of Fig.14 For the free LSTM without updates, the wafer map 1706 is generated by the same Fig.15 The wafer map 1708 is generated by updating the corresponding higher-order parameters of the drawing. Fig.16 The plot corresponding to the overlay higher-order parameter update is generated, Fig.16 The plots are for the LSTM predictions with and without updates. The latter has the best overlay correction performance, as shown by the lighter pattern.

[0230] Embodiments provide a method to include high order parameters into alignment corrections without measuring every wafer. Embodiments also improve methods for updating overlay measurements.

[0231] Alternatively or in addition to using the method for updating (higher order) parameters, the method of the present invention may be used to update parameters of a model that is used to update said parameters. For example, batch control of a semiconductor manufacturing process is typically based on the determination of process corrections using periodically measured process (related) parameters. In order to prevent too large fluctuations in the process corrections, an exponentially weighted moving average (EWMA) scheme is typically applied to a set of historical process parameter measurement data, which set includes not only the last obtained measurement of the process parameter. The EWMA scheme may have an associated set of weighting parameters, one of which is a so-called "smoothing constant" λ. The smoothing constant determines the extent to which the measured process parameter value is used for future process corrections, or alternatively, how far back in time the measured process parameter value is used to determine the current process correction. The EWMA scheme can be expressed as: Z i =λ·X i +(1-λ)·Z i-1 , where Z i-1For example, X may represent the process parameter value previously determined to be most suitable for a calibration run (usually a number of substrates) 'i-1'. i Being the process parameter as measured for run 'i', Zi is then predicted to represent the value of the process parameter that is most suitable for correcting run 'i' (the run following run 'i-1').

[0232] For example, more information on the use of EWMA in process control is provided in "Automated Process Control optimization to control Low Volume Products based on High Volume Products data), Proceedings of SPIE 5755, May 17, 2005, doi: 10.1117 / 12.598409, which is incorporated herein by reference in its entirety.

[0233] The value of the smoothing constant directly affects the predicted optimal process parameter used to determine the process correction for run 'i'. However, process fluctuations may occur which may affect the optimal value of the smoothing constant (or any other parameter associated with the model used to weight the historical process parameter data).

[0234] It is proposed to use a recursive neural network or LSTM as described in the previous embodiments to predict one or more values ​​of a first parameter associated with a semiconductor manufacturing process based on historical measurements of the first parameter. Instead of or in addition to determining a control scheme for subsequent operations of a process step in a semiconductor manufacturing process, it is proposed to update one or more parameters associated with a weighted model based on the predicted value of the first parameter. The one or more parameters may include a smoothing constant. For example, the smoothing constant may be determined based on the level of consistency between the predicted value of the first parameter using a recursive neural network (e.g., typically an LSTM) and the value of the first parameter predicted using a weighted model (e.g., typically an EWMA-based model). The weighting parameter (e.g., typically a smoothing constant) that gives the best consistency is selected. When the smoothing constant is used as a benchmark for LSTM-based predictions, regular reassessment of the quality of the smoothing constant ensures that the EWMA model is optimally configured at any point in time.

[0235] In one embodiment, a method for predicting a value of a first parameter associated with a semiconductor manufacturing process is disclosed, the method comprising: obtaining a first value of the first parameter based on measurements using a first sampling scheme; determining a predicted value of the first parameter based on the first value using a recurrent neural network; and determining a value of a parameter associated with a model used by a controller of the semiconductor manufacturing process based on the predicted value of the first parameter and the obtained first value of the first parameter.

[0236] In one embodiment, the determination of the previous embodiment is based on comparing the predicted value of the first parameter with a value of the first parameter obtained by applying the model to the obtained first value of the first parameter.

[0237] In one embodiment, the model is an exponentially weighted moving average model and the parameter is a smoothing constant associated with the exponentially weighted moving average model.

[0238] Other embodiments of aspects of the invention are disclosed in the following numbered clause list:

[0239] 23. A method for predicting a value of a first parameter associated with a semiconductor manufacturing process, the method comprising:

[0240] - obtaining a first value of a first parameter based on measurements using a first sampling scheme; and

[0241] -Determining a predicted value of the first parameter based on the first value using a recurrent neural network.

[0242] 24. The method of clause 23, further comprising: determining a control scheme for a subsequent operation of a process step in the semiconductor manufacturing process using a predicted value of a first parameter determined based on a first value obtained from a measurement at a first operation of the process step in the semiconductor manufacturing process.

[0243] 25. The method according to clause 24, further comprising:

[0244] - obtaining a second value of a second parameter based on measurements using a second sampling scheme that is less spatially dense and more frequent than the first sampling scheme; and - using the second value of the second parameter to determine a control scheme for subsequent operation of the process step.

[0245] 26. The method of clause 25, wherein the second value is obtained based on a measurement at a subsequent operation of the process step.

[0246] 27. A method according to clause 26, wherein the second value is obtained based on measurements performed on the same substrate supported on the same substrate stage, at which subsequent operations of the process step are performed by a control scheme determined using the second value.

[0247] 28. The method of any one of clauses 25 to 27, wherein the semiconductor manufacturing process is a batch process for patterning substrates, and wherein the measurement frequency of the first sampling scheme is every 5 to every 10 batches, and the measurement frequency of the second sampling scheme is once per batch.

[0248] 29. A method according to any one of clauses 25 to 28, wherein the semiconductor manufacturing process is a process of patterning a substrate using exposure fields, and wherein the spatial density of the first sampling scheme is 200 to 300 measurement points per field, and the spatial density of the second sampling scheme is 2 to 3 measurement points per field.

[0249] 30. A method according to any of clauses 23 to 29, further comprising obtaining subsequent values ​​of the first parameter based on measurements using the first sampling scheme, and using the subsequent values ​​to update the recurrent neural network.

[0250] 31. A method according to any of clauses 23 to 30, wherein the parameters comprise exposure magnification parameters and the process steps comprise photolithographic exposure.

[0251] 32. A method according to any one of clauses 23 to 31, wherein the recursive neural network comprises a long short-term memory network.

[0252] 33. A semiconductor manufacturing process comprising a method for predicting a value of a parameter associated with the semiconductor manufacturing process according to any of clauses 23 to 32.

[0253] 34. A lithographic apparatus comprising:

[0254] - an illumination system configured to provide a projection radiation beam;

[0255] - a support structure configured to support a patterning device configured to pattern the projection beam according to a desired pattern;

[0256] - a substrate table configured to hold a substrate;

[0257] - a projection system configured to project the patterned light beam onto a target portion of the substrate; and

[0258] - a processing unit, configured to:

[0259] The method according to any of clauses 23 to 32 predicts a value of a parameter associated with a semiconductor manufacturing process.

[0260] 35. A computer program product comprising machine-readable instructions for causing general purpose data processing apparatus to perform the steps of the method according to any of clauses 23 to 32.

[0261] 36. The method of any of clauses 23 to 32, further comprising determining a value of a parameter associated with a model used by a controller of a semiconductor manufacturing process based on the predicted value of the first parameter and a first value of the first parameter obtained by measurement using a first sampling scheme.

[0262] 37. A method according to clause 36, wherein the determination is based on a comparison of the predicted value of the first parameter with a value of the first parameter obtained by applying the model to the obtained first value of the first parameter.

[0263] 38. A method according to clause 36 or 37, wherein the model is an exponentially weighted moving average model and the parameter is a smoothing constant associated with the exponentially weighted moving average model.

[0264] 39. A method for configuring a semiconductor manufacturing process, the method comprising:

[0265] - obtaining a first value of a first parameter based on measurements associated with a first operation of a process step in a semiconductor manufacturing process and a first sampling scheme;

[0266] - based on the first value, determining a predicted value of the first parameter using a recurrent neural network; and

[0267] - using the predicted value of the first parameter to configure subsequent operation of a process step in the semiconductor manufacturing process.

[0268] 40. The method according to clause 39, further comprising:

[0269] - obtaining a second value of the second parameter based on measurements using a second sampling scheme, the second sampling scheme being less spatially dense than the first sampling scheme; and

[0270] - using the second value of the second parameter for configuring a subsequent operation of a process step in the semiconductor manufacturing process.

[0271] 41. The method of clause 40, wherein the second value is obtained based on a measurement at a subsequent operation of the process step.

[0272] 42. A method according to clause 41, wherein the second value is obtained based on measurements performed on the same substrate supported on the same substrate stage at which subsequent operations of the process step are performed, wherein the subsequent operations of the process step are configured using the second value.

[0273] 43. The method of clause 40, wherein the semiconductor manufacturing process is a batch process for patterning substrates, and wherein the measurement frequency of the first sampling scheme is every 5 to every 10 batches, and the measurement frequency of the second sampling scheme is once per batch.

[0274] 44. A method according to clause 40, wherein the semiconductor manufacturing process is a process of patterning a substrate using exposure fields, and wherein the spatial density of the first sampling scheme is 200 to 300 measurement points per exposure field, and the spatial density of the second sampling scheme is 2 to 3 measurement points per exposure field.

[0275] General Definition

[0276] Although specific reference may be made herein to the use of lithographic apparatus in the manufacture of ICs, it should be understood that the lithographic apparatus described herein may have other applications. Other possible applications include the manufacture of integrated optical systems, guidance and detection patterns for magnetic domain memories, flat panel displays, liquid crystal displays (LCDs), thin film magnetic heads, etc.

[0277] Although specific reference may be made herein to embodiments of the invention in the context of an inspection apparatus or a metrology apparatus, embodiments of the invention may be used in other apparatuses. Embodiments of the invention may form part of a mask inspection apparatus, a lithography apparatus, or any apparatus that measures or processes an object such as a wafer (or other substrate) or a mask (or other pattern forming device). It should also be noted that the term metrology apparatus or metrology system encompasses or may be replaced with the term inspection apparatus or inspection system. A metrology apparatus or inspection apparatus as disclosed herein may be used to detect defects on or within a substrate and / or defects in a structure on a substrate. In such an embodiment, for example, a feature of a structure on a substrate may be related to a defect in the structure, the absence of a particular portion of the structure, or the presence of an unwanted structure on the substrate.

[0278] Although specific reference is made to a "metrology device / tool / system" or an "inspection device / tool / system," these terms may refer to the same or similar types of tools, devices, or systems. For example, an inspection device or metrology device including embodiments of the present invention may be used to determine characteristics of a physical system, such as a structure on a substrate or wafer. For example, an inspection device or metrology device including embodiments of the present invention may be used to detect defects in a substrate or defects in a structure on a substrate or wafer. In such embodiments, the characteristics of the physical structure may be related to a defect in the structure, the absence of a particular portion of the structure, or the presence of an unwanted structure on the substrate or wafer.

[0279] Although specific reference may have been made above to the use of embodiments of the invention in the context of optical lithography, it will be appreciated that the invention is not limited to optical lithography and may be used in other applications, such as imprint lithography, where the context permits.

[0280] Although the targets or target structures (more generally, structures on a substrate) described above are metrology target structures specifically designed and formed for measurement purposes, in other embodiments, the characteristic of interest may be measured on one or more structures that are functional parts of a device formed on a substrate. Many devices have regular grating-like structures. The terms structure, target grating, and target structure as used herein do not necessarily require that the structure has been provided specifically for the measurement being performed. With respect to multi-sensitivity target embodiments, different product features may include many areas where the sensitivity changes (spacing changes, etc.). Further, the spacing p of the metrology target is close to the resolution limit of the optical system of the scatterometer, but may be much larger than the size of a typical product feature made by a lithography process in the target portion C. In practice, the lines and / or spaces of the overlaid grating within the target structure may be made to include smaller structures similar in size to the product features.

[0281] Although specific embodiments of the present invention have been described above, it should be appreciated that the present invention may be practiced in a manner other than that described. The above description is intended to be illustrative rather than restrictive. Therefore, it will be apparent to those skilled in the art that modifications may be made to the described invention without departing from the scope of the claims set forth below.

Claims

1. A method for configuring a semiconductor manufacturing process, the method comprising: - obtaining a first value of a first parameter based on a measurement associated with a first operation of a process step in said semiconductor manufacturing process; - determining a predicted value of the first parameter using the recurrent neural network based on using the first value as an input to the recurrent neural network; as well as - using said predicted value of said first parameter for configuring subsequent operation of said process step in said semiconductor manufacturing process.

2. The method of claim 1 , wherein the measurements associated with the first operation of the process step are associated with a first sampling scheme, and the method further comprises: - obtaining a second value of a second parameter based on measurements using a second sampling scheme, the second sampling scheme being less spatially dense than the first sampling scheme; as well as - using said second value of said second parameter for configuring said subsequent operation of said process step in said semiconductor manufacturing process.

3. The method of claim 2, wherein the second value is obtained based on a measurement at the subsequent operation of the process step.

4. A method according to claim 3, wherein the second value is obtained based on measurements performed on the same substrate supported on the same substrate stage, at which the subsequent operation of the process step is performed, wherein the subsequent operation of the process step is configured using the second value.

5. The method of claim 2, wherein the semiconductor manufacturing process is a batch process for patterning a substrate, and wherein the measurement frequency of the first sampling scheme is every 5 to every 10 batches, and wherein the measurement frequency of the second sampling scheme is once per batch.

6. The method of claim 2, wherein the semiconductor manufacturing process is a process of patterning a substrate using exposure fields, and wherein the spatial density of the first sampling scheme is 200 to 300 measurement points per exposure field, and wherein the spatial density of the second sampling scheme is 2 to 3 measurement points per exposure field.

7. The method according to claim 1, further comprising: A subsequent value of the first parameter is obtained based on the measurement, and the recurrent neural network is updated using the subsequent value.

8. The method of claim 1, wherein the recurrent neural network comprises a long short-term memory network.

9. A semiconductor manufacturing process comprising a method for configuring a semiconductor manufacturing process according to the method of claim 1.

10. The method according to claim 1, further comprising: A value of a parameter associated with a model used by a controller of a semiconductor manufacturing process is determined based on the predicted value of the first parameter and the obtained first value of the first parameter.

11. The method of claim 10, wherein the determining is based on comparing the predicted value of the first parameter with a value of the first parameter obtained by applying the model to the obtained first value of the first parameter.

12. The method of claim 10, wherein the model is an exponentially weighted moving average model and the parameter is a smoothing constant associated with the exponentially weighted moving average model.

13. A computer program product comprising machine-readable instructions for causing a general purpose data processing apparatus to perform the following steps: - obtaining a first value of a first parameter based on a measurement associated with a first operation of a process step in a semiconductor manufacturing process; - determining a predicted value of the first parameter using the recurrent neural network based on using the first value as an input to the recurrent neural network; and - using said predicted value of said first parameter for configuring subsequent operation of said process step in said semiconductor manufacturing process.

14. The computer program product of claim 13, wherein the recurrent neural network comprises a long short-term memory network.

15. The computer program product of claim 13, further comprising machine-readable instructions for determining a value of a parameter associated with a model used by a controller of a semiconductor manufacturing process, wherein determining the value of the parameter is based on the predicted value of the first parameter and the obtained first value of the first parameter.

16. The computer program product of claim 15, wherein determining the value of the parameter associated with the model used by the controller is based on comparing the predicted value of the first parameter with a value of the first parameter obtained by applying the model to the first value of the first parameter obtained.

17. The computer program product of claim 15, wherein the model is an exponentially weighted moving average model.

18. A computer program product comprising machine-readable instructions for causing a general purpose data processing apparatus to perform the following steps: - obtaining a first value of a first parameter based on measurements associated with a process step in a semiconductor manufacturing process; - determining a predicted value of the first parameter using the recurrent neural network based on using the first value as an input to the recurrent neural network; and - using said predicted value of said first parameter for configuring a controller, said controller being used for batch control of said semiconductor manufacturing process.

19. The computer program product of claim 18, wherein the step of configuring the controller comprises configuring an exponentially weighted moving average model, the exponentially weighted moving average model being used for the batch control of the semiconductor manufacturing process.

20. A method for configuring a semiconductor manufacturing process, the method comprising: - obtaining a first value of a first parameter based on measurements associated with a process step in said semiconductor manufacturing process; - determining a predicted value of the first parameter using the recurrent neural network based on using the first value as an input to the recurrent neural network; as well as - using the predicted value of the first parameter to configure an exponentially weighted moving average model, the exponentially weighted moving average model being used for batch control of the semiconductor manufacturing process.

Citation Information

Patent Citations

  • Methods for the preparation of organotin compounds

    EP0002019A1

  • Predicting a value of a semiconductor manufacturing process parameter

    EP3650939A1

  • Method and apparatus for inspection and metrology

    US20180067900A1

  • Lithographic apparatus and device manufacturing method

    US6952253B2

  • Methods and apparatus for measuring a property of a substrate

    WO2013092106A1