Systems and methods for designing and creating photolithographic masks

By directly generating mask designs using a gradient optimizer-based method, the problem of generating small-pitch masks in matter wave lithography is solved, achieving efficient mask generation that adapts to complex circuit designs and reducing computational resources and time.

CN123125541APending Publication Date: 2026-11-17REIS LITHOGRAPHY TECHNOLOGIES
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202580016126.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-02-20
Filing Date
2025-02-17
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to generate masks for matter wave lithography, especially those with small-pitch holes, and machine learning methods require significant computational resources and training data, making them ill-suited to the variability of complex circuit designs.

Method used

A gradient optimizer-based method is used to directly generate mask designs from the desired pattern without training, reducing computational resources and time, and is applicable to various lithography systems.

Benefits of technology

It achieves efficient generation of 3D physical lithography masks, applicable to different types of lithography systems, reducing computational resources and generation time, and adapting to complex circuit designs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123125541A_ABST
    Figure CN123125541A_ABST
Patent Text Reader

Abstract

A computer-implemented method and data processing system for designing and creating a lithographic mask (4) configured to be disposed between a source (2) having a beam (21) and a target (5), wherein the target has a predefined desired pattern (51), wherein the method includes: creating an initial mask template (41) having mask template properties (401), repeating until an intermediate error (62) is an acceptable error (60); predicting an intermediate pattern (52) generated in the target (5) by the mask template (42), determining an intermediate error (62) of the intermediate pattern (52) relative to the desired pattern (51), and improving the mask template properties (401) of the mask template (42) based on the error (60).
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of photolithography, and more particularly to the design and / or creation of photolithographic masks in a microlithography system for use in, for example, the production of integrated circuits. Background Technology

[0002] Integrated circuit manufacturing is currently based on pattern generation using photolithography, a mask-based process: an enlarged version or a portion of the desired wafer pattern is printed onto a substrate (photomask). The pattern on the mask can be distorted due to aberrations and other effects. Light (photons) is transmitted through or reflected off the photomask using optical components (refractive or reflective lenses), creating a reduced image of the pattern on a silicon wafer. The silicon wafer is coated with a photosensitive material (resist), which reacts with the photons to create a permanent imprint of the image within the resist. This resist is then used as a physical mask in subsequent manufacturing steps. The two main light sources used in the semiconductor industry today are deep ultraviolet (DUV) light with a wavelength of 193 nm and extreme ultraviolet (EUV) light with a wavelength of 13.5 nm.

[0003] As an alternative to photolithography, some published literature proposes using particle beams instead, because the high energy of photons limits the feature size achievable by EUV to approximately 6 nm. This is because the secondary electrons generated by the photon can travel several nm before inducing the reaction.

[0004] In practice, the generated or desired pattern is determined by the circuit design of the manufacturing project (such as, for example, integrated circuits). Therefore, a mask needs to be produced based on the desired pattern. This is often referred to as a reverse engineering problem.

[0005] Most particle-based lithography is based on grid-based binary holography, in which the openings are all the same size and are placed on the film at positions defined by a regular, square grid structure.

[0006] Nesse, T., Simonsen, I., Holst, B.: Nanometer-Resolution Mask Lithography with Matter Waves: Near-Field Binary Holography, Phys. Rev. Applied 11, 024009 (2019), https: / / doi.org / 10.1103 / PhysRevApplied.11.024009. This paper reveals a theoretical study on grid-based binary holography, which uses small masks with square and hexagonal grid structures. The distance between the mask and the pattern plane is only 40 μm, and it has extremely small holes of less than 1 nm.

[0007] While optical lithography uses well-known methods for producing photomasks with opaque and transparent regions, particle-based lithography, such as metastable atoms, can use solid masks with holes. The following describes some developments and mature technologies in mask production in this field.

[0008] Producing physical masks for manipulating metastable atoms using diffraction is a well-established technique.

[0009] Fujita et al., in their paper *Manipulation of anatomic beam by a computer-generated hologram* (Nature 380 691 (1996)), revealed the use of a mask made of solid material to manipulate atoms to generate specific aperiodic patterns. Here, a mask with a 300 nm diameter through hole was fabricated in a 100 nm thick SiN film. The mask was fabricated by first coating a resist layer onto the film, then using an electron beam to write the desired mask pattern into the resist. The through hole was subsequently created in the film using CF4 plasma etching. This mask was used to manipulate a metastable neon atomic beam.

[0010] S.D. Eder et al., in their paper "Focusing of a neutral helium beam with a photon-sieve structure," Phys. Rev. A 91, 043608 (2015), revealed the manipulation of a ground-state atomic beam using a so-called atomic sieve fabricated through a similar process. The mask consisted of 31,991 holes with diameters ranging from 1,840 to 150 nm.

[0011] Y. Yang et al., Nanostructured-membrane electronphase plates, Ultramicroscopy 217 113053 (2020) used electron beam lithography and reactive ion etching to pattern SiN films with a thickness as low as 10 nm into holes with a spacing as low as approximately 25 nm and 50 nm.

[0012] MG Stanford et al., Advanced nanoscale patterning and material synthesis with gas field helium and neon ion beams, J. Vac. Sci. Tech. B 35 030802 (2017), describe a different technique in which a helium ion beam performs so-called direct writing to create through-holes (pores) smaller than 10 nm in a film.

[0013] Previous theories of binary holography for scalar waves used a grid of apertures of equal size as the basis for the mask, employing system dimensions and wavelengths that satisfy the well-known Fraunhofer approximation. In the Fraunhofer approximation, the inverse problem is simplified to a Fourier transform of the desired pattern, which produces the required mask. Sampling the mask according to the Nyquist-Shannon sampling theorem provides its mode bandwidth and thus the minimum possible spacing. In one dimension, this can be accomplished through analysis as described in previous techniques. This can also be applied to matter waves if the apertures in the mask are too large for the dispersive forces to be negligible.

[0014] However, the above approach is not applicable to matter waves and masks with small apertures requiring fine spacing, because the dispersion force interaction between particles and diffracting objects induces a complex phase distribution of the matter wave, which is not explained in the mask generation theory of prior art. Furthermore, the Fourier transform method becomes ineffective when the Fraunhofer approximation is no longer satisfied.

[0015] Machine learning has been proposed as a way to solve the inverse problem of masks with small holes.

[0016] Fiedler, J., Palau, A., Osestad, E., Parviainen, P., Holst, B.: Realistic mask generation for matter-wave lithography via machine learning. ARXIV.ORG, Cornell University Library, July 15, 2022. https: / / arxiv.org / abs / 2207.08723 This paper proposes using machine learning to generate 1D masks for metastable atoms. A deep neural architecture is trained to produce an initial approximation of the mask. This approximation is then used to generate an initial population of an optimized genetic algorithm that can converge to arbitrary precision.

[0017] Previous mask generation methods could only be used to reverse engineer 1D patterns with a specific dimension, i.e., the dimension in which the neural network was trained. If patterns with different resolutions or extensions must be reverse engineered, the network must be trained again, a process that requires significant computational resources.

[0018] Furthermore, circuit designs are often quite complex and varied. Therefore, it is difficult to train a neural network to handle all the different designs, even if it has the same dimensions as the trained neural network. Summary of the Invention

[0019] One objective of this invention is to solve problems associated with prior art.

[0020] Specifically, this invention discloses an improved method for generating a mask based on a desired pattern from a design process. This mask design can then be used to produce a three-dimensional physical lithography mask for generating a realistic pattern in a target, which can be, for example, a wafer with a resist layer.

[0021] Previous mask generation techniques relied on large amounts of training data, facilitated by the fact that all patterns are obtained from Fraunhofer bodies. According to the present invention, no training is required. Instead, the method and system according to the invention take any desired pattern as input. A gradient-based optimizer is used directly for each new pattern. This means there are no dimensional constraints when creating the mask. Furthermore, computational resources and mask generation time can be significantly reduced.

[0022] Although this invention only reveals the prior art of creating masks by reverse engineering in particle lithography in Fraunhofer bodies, it can be applied to all types of lithography systems, including photonic, particle-based, and scalar wave-based systems, as well as any physical volume-based system of this kind.

[0023] Two-dimensional patterns of finite length are needed to encode information and logic gates in circuit designs. This invention can be used to obtain any practical two-dimensional pattern used in novel circuit designs.

[0024] The present invention, which addresses the above-mentioned problems, is a computer-implemented method and system for creating photolithographic masks according to the independent claims. Attached Figure Description

[0025] Figure 1 An example of a photolithographic pattern generation system using a mask 4 is schematically illustrated, wherein the mask has been created using the system and method according to the invention. The design process of the mask is illustrated in a simplified sketch above the photolithographic pattern generation system.

[0026] Figure 2 A flowchart illustrates the creation of a photomask according to an embodiment of the present invention.

[0027] Figure 3 The upper left corner displays an instance of the desired pattern 51, and the lower part displays the accepted mask template 43 and the accepted binary mask template 44. The upper part displays the corresponding intermediate patterns 52 of the accepted mask template 43 and the accepted binary mask template 44. Therefore, the intermediate pattern 52 shown here is the expected result of masking the target 5 with the same properties as the accepted mask template 43 or the accepted binary mask template 44 to produce the pattern 53. The corresponding system parameters for this setting are listed in... Figure 3 The bottom left corner.

[0028] It can be seen that the binarying of the accepted mask template may result in a slight decrease in the quality of the intermediate pattern.

[0029] Figure 4 This is a close-up view of the accepted binary mask template 44, where the black dots correspond to binary region states open to the beam used in the photolithography process. In the case of a particle beam, the binary region states would correspond to region 3, which creates the holes or openings in the mask 4. The grid lines are for illustrative purposes only.

[0030] Figure 5This is a summary diagram illustrating some mask template properties of the initial mask template 41, mask template 42, accepted mask template 43, accepted binary mask template 44, and physical mask 4. In this case, region state 4a is randomly generated in the initial mask template 41 and iteratively improved through gradient descent in mask template state 42, as indicated by the circular arrows, until mask template 42 generates a pattern sufficiently similar to the desired pattern, where mask template 42 is referred to as accepted mask template 43. To generate physical mask 4, accepted mask template 43 is binaryized into accepted binary mask template 44, which serves as the design template for generating physical mask 4. Region position 4b in the mask indicates the position of region 3 corresponding to region state 4a.

[0031] In optical systems, Figure 3 The black dots can represent transparent areas within an originally opaque mask. As an alternative to binary representation, the translucency of areas within the mask can be altered based on the region state of the accepted mask template 43. Detailed Implementation

[0032] In the following description, various examples and embodiments of the invention are set forth to provide those skilled in the art with a more thorough understanding of the invention. The specific details described in the context of the various embodiments and with reference to the accompanying drawings should not be construed as limiting. Rather, the scope of the invention is defined in the appended claims.

[0033] The embodiments described below are numbered. Furthermore, dependent embodiments defined relative to the numbered embodiments are described. Unless otherwise expressly stated, any embodiment that can be combined with one or more numbered embodiments can also be directly combined with any dependent embodiment of the referenced numbered embodiment.

[0034] Initially, and prior to mask generation, a desired pattern 51 for the mask generation process should be available. The desired pattern is typically the result of a design process (such as integrated circuit design), where the desired pattern may represent an area to be etched into a substrate (such as a silicon wafer) to obtain a circuit layer after etching the resist on the substrate.

[0035] Figure 3 This shows an example of this desired pattern 51 for illustrative purposes.

[0036] Many parameters may be related to the mask design. These parameters may include, for example, the write field of the desired pattern on the patterned target, the distance between the source and the mask, the distance between the mask and the patterned target, the source type and wavelength, the source beam width, the mask height, width, and thickness, etc. These parameters are referred to as system parameters in this document.

[0037] Used to generate such Figure 3 An example of system parameters for generating the pattern "LACE LITHOGRAPHY" with a pixel resolution of 35 nm can be given as follows: Mask size: 39.28 μm Mask thickness: 10 nm Pore ​​radius: 22 nm L1: 0.5 m L2: 0.000275 m Distance between the centers of the regions: 70 nm.

[0038] In a specific embodiment EM1-1, the present invention is a computer-implemented method for creating a photolithographic mask 4, which is configured to be disposed between a source 2 having a beam 21 and a target 5, wherein the target has a predefined desired pattern 51.

[0039] The method includes; Create an initial mask template 41 with multiple mask template properties 401. Starting from the initial mask template, repeat until the intermediate error of 62 is an acceptable error of 60; a - Predict the intermediate pattern 52 generated in the target 5 by the mask template 42. b - Determine the intermediate error 62 of the intermediate pattern 52 relative to the desired pattern 51, and c - Improve the mask template properties 401 of mask template 42 based on error 60.

[0040] EM1-2. A computer implementation method as in EM1-1, wherein the mask template property 401 includes a region state 42a, wherein the region state indicates whether a corresponding region 3 in the mask 4 should be open, closed, or partially open to allow the bundle 21 to pass through.

[0041] EM1-3. A computer implementation method as in EM1-2, wherein the region state 4a is represented by a numerical value selected from a predefined set of numbers.

[0042] EM1-4. A computer implementation method as in EM1-3, wherein the boundary values ​​of the set represent the respective open and closed regions 3 in the mask 4.

[0043] The set of numbers can be, for example, a column of decimal digits or integers.

[0044] If decimal numbers are used, they can be represented as floating-point numbers.

[0045] EM1-5. A computer-implemented method as described in any of EM1-2 to EM1-4, wherein a region state 4a with an open value indicates that the region has a through hole, and a closed state indicates that the region is solid. As previously indicated, a solid mask with a hole is used in particle lithography (such as atomic beam lithography).

[0046] EM1-6. A computer-implemented method as described in any of EM1-2 to EM1-4, wherein a region state 4a with an open value indicates that the region is transparent, and a closed state indicates that the region is opaque to the beam 21. The region state may represent transparent and opaque regions used in optical lithography.

[0047] EM2-1. A computer implementation method of any one of EM1-2 to EM1-6, wherein the masking template properties include the region position 42b of each region state 42a, wherein the region position defines the physical position 4b of region 3 in mask 4.

[0048] EM2-2. The computer implementation method of EM2-1, wherein the region location 4b is represented by a set of coordinates.

[0049] EM3-1. A computer implementation method as in any of EM1-1 to EM2-2, wherein only a single initial mask template 41 is created.

[0050] EM3-2. A computer implementation method as described in any of EM1-1 to EM3-1, wherein the value of region state 42a is randomly generated for the initial masking template 41.

[0051] EM3-3. A computer implementation method as in any of EM1-1 to EM3-1, wherein the value of region state 42a is equal for the initial masking template 41.

[0052] EM3-4. A computer-implemented method as described in any of EM1-1 to EM3-3, wherein the step of creating an initial mask template includes using parameters representing any one of the width, height, and thickness of mask 4 or a writable region of the mask.

[0053] EM3-5. A computer implementation method as described in any of EM2-1 to EM3-4, wherein the number of region position 42b is a function calculated to represent any one of the width, height, and thickness of mask 4 or a writable region of mask 4.

[0054] EM3-6. A computer implementation method as described in any of EM2-1 to EM3-5, wherein the width of region 3 is predefined.

[0055] EM3-7. A computer-implemented method as described in any of EM1-1 to EM3-6, wherein the initial single mask template 41 is untrained. That is, the region state 42a is not determined during the initial training process.

[0056] EM4-1. In any of the computer implementations of EM1-1 to EM3-7, the step of predicting the intermediate pattern 52 is performed by numerically calculating at least some contributions from the mask template property 401. This numerical calculation may include numerical integration.

[0057] EM4-2. A computer implementation method as in EM4-1, wherein the step of predicting the intermediate pattern 52 is performed by numerically calculating the region state 42a value of the mask template property 401. This numerical calculation may include numerical integration.

[0058] EM4-3. A computer-based implementation method as in EM4-2, wherein the intermediate pattern 52 is obtained by calculating the intensity I through a self-masking template 42, where I = |SUM(C_h)| W_h)|^2, where W_h is the complex intensity generated by opening each individual hole across the entire screen, and C_h is the region state value within the boundary of the set of numbers used.

[0059] EM4-4. A computer-implemented method as described in any of EM4-1 to EM4-3, comprising changing the size of region 3. The maximum size of this region is determined by the mesh of the mask defined by system parameters.

[0060] EM5-1. A computer-implemented method as described in any of EM1-1 to EM4-4, wherein the step of creating the initial mask template 41 includes; The values ​​of the region state 42a are preprocessed using a differentiable function to compress them into a predefined interval, where the boundaries of the interval represent the open and closed states of the cell.

[0061] EM5-2. A computer implementation method similar to EM5-1, wherein the preprocessing is performed by a sigmoid function.

[0062] EM5-3. A computer implementation method similar to EM5-2, wherein the value of region state 42a is multiplied by a number less than 1 before processing by the S-shaped function to map a number closer to the set boundary.

[0063] EM5-4. A computer implementation method as in EM5-3, wherein the boundaries of the set are 0 and 1 respectively.

[0064] EM5-5. A computer implementation method as in EM5-4, wherein 1 represents a fully open cell and 0 contains a fully closed cell, or vice versa.

[0065] EM6-1. A computer-implemented method as described in any of EM1-1 to EM5-5, wherein the step of determining the intermediate error 62 includes; The intermediate error is determined by a differentiable error function.

[0066] EM6-2. A computer implementation method as in EM6-1, wherein the variation of the intermediate error 62 is associated with the variation of the masking template properties 401.

[0067] EM6-3. The computer implementation method of EM6-1 or EM6-2, wherein the intermediate error is determined by the following formula; The intermediate error 62 = Loss(I(σ(x)) – P), where Loss is the loss function, I is the numerical calculation process, x is the mask template property and P is the desired pattern 52.

[0068] EM6-4. A computer implementation method as in EM6-3, wherein the loss function is any one of mean squared error, mean absolute error, and binary cross entropy.

[0069] EM6-5. A computer implementation method such as EM6-3 or EM6-4, wherein the loss function penalizes the values ​​of region states 42a that are close to the middle of the set and forces these values ​​to be closer to the boundaries of the set. For example, if some region state values ​​are 0.4356734 and 0.5784563, the new values ​​may be 0.2000000 and 0.8000000, respectively.

[0070] EM6-6. The computer implementation method of EM6-5, wherein the intermediate error is determined by the following formula; Intermediate error 62 = Loss(I(σ_T(x) - 0.5)) 2 – P), where σ_T is a low-temperature S-type function.

[0071] EM7-1. A computer-implemented method as described in any one of EM1-1 to EM6-6, comprising: Assign a binary number to each region state 42a of the accepted mask template 43.

[0072] EM7-2. A computer implementation method as in EM7-1, wherein the state values ​​of such regions are rounded to the nearest binary number.

[0073] EM8-1. A computer implementation method as described in any of EM1-1 to EM7-2, wherein the masking template 42 includes a matrix containing the region states 42a.

[0074] EM8-2. A computer implementation method as in EM8-1, wherein the matrix is ​​two-dimensional.

[0075] EM8-3. A computer implementation method such as EM8-1 or 8-2, wherein the dimensions of the matrix are predefined.

[0076] EM8-4. A computer implementation method as described in any of EM8-1 to EM8-3, wherein the desired pattern 51 has a predefined feature size, and the dimension of the matrix is ​​a function of the feature size.

[0077] EM8-5. A computer implementation method as described in any of EM1-1 to EM7-2, wherein the masking template 42 includes a dictionary structure containing the region states 4a.

[0078] EM8-6. A computer implementation method as in EM8-5, wherein the dictionary structure includes a region location 42b corresponding to the region state 42a.

[0079] EM9-1. A method as described in any of the above embodiments; comprising generating a photolithographic mask 4 having a mask template property 401 of an accepted mask template 43, the accepted mask template 43 being similar to a mask template 42.

[0080] Since the methods and systems of the present invention can be used for different types of lithography and optics, masks can be produced using different techniques. While optical lithography uses well-known methods for producing masks with opaque and transparent areas, particle-based lithography (such as metastable atom lithography) can use perforated solid masks produced as explained in the references in the preceding paragraphs.

[0081] The present invention may also be described as in other independent embodiments, such as: ED1. A data processing system comprising means for performing any one of EM1-1 to EM9-1.

[0082] ED2: A computer program comprising instructions that, when executed by a computer, cause the computer to perform any one of EM1-1 to EM9-1.

[0083] ED3: A computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform any one of EM1-1 to EM9-1.

[0084] ED4: A computer-based implementation method for creating a photolithographic mask design, comprising the steps of any one of EM1-1 to EM9-1.

[0085] A specific embodiment will now be described with reference to the accompanying drawings.

[0086] Figure 1 A specific embodiment of a photolithographic pattern generation system 1 is schematically illustrated, wherein a mask 4 created according to the present invention can be applied to obtain a desired pattern 51. System 1 includes an incident particle beam 21 from a particle source 2 and a mask 4 disposed within the incident particle beam 21 between the particle source 2 and a pattern target 5. The generated pattern 53 on the pattern target is produced by particles diffracted by the mask 4. The source distance L1 from the particle source 2 to the mask 4 and the pattern target distance L2 from the mask 4 to the pattern target 5 are indicated, respectively. Furthermore, the mask 4 includes a through-hole 3.

[0087] The generated pattern 53 may differ slightly from the desired pattern 51. For example, the generated pattern may have lower contrast in certain areas than the desired pattern, or it may have features such as those along... Figure 1 The small irregularities at the edges of the pattern indicate some noise. However, as long as the generated pattern 53 is within the predefined tolerances, the minor differences inherent in all types of lithography are acceptable.

[0088] Given the desired pattern 51 and system parameters, mask design and production can be carried out. Figure 2 The diagram illustrates how the process may be performed in one embodiment. The initial step is computer-implemented and includes generating an initial mask template 41, which in this embodiment may be a two-dimensional random matrix of floating-point numbers in a lattice, where the lattice values ​​are region state 42a values.

[0089] The dimension of a random matrix depends on the system parameters.

[0090] Then, each random floating-point number (i.e., region state) in the initial mask template 41 is compressed or normalized to the interval [0,1], where 1 can represent a fully open hole in the corresponding region 3 of the mask 4, and 0 can represent a solid or sealed hole. Grid values ​​(i.e., states between 0 and 1) represent partially open regions. However, since holes cannot be partially open, the grid values ​​need to be refined until a mask pattern consisting only of closed and open values ​​is found to obtain the desired pattern.

[0091] Compression is achieved using a sigmoid function, where x is a random floating-point number. To map numbers closer to 0 and 1, a dedicated sigmoid function σ(x / T) with a temperature parameter T can be used.

[0092] Other types of functions can be used for compression, as long as they are differentiable.

[0093] Therefore, at this stage, the matrix (which we will call mask template 42 for clarity) contains floating-point numbers in the range [0, 1].

[0094] Although these numbers are decimal numbers between 0 and 1, rather than binary, the cells in the mask template 42 can still be processed in a manner similar to a binary matrix.

[0095] To determine the intermediate pattern 52, the intensity I from the masking template 42 is calculated, where I = |SUM(C_h)| / (C_h) W_h)|^2, where W_h is the complex intensity produced by opening each individual hole across the entire screen, and C_h is the region state value within the boundary of the set of numbers used.

[0096] Then, by determining the amount of error given by Loss(I(σ(x)) – P), the intermediate pattern 52 is compared with the desired pattern 51, where Loss is the loss function, I is the numerical calculation process, x is the masking parameter and P is the desired pattern 52.

[0097] The loss function has a negative gradient in the direction of minimum loss.

[0098] Given the gradient of the loss function, the masking property (i.e., the value of the region state 42a) can be updated by applying gradient descent methods, such as gradient descent, Newton's method, and quasi-Newton methods (including BFGS, conjugate gradient, global optimization methods using gradient descent, etc.).

[0099] The updated mask is again a mask template 42, in which the intermediate pattern 52 can be determined and iteratively improved by repeating the above steps until the difference between the intermediate pattern 52 and the desired pattern 51 is within a predefined limit, and an accepted mask template 43 is obtained. Figure 1 In the diagram, circular arrows are used to indicate gradient descent iterations. Furthermore, the question mark between the expected pattern 51 and the intermediate pattern 52 indicates that iterations continue until the difference or error is within acceptable limits.

[0100] In cases where the mask has through holes, the cell values ​​of the acceptable mask need to be binaryed before generation, because only binary values ​​are meaningful in this case.

[0101] Once this iterative process is complete, the accepted mask template 43 can be binarized by assigning 0 or 1 to each decimal number represented by a floating-point number in the matrix unit. This binarization will introduce errors, and if the binarized accepted mask template is integrated, the pattern will not be exactly the same as the non-binarized accepted mask template.

[0102] This error can be minimized by modifying the function that maps the mask values ​​to numbers between 0 and 1. This function can be designed so that for most real-world inputs, the output is either 0 or 1, for example, a sigmoid function at low temperatures. This will modify the loss function described above.

[0103] An example of a binary term could be: Intermediate error 62 = Loss(I(σ_T(x) - 0.5)) 2 – P), where σ_T is an S-shaped function with a temperature parameter, making most of its values ​​close to 0 or 1 (corresponding to the low temperature parameter).

[0104] The loss function now penalizes numbers close to 0.5, forcing them to become 0 or 1.

[0105] Alternatively, the threshold used for binarization can be changed from 0.5 to any other threshold. The optimal threshold will be the one that produces a pattern after binarization that reduces the final error between the desired pattern and the pattern generated by the binarized mask.

[0106] In the exemplary embodiments, various features and details are shown in combination. The fact that several features are described for a particular instance should not be construed as implying that such features must be included together in all embodiments of the invention. Rather, features described with reference to different embodiments should not be construed as mutually exclusive. As will readily apparent to those skilled in the art, the inventors have contemplated any subset of features described herein that are not explicitly interdependent and such embodiments are intended to be part of the disclosure. However, explicitly describing all such embodiments does not contribute to understanding the principles of the invention; therefore, for simplicity or brevity, certain arrangements of features have been omitted.

Claims

1. A computer-implemented method for creating a photolithographic mask (4), said photolithographic mask (4) being configured to be disposed between a source (2) having a beam (21) and a target (5), wherein, The target has a predefined desired pattern (51), wherein the method includes: - Create an initial mask template (41) with mask template properties (401). - Repeat until the intermediate error (62) is an acceptable error (60). - Predict the intermediate pattern (52) generated in the target (5) by the mask template (42). - Determine the intermediate error (62) of the intermediate pattern (52) relative to the desired pattern (51), and - Improve the masking template properties (401) of the masking template (42) based on the error (60).

2. The computer-implemented method as described in claim 1, wherein, The mask template properties (420) include a region state (42a), wherein the region state indicates whether a corresponding region (3) in the mask (4) should be open, closed or partially open to allow the bundle (21) to pass through.

3. The computer-implemented method as described in claim 2, wherein, The region state (4a) is represented by a numerical value selected from a predefined set of numbers.

4. The computer-implemented method as described in claim 3, wherein, The boundary values ​​of the set of numbers represent the corresponding open and closed regions (3) in the mask (4).

5. The computer-implemented method as described in claim 4, wherein, The mask template properties include the region position (42b) of each region state (42a), wherein the region position defines the physical position (4b) of the region (3) in the mask (4).

6. The computer-implemented method as described in any one of claims 1 to 5, wherein, The step of predicting the intermediate pattern (52) is performed by numerically calculating at least some of the contributions from the mask template properties (400).

7. The computer-implemented method as described in any one of claims 1 to 6, wherein, The steps for creating the initial mask template (41) include: - The value of the region state (42a) is preprocessed using a differentiable function to compress the value of the region state (42a) into a predefined interval, wherein the boundary of the interval represents the open and closed states of the cell.

8. The computer-implemented method as described in any one of claims 1 to 7, wherein, The steps for determining the intermediate error (62) include: - The intermediate error is determined by a differentiable error function.

9. The computer-implemented method as described in claim 8, wherein, The intermediate error is determined by the following formula: The intermediate error (62) = Loss(I(σ(x))-P), where Loss is the loss function, I is the numerical calculation process, x is the mask template property and P is the desired pattern (52).

10. The computer-implemented method according to any one of claims 1 to 9, comprising: - Assign a binary number to each region state (42a) of the accepted mask template (43).

11. The method of claim 10, comprising generating the photolithographic mask (4) having mask properties (400) of the accepted mask template (43), the accepted mask template (43) being similar to the mask template (42).

12. A data processing system comprising means for performing the method as described in any one of claims 1 to 10.

13. A computer program comprising instructions that, when executed by a computer, cause the computer to perform the method as claimed in any one of claims 1 to 10.

14. A computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform the method as claimed in any one of claims 1 to 10.

15. A computer-based method for creating a photolithographic mask design, comprising the method as described in any one of claims 1 to 10.