CN117744035A - Multispectral optical key dimension acquisition method, multispectral optical key dimension acquisition equipment and multispectral optical key dimension acquisition medium - Google Patents
Multispectral optical key dimension acquisition method, multispectral optical key dimension acquisition equipment and multispectral optical key dimension acquisition medium Download PDFInfo
- Publication number
- CN117744035A CN117744035A CN202410187283.9A CN202410187283A CN117744035A CN 117744035 A CN117744035 A CN 117744035A CN 202410187283 A CN202410187283 A CN 202410187283A CN 117744035 A CN117744035 A CN 117744035A
- Authority
- CN
- China
- Prior art keywords
- spectrum
- data
- calculation
- regression analysis
- spectral data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 133
- 230000003287 optical effect Effects 0.000 title claims abstract description 48
- 238000001228 spectrum Methods 0.000 claims abstract description 94
- 230000008569 process Effects 0.000 claims abstract description 64
- 238000004364 calculation method Methods 0.000 claims abstract description 58
- 238000000611 regression analysis Methods 0.000 claims abstract description 39
- 238000004458 analytical method Methods 0.000 claims abstract description 15
- 230000003595 spectral effect Effects 0.000 claims description 28
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000010183 spectrum analysis Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 2
- 230000005540 biological transmission Effects 0.000 claims 1
- 230000001133 acceleration Effects 0.000 description 14
- 230000006870 function Effects 0.000 description 14
- 230000000694 effects Effects 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 10
- 239000010410 layer Substances 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 239000004065 semiconductor Substances 0.000 description 3
- 239000002356 single layer Substances 0.000 description 3
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- FFBHFFJDDLITSX-UHFFFAOYSA-N benzyl N-[2-hydroxy-4-(3-oxomorpholin-4-yl)phenyl]carbamate Chemical compound OC1=C(NC(=O)OCC2=CC=CC=C2)C=CC(=C1)N1CCOCC1=O FFBHFFJDDLITSX-UHFFFAOYSA-N 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 239000003990 capacitor Substances 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Spectrometry And Color Measurement (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a method, equipment and medium for acquiring multi-spectrum optical critical dimensions, which comprises the steps of firstly acquiring spectrum data, then inputting multiple copies of spectrum data into an analysis module for regression analysis to obtain the optical critical dimensions corresponding to each spectrum data, wherein the regression analysis comprises the following steps: and starting multi-process calculation, wherein the number of processes is smaller than or equal to the number of the spectrum data, then distributing the multiple spectrum data to each process, wherein each process receives at least one spectrum data and carries out task binding, each process respectively starts multiple threads, and sequentially carries out single spectrum regression analysis on the received multiple spectrum data. Through multi-layer parallel calculation, the calculation time of multispectral regression analysis can be shortened as far as possible within the allowable range of calculation power, the software performance is improved, and the application cost is reduced.
Description
Technical Field
The present invention relates to the field of optical critical dimensions, and in particular, to a method, an apparatus, and a medium for obtaining a multispectral optical critical dimension.
Background
The optical critical dimension (Optical Critical Dimension or OCD) technology is widely applied to the semiconductor manufacturing industry, is a non-contact measurement technology, can acquire the chip structure size and shape through a series of complex algorithms on the premise of not damaging a sample to be tested by utilizing an optical principle, and plays an important role in quality control, process optimization, cost reduction, yield improvement and the like of semiconductor products. Regression analysis is a common inverse problem solving method in the OCD technology, and is characterized in that an OCD measurement is converted into a numerical optimization problem by establishing a mathematical physical model between a theoretical spectrum and chip geometric parameters, and the degree of optimization is evaluated by comparing the goodness of fit of the theoretical spectrum and an actual measured spectrum, so that the OCD value of the chip physical model is solved. The mathematical physical model of the geometric structure and the theoretical spectrum is established, and the common methods include a strict coupled wave analysis method, a time domain finite difference method, a finite element method and the like.
The above-described techniques have matured in the semiconductor industry and methods for analyzing individual spectra and obtaining chip geometry using the techniques have been widely demonstrated. However, because the calculation amount of the optimized mathematical physical model is large, the time cost is always a concern, and particularly in some practical application scenarios, such as the testing of the preamble range of the generated spectrum library, the verification of the stability of the built chip geometric model, the cross verification of the quality and precision of the generated spectrum library, and the like, the multispectral regression analysis problem is involved.
In the existing multispectral calculation, a serial or partially parallel method is generally adopted. The serial method is to calculate each spectrum data in turn, the last spectrum calculation can be completed to calculate the next spectrum, the serial calculation is also carried out in the regression model calculation of each spectrum, and all calculation tasks are completed by one process and one thread. The partial parallel is to calculate each spectrum data in turn, and the last spectrum calculation is completed before the next spectrum calculation can be started, but parallel acceleration is used in a certain part of the single spectrum regression analysis. Although the calculation time can be shortened to a certain extent, the acceleration effect is limited. It can be seen that the time consumption of the multispectral problem will increase linearly compared to the single-spectrum problem. Considering the disadvantage that single spectrum calculation itself is time consuming, the excessive waiting time caused by the superposition effect greatly influences the software use experience.
Disclosure of Invention
In view of some or all of the problems in the prior art, a first aspect of the present invention provides a method for obtaining a multispectral optical critical dimension, including:
acquiring spectrum data;
and inputting the multiple spectral data into an analysis module for regression analysis to obtain optical key dimensions corresponding to the spectral data, wherein the regression analysis comprises:
starting multi-process calculation, wherein the number of processes is less than or equal to the number of parts of the spectrum data;
distributing the multiple spectral data to each process, wherein each process receives at least one spectral data and performs task binding;
and each process respectively starts multiple threads, and sequentially carries out single spectrum regression analysis on the received multiple spectrum data.
Further, the spectral data includes the light intensity and the phase change of the reflected light after the light is incident on the sample to be detected.
Further, initiating the multi-process calculation includes:
the multi-process is started through the information transfer interface (Message Passing Interface, MPI).
Further, multiple spectral data are distributed to each process by a modulo operation method.
Further, the single spectral regression analysis includes:
and inputting the spectrum data into a regression model, and obtaining the key optical size after multi-step iteration.
Further, the single spectrum regression analysis adopts a strict coupled wave analysis method.
Further, turning on multithreading includes:
dividing matrix calculation in single spectrum regression analysis into a plurality of sub-matrix calculations, and distributing each sub-matrix calculation to each thread to execute calculation;
and/or splitting the loop task in the single spectrum analysis into a plurality of subcycles, and distributing each subcycle to each thread to execute calculation.
Further, the acquiring method further includes:
the spectrum data is preprocessed and then input to an analysis module, wherein the preprocessing comprises data cleaning, noise filtering and data standardization.
Further, the acquiring method further includes:
pre-training a regression model to optimize model parameters, wherein the model parameters include: hyper-parameters of linear/nonlinear regression models, convergence criteria for goodness of fit of theoretical spectra to actual spectra, maximum number of iterations, and various dimensional ranges of optical critical dimensions.
A second aspect of the present invention provides an apparatus for obtaining a multispectral optical critical dimension for performing the method of obtaining a multispectral optical critical dimension as described above, the apparatus comprising:
a spectrum data acquisition module for acquiring spectrum data;
and an analysis module for enabling multi-process parallel regression analysis based on the spectral data to determine an optical critical dimension, wherein multi-thread parallel computation is enabled in each process.
Further, the apparatus includes a data processing module for preprocessing the acquired spectral data to identify anomalies or missing data, and/or to filter out interference from unwanted signals, and/or to scale the data to a specified range.
A third aspect of the invention provides a computer readable storage medium for acquiring multispectral optical critical dimensions, for storing a computer program which, when run on a processor, performs the method of acquiring multispectral optical critical dimensions as described above.
According to the method, the equipment and the medium for acquiring the multispectral optical key dimension, multispectral regression analysis is regarded as a total calculation task, the task is decomposed in a grading manner through corresponding parallel tools, the task is divided into a plurality of single-spectrum tasks from a higher level, the bottom calculation in each single-spectrum task is divided into a plurality of sub-tasks, parallel calculation is carried out on two levels, and finally, the result is collected and the total task is completed, so that the effects of multilayer parallelism and multilayer acceleration are achieved. By adopting the multi-layer parallel architecture, the calculation time of multispectral regression analysis can be shortened as far as possible within the allowable range of calculation power, the software performance is improved, and the application cost is reduced.
Drawings
To further clarify the above and other advantages and features of embodiments of the present invention, a more particular description of embodiments of the invention will be rendered by reference to the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. In the drawings, for clarity, the same or corresponding parts will be designated by the same or similar reference numerals.
FIG. 1 is a process diagram of a method for obtaining a multi-spectral optical critical dimension according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for obtaining a multi-spectral optical critical dimension according to an embodiment of the present invention;
fig. 3 is a schematic diagram showing a comparison of the open-loop numbers and the calculation time lengths of the thread numbers in a method for obtaining the multispectral optical key dimension according to an embodiment of the invention.
Detailed Description
In the following description, the present invention is described with reference to various embodiments. One skilled in the relevant art will recognize, however, that the embodiments may be practiced without one or more of the specific details, or with other alternative and/or additional methods or components. In other instances, well-known structures or operations are not shown or described in detail to avoid obscuring aspects of the invention. Similarly, for purposes of explanation, specific numbers and configurations are set forth in order to provide a thorough understanding of embodiments of the present invention. However, the invention is not limited to these specific details. Furthermore, it should be understood that the embodiments shown in the drawings are illustrative representations and are not necessarily drawn to scale.
Reference throughout this specification to "one embodiment" or "the embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment.
It should be noted that the embodiments of the present invention describe the steps of the method in a specific order, however, this is merely for the purpose of illustrating the specific embodiments, and not for limiting the order of the steps. In contrast, in different embodiments of the present invention, the sequence of each step may be adjusted according to the adjustment of the actual requirement.
In the present invention, the modules of the system according to the present invention may be implemented using software, hardware, firmware or a combination thereof. When implemented in software, the functions of the modules may be performed by a computer program flow, e.g., the modules may be implemented by code segments (e.g., code segments in a language such as C, C ++) stored in a storage device (e.g., hard disk, memory, etc.), which when executed by a processor, perform the corresponding functions of the modules. When a module is implemented in hardware, the functionality of the module may be implemented by providing corresponding hardware structures, such as by hardware programming of a programmable device, e.g., a Field Programmable Gate Array (FPGA), or by designing an Application Specific Integrated Circuit (ASIC) comprising a plurality of transistors, resistors, and capacitors, etc. When implemented in firmware, the functions of the module may be written in program code form in a read-only memory of the device, such as EPROM or EEPROM, and the corresponding functions of the module may be implemented when the program code is executed by a processor. In addition, some functions of the module may need to be implemented by separate hardware or by cooperation with the hardware, for example, a detection function is implemented by a corresponding sensor (e.g., a proximity sensor, an acceleration sensor, a gyroscope, etc.), a signal transmission function is implemented by a corresponding communication device (e.g., a bluetooth device, an infrared communication device, a baseband communication device, a Wi-Fi communication device, etc.), an output function is implemented by a corresponding output device (e.g., a display, a speaker, etc.), and so on.
In order to improve the calculation efficiency of multispectral regression analysis and further rapidly and accurately acquire the multispectral optical key size, the invention provides a multispectral optical key size acquisition method and multispectral optical key size acquisition equipment, and the multispectral regression analysis is performed by adopting a multilayer parallel architecture, so that the effects of multilayer parallel and multilayer acceleration are achieved.
The technical scheme of the invention is further described below with reference to the accompanying drawings of the embodiments.
FIG. 1 is a process diagram of a method for obtaining a multi-spectral optical critical dimension according to an embodiment of the present invention. In the embodiment of the invention, as shown in fig. 1, the multi-spectrum regression analysis is regarded as a total calculation task, a multi-process is started through a parallel tool, the total task is disassembled in a grading manner, a plurality of single-spectrum tasks are divided from a higher layer to be distributed to each process, then multi-thread calculation is started in each process, namely, the bottom calculation in each single-spectrum task is divided into a plurality of sub-tasks, parallel calculation is carried out on two layers, and finally, the result is collected and the total task is completed, so that the effects of multi-layer parallelism and multi-layer acceleration are achieved.
FIG. 2 is a flow chart of a method for obtaining a multispectral optical critical dimension according to an embodiment of the invention. As shown in fig. 2, a method for obtaining a multispectral optical critical dimension includes:
first, in step 201, spectral data is acquired. The method comprises the steps that spectrum data of a chip to be detected are obtained through a spectrum data obtaining module, and in one embodiment of the invention, the spectrum data comprise two groups of optical characteristic parameters alpha and beta, wherein the alpha and beta are calculated by the light intensity and phase change of reflected light;
next, at step 202, data is transmitted. The multiple spectral data are input into an analysis module. In order to improve the performance, in one embodiment of the present invention, the spectrum data is further preprocessed by a data preprocessing module before the data is transmitted. In one embodiment of the invention, the preprocessing includes data cleaning, noise filtering, and normalization. The data cleaning is used for identifying data abnormality or missing so as to ensure the quality of input data; noise filtering can reduce interference of bad signals; the numerical value can be scaled to a certain range by standardization, so that subsequent calculation is convenient;
next, at step 203, regression analysis is performed. And the analysis module carries out regression analysis based on the spectrum data to obtain the multispectral optical key size. As previously described, in embodiments of the present invention, the regression analysis employs multiple layers of parallel computing. Specifically, as shown in fig. 1, the regression analysis includes:
first, the overall task is disassembled, and task allocation is performed. And starting multi-process calculation, wherein the number of processes is smaller than or equal to the number of the spectrum data, and then distributing the multiple spectrum data to each process, wherein each process receives at least one spectrum data and performs task binding. In the step, a first parallel level is realized, wherein the multispectral total task is divided into a plurality of single-spectrum subtasks according to different spectrums, processes and the subtasks are mapped, each process corresponds to one single-spectrum calculation task and is an independent calculation unit which is responsible for executing all calculation processes of the current single-spectrum regression analysis, and the OCD value of the geometric model corresponding to the current spectrum is obtained, so that coarse-granularity parallelism is realized. Specifically, the multispectral total task can be considered as a task that includes N pieces of actual spectral data, including N sets of specific values of alpha_true and beta_true, which are stored in array form in computer memory. To facilitate task management and allocation, they may be named alpha_true_1, beta_true_1, alpha_true_2, beta_true_2, …, alpha_true_ N, beta _true_N in order. In one embodiment of the invention, a multi-process is started using an information delivery interface (Message Passing Interface, MPI) as a parallelization tool. The MPI library provides rich parallel tools, in one embodiment of the present invention, multiple processes can be started through the mpi_init, and the number of processes is controlled through the mpi_comm_size, so that the process number P can be set manually through the mpiexec instruction, and the default number of processes of the system is p=1, and then P processes are numbered as process 0, process 1. Meanwhile, after the MPI multi-process is started, the main process sends spectrum data to each process through an MPI_Send function, alpha_true_1 and beta_true_1 are sent to process 1, alpha_true_2 and beta_true_2 are sent to process 2, and so on until N spectrums are all distributed. In a single process, an MPI_Comm_rank function is used for acquiring an identification ID of a current process, and an MPI_Recv function is used for receiving spectrum data sent by a main process, so that the process and spectrum subtask are bound. Since the starting of the MPI process requires additional resource consumption and time loss, which may result in a decrease in the overall performance of the parallel architecture, in order to avoid excessive consumption of resources, in one embodiment of the present invention, the number of processes should be limited to a certain extent, which may be smaller than the total number of spectral subtasks. In one embodiment of the present invention, when the number of spectra is greater (N > > P), the method of modulo arithmetic is used to perform task allocation and process binding, that is, a plurality of single spectrum analysis tasks are sequentially executed in the same process, it should be understood that in other embodiments of the present invention, task allocation may be performed in other manners, and the principle is to ensure that the difference of calculation amounts between processes is not too large, and task allocation is as uniform as possible;
and, next, task secondary disassembly, single-process in-multithreading. And each process respectively starts multiple threads, and sequentially carries out single spectrum regression analysis on the received multiple spectrum data. The regression analysis of the single spectrum refers to searching OCD values which can be matched with measured spectrum data by using a regression model through a mathematical modeling method after spectrum data is acquired, and in the iteration process of the regression model, the corresponding theoretical spectrum is required to be calculated according to the OCD values of the current iteration, and then the error between the theoretical spectrum and the actual spectrum is calculated and is used as a standard for evaluating the goodness of fit. For example, an AI algorithm such as a rigorous coupled wave analysis method, a time domain finite difference method, a finite element method, or machine learning may be used. In one embodiment of the present invention, for the single spectrum regression analysis, it may be understood that a set of actual spectrums alpha_true and beta_true measured from the device end are input into a regression model, and after multiple steps of iteration, a set of OCD values is finally output. It can be seen that in the regression analysis of the single spectrum, there are still a lot of complex mathematical physical models and matrix operations, and the execution time of each process can be shortened by the multithreading technology. Based on the method, a second parallel level is realized through the step, a proper parallel tool is adopted to start multithreading, a large number of circulation or large-scale matrix operations in a single spectrum regression algorithm are further disassembled into a plurality of subtasks, the multithreading is started in a single process, and the parallel acceleration effect and the first parallel level are mutually overlapped, so that the calculation time is shortened to the greatest extent. That is, on the basis of the subtasks divided by the total tasks of the previous step, the subtasks are divided again, that is, the subtasks of the subtasks are divided more finely. Compared to the first level, the parallel tool at the thread level has more choices, and CUDA (hardware is NVIDIA GPU), openMP, pthread, TBB (Intel Threading Building Blocks) and the like can be selected according to different configurations and specific requirements of hardware/software/operating systems and the like. How to decompose the subtasks and which parallel tool to choose to implement the level of parallelism depends on multiple factors such as the choice of regression model, the positive problem solving algorithm of geometry and theoretical spectrum, the hardware/software configuration that can be invoked, programming language, the operating system of the computer, etc. In one embodiment of the present invention, when configured with an NVIDIA GPU graphics card, due to its great advantage in matrix operation, a large-scale matrix can be split into multiple small-scale submatrices and GPU acceleration matrix operation is invoked using CUDA programming language. In yet another embodiment of the present invention, when the CPU is used to execute the calculation task, the loop task with longer time consumption in the algorithm may be disassembled into several sub-loops, and the sub-loops are allocated to each thread to execute the calculation by using the tools such as openMP, pthread and Intel TBB, so as to obtain better acceleration effect. It should be understood that the problems of resource competition, memory read-write and the like which are common to the multithreading technology must be avoided, the error of the calculation result caused by the thread conflict is prevented, and the robustness of the algorithm is ensured. Similar to the process number setting, in one embodiment of the present invention, the process number should be limited to a reasonable range, and specific values can be obtained according to actual algorithm and calculation test. After determining a specific multithreading implementation, the single spectrum regression analysis calculation in the process can be performed, and the corresponding OCD value is obtained based on the actual spectrum data allocated to the single spectrum regression analysis calculation. In the embodiment of the invention, the naming method of the OCD value is unified with the spectrum data so as to collect the result and avoid confusion: OCD_1 is obtained according to alpha_true_1 and beta_true_1, OCD_2 is obtained according to alpha_true_2 and beta_true_2, and the like. After a single process has completed all its assigned computing tasks, the OCD data is sent to the master process using the mpi_send function. Furthermore, to improve performance, in one embodiment of the invention, the regression model is also pre-trained to optimize model parameters, wherein the model parameters include: hyper-parameters of a linear/nonlinear regression model, convergence criteria of theoretical spectrum and actual spectrum fitting goodness, maximum iteration times, various dimensional ranges of optical key dimensions and the like;
and, finally, at step 204, the results are collected. And collecting the calculation results of each process to complete the multispectral calculation task. And collecting the calculation results scattered in each process, namely the final result of the total task of the multispectral regression analysis. When the MPI is used, in the main process, the MPI_Recv function is used to collect the OCD calculation results sent by all other processes, and all the results are combined and post-processed to give the final multispectral OCD calculation result. And ending all parallel computing parts by using the MPI_Finalize function, releasing the resources occupied by the MPI, cleaning the memory and completing the whole multi-layer parallel architecture. In one embodiment of the present invention, the results are also visually presented to the user by means of visual drawing, chart presentation, etc. based on the total calculated results of the multispectral OCD.
The present invention also provides an apparatus for obtaining a multispectral optical critical dimension for performing the method for obtaining a multispectral optical critical dimension as described above, the apparatus comprising:
a spectrum data acquisition module for acquiring spectrum data;
and an analysis module for enabling multi-process parallel regression analysis based on the spectral data to determine an optical critical dimension, wherein multi-thread parallel computation is enabled in each process.
In one embodiment of the invention, the apparatus further comprises a data processing module for preprocessing the acquired spectral data to identify anomalies or missing data, and/or to filter out interference of unwanted signals, and/or to scale the data to a specified range.
The present invention also provides a computer readable storage medium for acquiring a multispectral optical critical dimension for storing a computer program which, when run on a processor, performs the method of acquiring a multispectral optical critical dimension as described above.
In order to fully explain the effect of the acquisition method, the acquisition method is tested by adopting actual data, wherein an MPI+openMP hybrid programming multi-layer parallel architecture is adopted, a regression algorithm adopts a nonlinear least square method, an internal geometric structure and a mathematical physical model of a theoretical spectrum adopt a strict coupled wave analysis method, and the test result is as follows:
when the total task spectrum number=100, if the single process is executed in a single thread, namely all the tasks are serial, the total time is >5000s; if the single-process multithreading is adopted, namely a single-layer parallel architecture, the time is shortened to about 1200s; if the multi-process multithreading is adopted, namely the multi-layer parallel architecture provided by the invention, the time can be further shortened to be less than 300s, and compared with the original algorithm, the time is shortened by tens of times.
In practical application scenarios, the total number of spectra is typically not too large. Limiting the total task spectrum number to be within 20, the test result is shown in fig. 3, and it can be seen that the multi-layer parallel architecture always has a good acceleration effect, and the calculation time is significantly short compared with the single-layer parallel architecture (the abscissa process number=1 in the figure). Meanwhile, on the premise of the same calculated amount, the numerical value and the acceleration effect of the process number and the thread number are not necessarily in direct proportion, and an optimal process/thread combination exists, so that the acceleration effect can be maximized. How to determine the optimal parallel parameter combination is obtained by combining with a specific problem test, and is not described in detail.
In summary, the method and the device for acquiring the multispectral optical critical dimension of the invention can ensure that the method and the device are better than serial or single-layer parallel in terms of calculation time consumption, the actually and finally achieved acceleration ratio is related to a plurality of aspects such as specific algorithm logic, hardware configuration, programming language, parallel strategy and the like, and the maximum acceleration ratio complies with the armda law.
While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example only, and not limitation. It will be apparent to those skilled in the relevant art that various combinations, modifications, and variations can be made therein without departing from the spirit and scope of the invention. Thus, the breadth and scope of the present invention as disclosed herein should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.
Claims (9)
1. The method for acquiring the multispectral optical key dimension is characterized by comprising the following steps:
pre-training a regression model to optimize model parameters, wherein the model parameters include: hyper-parameters of a linear/nonlinear regression model, convergence criteria of theoretical spectrum and actual spectrum fitting goodness, maximum iteration times, and various dimensional ranges of optical critical dimensions;
acquiring spectrum data;
and inputting the multiple spectral data into an analysis module for regression analysis to obtain optical key dimensions corresponding to the spectral data, wherein the regression analysis comprises:
starting multi-process calculation, wherein the number of processes is less than or equal to the number of parts of the spectrum data;
distributing the multiple spectral data to each process, wherein each process receives at least one spectral data and performs task binding;
and each process respectively starts multiple threads, and sequentially carries out single spectrum regression analysis on the received multiple spectral data, wherein the single spectrum regression analysis comprises inputting the spectral data into a regression model, obtaining a key optical size after multi-step iteration, and starting multiple threads comprises splitting matrix calculation in the single spectrum regression analysis into a plurality of sub-matrix calculation, distributing each sub-matrix calculation to each thread for calculation, and/or splitting a loop task in the single spectrum analysis into a plurality of sub-loops, and distributing each sub-loop to each thread for calculation.
2. The acquisition method of claim 1, wherein the spectral data includes a light intensity and a phase change of reflected light after the light is incident on the sample to be detected.
3. The acquisition method of claim 1, wherein starting the multi-process calculation comprises the steps of:
and starting a plurality of processes through the information transmission interface.
4. The acquisition method according to claim 1, wherein the plurality of spectral data are assigned to each process by a modulo operation method.
5. The method of claim 1, wherein the single spectral regression analysis employs rigorous coupled wave analysis.
6. The acquisition method of claim 1, further comprising the step of:
the spectrum data is preprocessed and then input to an analysis module, wherein the preprocessing comprises data cleaning, noise filtering and data standardization.
7. An apparatus for acquiring multispectral optical critical dimensions, configured to perform the acquisition method of any of claims 1 to 6, the apparatus comprising:
a spectral data acquisition module configured to acquire spectral data;
and an analysis module configured to enable multi-process parallel regression analysis based on the spectral data to determine an optical critical dimension, wherein each in-process enables multi-thread parallel computation.
8. The apparatus of claim 7, further comprising a data processing module configured to pre-process the acquired spectral data to identify anomalies or missing data, and/or to filter out interference from unwanted signals, and/or to scale the data to a specified range.
9. A computer-readable storage medium for acquiring multispectral optical critical dimensions, characterized in that a computer program is stored, which computer program, when run on a processor, performs the acquisition method according to any of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410187283.9A CN117744035B (en) | 2024-02-20 | 2024-02-20 | Multispectral optical key dimension acquisition method, multispectral optical key dimension acquisition equipment and multispectral optical key dimension acquisition medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410187283.9A CN117744035B (en) | 2024-02-20 | 2024-02-20 | Multispectral optical key dimension acquisition method, multispectral optical key dimension acquisition equipment and multispectral optical key dimension acquisition medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117744035A true CN117744035A (en) | 2024-03-22 |
| CN117744035B CN117744035B (en) | 2024-04-26 |
Family
ID=90281582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410187283.9A Active CN117744035B (en) | 2024-02-20 | 2024-02-20 | Multispectral optical key dimension acquisition method, multispectral optical key dimension acquisition equipment and multispectral optical key dimension acquisition medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117744035B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106157317A (en) * | 2016-07-21 | 2016-11-23 | 武汉大学 | The high-resolution remote sensing image fusion rules method guided based on dispersion tensor |
| CN113035735A (en) * | 2021-03-01 | 2021-06-25 | 长鑫存储技术有限公司 | Method, system, medium, and electronic device for measuring semiconductor structure |
| CN114820581A (en) * | 2022-05-26 | 2022-07-29 | 清华大学 | Axisymmetric optical imaging parallel simulation method and device |
| CN115406530A (en) * | 2021-05-26 | 2022-11-29 | 华为技术有限公司 | Spectrum measuring method and device |
-
2024
- 2024-02-20 CN CN202410187283.9A patent/CN117744035B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106157317A (en) * | 2016-07-21 | 2016-11-23 | 武汉大学 | The high-resolution remote sensing image fusion rules method guided based on dispersion tensor |
| CN113035735A (en) * | 2021-03-01 | 2021-06-25 | 长鑫存储技术有限公司 | Method, system, medium, and electronic device for measuring semiconductor structure |
| CN115406530A (en) * | 2021-05-26 | 2022-11-29 | 华为技术有限公司 | Spectrum measuring method and device |
| CN114820581A (en) * | 2022-05-26 | 2022-07-29 | 清华大学 | Axisymmetric optical imaging parallel simulation method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117744035B (en) | 2024-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111427681B (en) | Real-time task matching scheduling system and method based on resource monitoring in edge computing | |
| DE102019106669A1 (en) | METHOD AND ARRANGEMENTS FOR MANAGING STORAGE IN CASCADED NEURONAL NETWORKS | |
| US20100223213A1 (en) | System and method for parallelization of machine learning computing code | |
| US20110083125A1 (en) | Parallelization processing method, system and program | |
| US8843932B2 (en) | System and method for controlling excessive parallelism in multiprocessor systems | |
| US20170330078A1 (en) | Method and system for automated model building | |
| CN110866589B (en) | Operation method, device and framework of deep neural network model | |
| US20190114770A1 (en) | Systems and methods for detecting cancer metastasis using a neural network | |
| CN104657111A (en) | Parallel computing method and device | |
| CN111597243A (en) | Data warehouse-based abstract data loading method and system | |
| US20170249240A1 (en) | Automated test planning using test case relevancy | |
| WO2019016656A1 (en) | Process for the automatic generation of parallel code | |
| CN112765017A (en) | Data query performance test method and device based on MySQL database | |
| CN110377519B (en) | Performance capacity test method, device and equipment of big data system and storage medium | |
| CN113168364A (en) | Chip verification method and device | |
| Jordà et al. | cuConv: CUDA implementation of convolution for CNN inference | |
| CN117744035B (en) | Multispectral optical key dimension acquisition method, multispectral optical key dimension acquisition equipment and multispectral optical key dimension acquisition medium | |
| US11467827B1 (en) | Index space mapping using static code analysis | |
| Gao et al. | Resource-guided configuration space reduction for deep learning models | |
| Srivastava et al. | Performance and memory trade-offs of deep learning object detection in fast streaming high-definition images | |
| Hamanaka et al. | An exploration of state-of-the-art automation frameworks for FPGA-based DNN acceleration | |
| Byrd et al. | Reducing the run-time of MCMC programs by multithreading on SMP architectures | |
| van den Berg et al. | iDSL: Automated performance prediction and analysis of medical imaging systems | |
| Cruz et al. | Fast evaluation of segmentation quality with parallel computing | |
| Sofranac et al. | Accelerating domain propagation: An efficient gpu-parallel algorithm over sparse matrices |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |