阅读说明：本技术 基于协方差矩阵自适应演化策略的光刻机匹配方法 (Photoetching machine matching method based on covariance matrix adaptive evolution strategy ) 是由 李思坤 陈俞光 唐明 王向朝 陈国栋 胡少博 于 2020-12-10 设计创作，主要内容包括：一种基于协方差矩阵自适应演化策略的光刻机匹配方法。本方法以空间像关键尺寸作为描述光刻机成像性能的参数,通过协方差矩阵自适应演化策略优化像素化描述的光刻机的照明光源,实现光刻间高精度匹配。本发明充分利用了协方差矩阵自适应演化策略在中等规模程度的复杂优化问题上的优势,改进了自由照明系统光刻机的匹配方法,提高了现有方法的匹配精度与效率。适用于具有自由照明系统的浸没式光刻机之间的匹配。(A lithography machine matching method based on a covariance matrix adaptive evolution strategy is disclosed. The method takes the key size of the space image as a parameter for describing the imaging performance of the photoetching machine, optimizes the illumination light source of the photoetching machine described in a pixelization manner through a covariance matrix adaptive evolution strategy, and realizes high-precision matching between photoetching. The method fully utilizes the advantages of the covariance matrix adaptive evolution strategy on the complex optimization problem of the medium scale degree, improves the matching method of the free illumination system photoetching machine, and improves the matching precision and efficiency of the existing method. The method is suitable for matching between immersion lithography machines with free illumination systems.)

1. A lithography machine matching method based on covariance matrix adaptive evolution strategy is characterized by comprising the following steps:

1) preparation work:

checking and adjusting the working state parameters of the reference photoetching machine and the photoetching machine to be matched so that the reference photoetching machine and the photoetching machine to be matched are in the optimal working state;

checking the working state of the glue spreading and developing machine, the working state of the CD detection system and the photoresist batch to ensure that the glue spreading and developing machine is in the optimal working state, and confirming that the photoresist batch is the same and the working state of the CD detection system is normal;

2) and (3) exposure verification:

adopting a one-dimensional through-pitch graphic mask or a two-dimensional graphic mask as test masks, wherein the number of the test masks is M;

adjusting the adjustable parameters of the reference photoetching machine and the photoetching machine to be matched to the same value, wherein the adjustable parameters comprise the shape of a light source, the numerical aperture of a projection objective and the wave aberration of the projection objective;

sequentially loading test masks to a reference photoetching machine and a photoetching machine to be matched for exposure and development, respectively measuring photoresist patterns CD of the photoresist patterns on a silicon wafer by using a CD detection system, and if the mean square difference value of the difference between the photoresist patterns CD generated by the exposure of the two photoetching machines is larger than the preset valueTarget value CD_RMSOr the maximum value of the difference between the photo-resist pattern CDs is larger than a preset maximum target value CD_MAXMatching the photoetching machine;

3) matching a photoetching machine:

reading a state file SFF of a reference photoetching machine, wherein the state file SFF comprises photoetching machine characteristic information such as Numerical Aperture (NA) of a photoetching machine projection objective, partial coherence factor of an illumination system, actually measured pupil distribution of the illumination system, exposure dose, defocusing amount, actually measured projection objective aberration, workpiece table inclination factor, mechanical vibration level of a photoetching machine optical system, laser bandwidth and the like;

space image intensity threshold T is carried out on photoetching simulation software according to a reference photoetching machine state file_rSetting and calculating threshold value T_rTest mask aerial image values of

Reading the state file of the photoetching machine to be matched, setting photoetching simulation software according to the state of the file, and setting the initial sampling step length (standard deviation) sigma_initAnd an evaluation threshold value F_s；

Optimizing a light source of the photoetching machine to be matched by utilizing a covariance adaptive evolution strategy: the reference photoetching machine measures the obtained light source pattern J_Ref(size N)_S×N_S) Performing coding calculation to generate initial coded light sourceThe coding mode is real number coding, and the coded light source is as follows:

wherein the content of the first and second substances,is the intensity value of the ith pixel point in the kth (k is 1, 2, …, N) sample light source, and emits lightThe brightness value of the area is 1, the brightness value of the non-luminous area is 0, and N is the sum of the number of the discrete light source point pixels; the specific steps of iterative calculation of the target light source graph of the photoetching machine to be matched are as follows:

(ii) decomposing the population of the g (g 1, 2, …)The k (k is 1, 2, …, lambda) individual decoding calculation corresponds to the corresponding pair of light source patterns

According to the pattern of the light sourceCalculating the threshold T by using photoetching simulation software which is set by the state file of the photoetching machine to be matched_rThe lower spatial image value is recorded asAnd calculating an evaluation functionThe formula is as follows:

② selecting the best (namely, the g is 1, 2, …) solution in the g generationMinimum) of individualsIts evaluation value is recorded asIf it isIf the evaluation value is less than the evaluation threshold value, the step (b) is carried out, otherwise, the step (c) is carried out;

third, according to the global step length (standard deviation) sigma of the g generation^(g)Collecting the g +1 generation populationWherein the individualFollowing a multivariate normal distribution

In the above formula, m^(g)Is the mean of the solution vectors of the g-th generation,is a multivariate normal distribution with a mean value of 0, C^(g)Is the solution vector of the g generationThe covariance matrix of (a);

substituting all the obtained individual solutions into an evaluation function to obtain corresponding evaluation valuesIt is sorted in the following order:

wherein the subscript i: λ denotes the ith position in λ individuals, and the weighted mean is calculated by taking the first μ ═ λ/2 individuals to update the mean, i.e.

Wherein ω is_iAre weights and

adaptive update step size (standard deviation) sigma^(g)：

Firstly, an accumulation step evolution path is learned from evolution information of a previous generation (g generation)

WhereinFor the g-th generation cumulative evolution path, c_σ＝(μ_eff+2)/(N+μ_eff+3) < 1 is the step accumulation constant,selecting a quality, C, for the effective variance^(g)Is a g-th generation covariance matrix;

then, according to the accumulated evolution path, the step length sigma is updated^(g)

Wherein d is_σFor the damping coefficient, approximately 1,for a normally distributed random vector norm expected length, the formula is as follows:

adaptive updating covariance matrix C^(g+1)：

Introduction of evolution pathAccumulating the inter-generation information during covariance updating, and constructing a Rank-1-Update updating mechanism:

utilizing the correlation relationship between the continuous evolution generation variation step sizes:

introducing a Rank-mu-Update updating mechanism, selecting the best mu individuals in the sub-population, and utilizing the mu individuals relative to the mean value m^(g)Such that the solution of the most recent algebra has a higher weight, thereby updating the covariance matrix:

updating the covariance matrix by combining two updating processes of Rank-mu-Update and Rank-1-Update, thereby not only fully utilizing the information between generations, but also fully utilizing the information of the whole population:

in the formula c_c，c_μ，c₁Respectively representing the learning rate or the accumulation constant of the covariance matrix, the Rank-mu-Update and the Rank-1-Update updating process;

seventhly, sampling based on multivariate normal distribution is carried out according to the updated step length, the covariance matrix and the corresponding evolution path, and lambda random samples are generatedReturning to the step I;

stopping iteration and marking the obtained individual as x_bestDecoding it to produce a light source shape J_bestAnd outputting the target light source shape as the target light source shape of the photoetching machine to be matched.

2. The lithography machine matching method based on the covariance matrix adaptive evolution strategy as claimed in claim 1, further comprising:

4) exposure verification

According to the solved target light source J of the photoetching machine to be matched_bestGenerating a parameter submenu of the photoetching machine to be matched, which needs to be adjusted, namely a nominal parameter set by a light source of the photoetching machine to be matched;

inputting the parameter submenu into a photoetching machine to be matched to adjust adjustable parameters, and loading a test mask by the photoetching machine to be matched to perform exposure and development;

measuring the photoresist pattern CD on the silicon wafer by using a CD detection system: if the mean square deviation value of the difference between the photoresist pattern CDs generated by the exposure of the two photoetching machines is smaller than the target value CD_RMSOr the maximum value of the difference between CDs is smaller than the target value CD_MAXIf the mask matching fails, the mask needs to be redesigned by performing Optical Proximity Correction (OPC) or light Source Mask Optimization (SMO).

3. The matching method of the lithography machine based on the covariance matrix adaptive evolution strategy as claimed in claim 1 or 2, wherein the working state parameters of the reference lithography machine and the lithography machine to be matched are projection objective cold aberration, illumination ellipticity, partial coherence factor of illumination, laser source stability, stray light level, illumination uniformity, mask stage and stage synchronization error information.

Technical Field

The invention relates to a free illumination system lithography machine, in particular to a lithography machine matching method based on a covariance matrix adaptive evolution strategy.

Background

Photolithography is the core process in integrated circuit manufacturing. To reduce the economic and time costs, lithographic processes are often developed on reference lithography machines. And applying a new photoetching process to a production line when the chip is produced in quantity. Generally, a production line has a plurality of lithography machines, which may have different models and suppliers, and the performances of the lithography machines in different models generally have obvious differences. Even the same model of lithography machine may cause large differences in lithography performance including imaging quality due to slight differences in hardware specifications. The performance of different photoetching machines can be different due to the difference, so that the process migration fails, and the production efficiency and the yield of products are influenced. In order to realize the quick transfer of the photoetching process, expand the capacity and improve the chip manufacturing yield, matching of a photoetching machine is needed, and the imaging performance of the photoetching machine to be matched and the imaging performance of a reference photoetching machine are consistent as much as possible by adjusting the adjustable parameters of the photoetching machine (the photoetching machine to be matched) on a production line.

Common lithography machine matching techniques are Critical Dimension (CD) measurement based matching techniques, photoresist model based matching techniques, and optical model based matching techniques. In the former two technologies, CD data on a silicon wafer is needed to represent the imaging performance difference between two lithography machines, and sensitivity information of adjustable parameters for matching is calculated to realize lithography machine matching. To ensure matching accuracy requires repeated multiple measurements of the CD under multiple conditions (e.g., multiple illumination patterns, multiple mask patterns, multiple pattern cycles, multiple exposure doses, multiple defocus positions), respectively, which consumes a significant amount of lithography time and measurement time. The matching technology based on the optical model (prior art 1, Yuan He, Erik Byers, and Scatt Light et al, Simulation-based patterning using scanner calibration and design data to reduce calibration on CD metrology, Proc. SPIE.7640,764014(2010)) utilizes the optical model of the photoetching machine to carry out adjustable parameter sensitivity calculation, does not need to carry out time-consuming CD measurement, avoids the influence of measurement noise and the calibration error of the photoresist model on the matching precision, is accurate and quick, has higher matching precision when the photoetching machine is the main influencing factor causing the unmatched graphs, and is a common technology in the production of high-end chips. The adjustable parameters comprise light source, projection objective numerical aperture, projection objective wave aberration and the like. The light source has high degree of freedom and is a main parameter for adjustment in matching of the photoetching machine. The prior art optimizes the Light source by Newton method or least square method to achieve Matching of lithography machines (prior art 2, Yuan He, Alexander Serebryakov and Scott Light et al, A Study on the Automation of Scanner Matching, Proc. SPIE.7973,79731H (2013); prior art 3, Lesikun, Lespajie, Dynasty, Yangxi, Yangxing, lithography machine Matching method, CN 108170006A). With the use of a pixilated light source with higher degree of freedom, the realization of lithography machine matching based on illumination sensitivity calculation is not applicable any more, but is realized by calculating the change of a CD and adding or deleting pixels of the light source according to the change, the prior art adopts a genetic algorithm (prior art 4, Higger, Lisikun, the dynasty, a lithography machine matching method, CN109031892B) to code the pixilated light source into a chromosome, and the difference between the lithography machine to be matched and a reference lithography machine is reduced by continuously updating the light source information through the genetic algorithm. However, for the high-precision pixelized light source matching, when the sampling precision of the light source is increased, the search space expanded by the genetic algorithm is exponentially multiplied, the search time is greatly increased, and the optimization efficiency is seriously influenced. And in some cases genetic algorithms tend to be locally optimal rather than globally optimal, adversely affecting the accuracy of the match.

Disclosure of Invention

The invention provides a photoetching machine performance matching method aiming at a photoetching machine with a free illumination system. The method carries out coding sampling on a pixilated light source, and continuously searches for an optimal solution in a sampling-updating-resampling mode through a covariance matrix self-adaptive evolution strategy, so that the difference of the photoetching performance between a photoetching machine to be matched and a reference photoetching machine is reduced. Aiming at the characteristics of high freedom degree of a light source and large parameter scale of a free illumination system, the method fully utilizes the advantages of the method on the medium-scale complex optimization problem and improves the matching precision and efficiency of the photoetching machine.

The technical solution of the invention is as follows:

a lithography machine matching method based on Covariance Matrix adaptive Evolution Strategy (CMA-ES) comprises the following steps:

1) and (3) checking a photoetching machine and a gluing developing machine:

checking the states of the reference photoetching machine and the photoetching machine to be matched: the method comprises the photoetching machine characteristic information such as cold aberration of a projection objective, the ellipticity of illumination, partial coherence factors of illumination, the stability of a laser light source, the stray light level, the illumination uniformity, the synchronization error of a mask table and a workpiece table and the like. Checking and confirming that the parameters of the reference photoetching machine and the photoetching machine to be matched are correctly set, and if the parameters are inconsistent with the parameters specified in the specification, adjusting the parameters in time to ensure that the reference photoetching machine and the photoetching machine to be matched work normally and are in the optimal working state. And checking the working state of the gluing developing machine, the working state of the CD detection system and the photoresist batch to ensure that the gluing developing machine works normally and is in the optimal working state. And confirming that the photoresist batches are the same and the working state of the CD detection system is normal.

2) And (3) exposure verification:

one-dimensional through-pitch pattern masks or a part of mass production two-dimensional pattern masks screened in advance are used as test masks, and the number of the test masks is M. And adjusting the adjustable parameters of the reference photoetching machine and the photoetching machine to be matched to the same value, wherein the adjustable parameters comprise the shape of a light source, the numerical aperture of the projection objective and the wave aberration of the projection objective. Sequentially loading test masks to a reference photoetching machine and a photoetching machine to be matched for exposure and development, respectively measuring the CDs of the photoresist patterns on the silicon wafer by using a CD detection system, and if the mean square difference value of the difference between the CDs of the photoresist patterns generated by the exposure of the two photoetching machines is larger than a target value CD_RMSOr the maximum value of the difference between CDs is greater than the target value CD_MAXThen the lithography machine needs to be matched.

3) Matching a photoetching machine:

the state file (SFF) of the reference lithography machine is read. The state file comprises the Numerical Aperture (NA) of the projection objective of the photoetching machine, the partial coherence factor of the illumination system, the actually measured pupil distribution of the illumination system, the exposure dose, the defocusing amount, the actually measured aberration of the projection objective, the inclination factor of the workpiece table, the mechanical vibration level of the optical system of the photoetching machine,Laser bandwidth, etc. Setting an aerial image intensity threshold T_r. Setting the photoetching simulation software according to the reference photoetching machine state file, and calculating the threshold value T_rTest mask aerial image CD values ofAnd reading the state file of the photoetching machine to be matched, and setting photoetching simulation software according to the state file. Setting initial sampling step size (standard deviation) sigma_initAnd an evaluation threshold value F_s。

And optimizing the light source of the photoetching machine to be matched by utilizing a covariance adaptive evolution strategy. The reference photoetching machine measures the obtained light source pattern J_Ref(size N)_S×N_S) Performing coding calculation to generate initial coded light sourceThe coding mode is real number coding, and the coded light source is as follows:

wherein the content of the first and second substances,the intensity value of the ith pixel point in the kth (k is 1, 2, …, N) sample light source is shown, the luminance value of the light-emitting area is 1, the luminance value of the non-light-emitting area is 0, and N is the sum of the number of the discrete light source point pixels. The specific steps of iterative calculation of the target light source graph of the photoetching machine to be matched are as follows:

(ii) decomposing the population of the g (g 1, 2, …)The k (k is 1, 2, …, lambda) individual decoding calculation corresponds to the corresponding pair of light source patternsAccording to the pattern of the light sourceCalculating the threshold T by using photoetching simulation software which is set by the state file of the photoetching machine to be matched_rLower aerial image CD value, noted And calculating an evaluation functionThe calculation formula is as follows:

② selecting the best (namely, the g is 1, 2, …) solution in the g generationMinimum) of individualsIts evaluation value is recorded asIf it isIf the evaluation value is less than the evaluation threshold value, the step (b) is carried out, otherwise, the step (c) is carried out.

Third, according to the global step length (standard deviation) sigma of the g generation^(g)Collecting the g +1 generation populationIn which the individualShould follow a multivariate normal distribution

In the above formula, m^(g)Is the mean of the solution vectors of the g-th generation,is a multivariate normal distribution with a mean value of 0, C^(g)Is the solution vector of the g generationThe covariance matrix of (2).

Substituting all the obtained individual solutions into an evaluation function to obtain corresponding evaluation valuesIt is sorted in the following order:

wherein the subscript i: λ denotes the ith position in λ individuals, and the weighted mean is calculated by taking the first μ ═ λ/2 individuals to update the mean, i.e.

Wherein ω is_iAre weights andω_i＞0(i＝1，2，…，μ)。

adaptive update step size (standard deviation) sigma^(g). In order to realize the overall scaling of the covariance matrix and improve the convergence speed of the algorithm, the accumulation path length control independent of the covariance matrix update, namely step length update, needs to be introduced. Evolution information from the previous generation (g-th generation) firstLearning cumulative step evolution paths

WhereinFor the g-th generation cumulative evolution path, c_σ＝(μ_eff+2)/(N+μ_eff+3) < 1 is the step accumulation constant,selecting a quality, C, for the effective variance^(g)Is the g-th generation covariance matrix. Then, according to the accumulated evolution path, the step length sigma is updated^(g)

Wherein d is_σIs a damping coefficient, approximately 1;normally distributed random vector norm expected length:

and sixthly, adaptively updating the covariance matrix. It is important to be able to update the covariance matrix by fully utilizing a series of evolving intergenerative information, so by introducing an evolution pathIntergenerative information at cumulative covariance update:

thereby constructing a Rank-1-Update updating mechanism to fully utilize the correlation relationship between the continuously evolving generation variation step sizes:

meanwhile, in order to fully utilize effective information provided by a large population and improve the global search capability, the best mu individuals in the sub-population are selected by introducing a Rank-mu-Update updating mechanism, and the mu individuals are utilized to be relative to the mean value m^(g)Such that the solution of the most recent algebra has a higher weight, thereby updating the covariance matrix:

in conclusion, the covariance matrix is updated by combining two updating processes of Rank-mu-Update and Rank-1-Update, so that the information between generations is fully utilized, and the information of the whole population is fully utilized:

in the above formula c_c，c_μ，c₁Respectively representing the learning rate or the accumulation constant of the updating process of the covariance matrix, the Rank-mu-Update and the Rank-1-Update.

Seventhly, sampling based on multivariate normal distribution is carried out according to the updated step length, the covariance matrix and the corresponding evolution path, and lambda random samples are generatedAnd returning to the step I.

Stopping iteration and marking the obtained individual as x_bestDecoding it to produce a light source shape J_bestAs to be matchedAnd outputting the target light source shape of the photoetching machine.

4) Exposure verification

According to the solved target light source J of the photoetching machine to be matched_bestAnd generating a parameter submenu of the photoetching machine to be matched, which needs to be adjusted, namely the nominal parameters set by the light source of the photoetching machine to be matched. And inputting the parameter submenu into the photoetching machine to be matched to adjust the adjustable parameters. And loading a test mask to be matched with the photoetching machine for exposure and development. The CD of the photoresist pattern on the silicon wafer is measured using a CD detection system. If the mean square deviation value of the difference between the photoresist pattern CDs generated by the exposure of the two photoetching machines is smaller than the target value CD_RMSOr the maximum value of the difference between CDs is smaller than the target value CD_MAXIf the mask matching fails, the mask needs to be redesigned by performing Optical Proximity Correction (OPC) or light Source Mask Optimization (SMO).

Compared with the prior art, the method uses a covariance adaptive evolution strategy (CMA-ES) to carry out photoetching machine matching, and the CMA-ES adopts a preferred truncation selection strategy, so that the matching efficiency and precision are higher, premature convergence to a certain degree can be avoided, and the method is suitable for the matching problem of a pixelized light source.

Drawings

FIG. 1 is a flow chart of a method for matching performance of a lithography machine using the present invention.

FIG. 2 is an illumination source of a reference lithography machine used in an embodiment of the present invention.

FIG. 3 is a test mask and matching mask pattern employed by an embodiment of the present invention.

FIG. 4 is a diagram of a matching lithography machine pupil used in an embodiment of the present invention.

FIG. 5 shows an illumination source of a lithography machine to be matched after the present invention has been implemented.

FIG. 6 is an illumination source of a lithography machine to be matched after matching of the lithography machine by using a genetic algorithm.

FIG. 7 is a CD error of line empty patterns of different periods of a reference lithography machine and a lithography machine to be matched before and after matching.

Detailed Description

The present invention will be further described with reference to the following examples and drawings, but the scope of the present invention should not be limited by these examples.

In the present embodiment, referring to the intensity distribution of the illumination light source of the lithography machine, as shown in fig. 2, the luminance value of the white area is 1, the luminance value of the black area is 0, the light source grid is 101 × 101, and the number of effective light source points S in the pupil range is 8048. The adopted test mask and the matching mask are one-dimensional through-pitch line empty pattern masks shown in figure 3, the line width of a mask pattern is 45nm, the type is a binary mask, the transmissivity of a white area is 1, and the transmissivity of a black area is 0. The mask pattern has a period of 45 in total of 120nm, 140nm, 160nm, … and 1000nm, i.e., M is 45, and the horizontal line segment marked in the figure is the cross-sectional position of the computed aerial image CD. The reference photoetching machine and the photoetching machine to be matched are both in an immersion type, and the working wavelength of the photoetching machine is 193 nm. The numerical aperture of a projection objective of the photoetching machine is set to be 1.35, the refractive index of the immersion liquid is 1.44, and the scaling factor R is 0.25. The matching steps of the photoetching machine are as follows:

1) and (3) checking a photoetching machine and a gluing developing machine:

checking the states of the reference photoetching machine and the photoetching machine to be matched: the inspected portions include projection objective cold aberration, ellipticity of illumination, partial coherence factor of illumination, laser source stability, stray light level, illumination uniformity, mask stage workpiece stage synchronization error, and the like. Checking and confirming that the parameters of the reference photoetching machine and the photoetching machine to be matched are correctly set, and if the parameters are inconsistent with the parameters specified in the specification, adjusting the parameters in time to ensure that the reference photoetching machine and the photoetching machine to be matched work normally and are in the optimal working state; and checking the work flow of the glue spreading developing machine, the working state of the CD detection system and the photoresist batch to ensure that the glue spreading developing machine works normally and is in the optimal working state, the photoresist batch is the same, and the working state of the CD detection system is normal.

2) And (3) exposure verification:

adjusting the adjustable parameters of the reference photoetching machine and the photoetching machine to be matched to the same value, respectively loading the test masks for exposure and development, measuring the CD of the photoresist pattern on the silicon wafer by using a CD detection system, and carrying out exposure production by the two photoetching machinesThe difference between the CD of the raw photoresist pattern is shown as the RMS value of 1.3566nm in terms of CD_RMSFor example, if the standard of 1nm is out of the allowable range of the process, the current lithography machine to be matched needs to be matched, and the next step is performed.

3) Matching a photoetching machine:

reading the state file (SFF) of the reference photoetching machine, setting photoetching simulation software according to the state file of the reference photoetching machine, and calculating the threshold value T_rThe CD value of the space image of the test mask is recordedAnd reading the state file of the photoetching machine to be matched, and setting photoetching simulation software according to the state file. Setting initial sampling step size (standard deviation) sigma_init0.008 and a fitness threshold F_s＝0.07nm。

The reference photoetching machine measures the obtained light source pattern J_Ref(size N)_S×N_S) Performing coding calculation to generate initial coding light sourceThe coding mode is real number coding, and the coded light source is as follows:

wherein the content of the first and second substances,the method comprises the following steps of (1) obtaining an intensity value of an ith pixel point in a kth (k is 1, 2, …, N) sample light source, iteratively calculating a target light source graph of the photoetching machine to be matched, wherein the luminance value of a light-emitting area is 1, the luminance value of a non-light-emitting area is 0, and N is the sum of the number of discrete light source point pixels, and the method comprises the following specific steps:

(ii) breeding the g (g is 1, 2, …) th generationRespectively decoding and calculating corresponding light source patternsAccording to the figureCalculating threshold T by using photoetching simulation software which is set by a photoetching machine state file to be matched_rLower aerial image CD value, notedAnd calculating an evaluation functionThe calculation formula is as follows:

② calculating the individual with the minimum fitness in the k (k is 1, 2, …, lambda is more than or equal to 2) th generation solutionIts fitness is recorded asIf it isEntering step (b), otherwise entering step (c).

③ according to the step length sigma^(g)Collecting samples from the multivariate normal distribution of the kth generation solution

The number of child solutions of the first generation is generally λ ═ 4+ floor (3 × log (n)), where floor denotes rounding up.

Substituting all the solutions into an evaluation function to obtain corresponding evaluation valuesThey are sorted in the following order.

Taking the updated mean of the first mu-lambda/2 individuals (rounding down), and taking the super-linear weight omega_i＝log[max(i，λ/2)]-log(i)。

Self-adapting step length updating. Calculating an evolution path from the accumulated information

Updating the step size σ^(g)：

Step-size integration constant set to c_σ＝(μ_eff+2)/(N+μ_eff+3) effective variance selection quality of

Sixthly, updating the adaptive covariance matrix. Calculating an evolution path from the accumulated information

Updating the covariance matrix of the g +1 th generation of samples:

the Rank-1-Update learning rate is set to

The Rank-mu-Update learning rate is set as

The covariance accumulation constant is set to

Seventhly, according to the new step length, the covariance matrix and the corresponding evolution path, carrying out a sampling process based on multivariate normal distribution to generate lambda random samplesAnd returning to the step I.

Stopping iteration, and marking the obtained solution vector as x_bestDecoding it to produce a light source shape J_bestOutputting the target light source shape as a target light source shape of the photoetching machine to be matched; the matched light source graph is shown in FIG. 5, the CD error obtained by simulation of the photoetching simulation software is shown in FIG. 7, and the CD error after matching is not more than 0.17 nm. For the same reference photoetching machine and photoetching machine to be matched, the time consumption is reduced from 1043.2 seconds to 304.76 seconds and is reduced by about 70.8 percent when the reference photoetching machine and the pixelized light source matching method based on the genetic algorithm (the prior art 4, the council Jie, Liskun, the dynasty and the photoetching machine matching method, CN109031892B), and the mean square root value of the CD error is reduced from 0.1894nm to 0.0695nm and is reduced by about 63.3 percent.

4) Exposure verification

Finally according to matchingAnd the subsequent lighting system light source graph produces a parameter submenu of the photoetching machine to be matched, and the parameter submenu is input into the photoetching machine to be matched so as to adjust corresponding adjustable parameters. After adjustment, the photoetching machine to be matched exposes the test mask, and the CD of the photoresist pattern on the silicon wafer is measured. The difference between the measured CD and the reference lithography CD is smaller than the CD_RMSAnd completing matching of the photoetching machine.

The above description is only one specific embodiment of the present invention, and the embodiment is only used to illustrate the technical solution of the present invention and not to limit the present invention. The technical solutions available to those skilled in the art through logical analysis, reasoning or limited experiments according to the concepts of the present invention are all within the scope of the present invention.