TWI299520B - Mix-and-match control for lithography overlay - Google Patents

Description

1299520 IX. Description of the Invention: [Technical Field of the Invention] The present invention relates to a method of mixing and matching control of lithographic overlays, and more particularly to a method of mixing and matching control of lithographic overlays. [Prior Art] The lithography system is mainly composed of a light source system, a lens system, a mask system, an alignment system, and a wafer stage system. In these subsystems, the factors that may cause stacking errors are: 1) The light source system is a light source required to generate lithography by using a mercury lamp. As the number of times of use increases, the intensity of the mercury lamp source may be degraded. In the case of insufficient exposure, there will be a phenomenon after lithography. The problem of deformation, (2) the lens system is composed of each lens group, and if the positioning deviation occurs in the positioning of the lens, it will affect the refraction of the light source. In addition, if the lens surface profile is not good, it will cause the light source to scatter, and these may be the original purpose of circular deformation. The current surface of the axis has a fairly rainy quality. There are still physical limits in the production. As the key dimensions of the graphics are not small, the requirements for the wheel surface of the county will become more and more stringent. (3) The reticle stage and the wafer stage are transported in the reticle. In the case of wafers, as the number of shipments increases, the mechanical components continue to be mixed, which may result in a stage phenomenon and a position of no _determination bit, resulting in a positional deviation of the upper circuit pattern stack 5. In addition to the stacking error sources that may occur in the aforementioned lithography systems, there are also stochastic factors such as S-factors, measurement equipment, personnel operations, and wafer surface states that may cause stacking errors. Therefore, under the consideration of process stability and yield improvement, the attribution and compensation of the error source become indispensable. In addition, if the lithography equipment can be brought to the same state by appropriate adjustment, the equipment can be replaced with each other, and the degree of freedom of the overall production schedule can be increased. The Photolithography Process plays a role in the transfer of circuit graphics in the semiconductor manufacturing process. Through the transfer and stacking of structural layers, the final required electronic components are formed, so the quality of the lithography process is good or bad. The size and graphical correctness of the critical dimensions (CriticalDimensi()n, eD) that can be made. In the lithography process, an "alignment," step is usually used to ensure that the front and rear structural layer patterns can indeed fall to the original desired position, thereby reducing the overlay error of the graphics when stacked. However, each lithography device In the actual operation process, due to external environmental factors, wafer conditions or random coma generated by the lithography equipment itself, different types of nuisance errors may occur between individual lithography equipments, and in order to reduce The scheduling behavior of the working time enables the stacking error of the front layer circuit pattern to accumulate to the manufacturing process of the next layer circuit pattern, and the lithography blending and matching main meaning is to expect the structural pattern produced by different lithography devices. It can maintain the original expected rectification effect without being affected by the stacking error accumulation after stacking. One of the ways to improve Mix-and_ Match can start from improving the state similarity between lithography devices' from unavoidable In the error of raising the pair, by adjusting the equipment by 1299520, the result of the combination of the front and back layers with high similarity of the lanes will reduce the stacking error of the front and back layers. Poor, this can also increase the utilization rate of lithography equipment and the degree of freedom of scheduling (Degree of freedQm). Currently in the yery Large Seale (VLSI) (4) lithography, its aligning device lychee Laser and COXChairge Coupled Diode), etc., and the alignment of the technology side, the surface, then the two pairs of marks on the wafer cutting track "·,,,,,匚" as in Figure A Show that 'the difference between the overlapped mark left by the front _ layer and the current four pairs of mark φ is "Q,", after measuring the displacement of the two marks by image processing technology, such as the first B In order to achieve accurate lithography, process engineers usually establish a man-pair error model based on some empirical rules and four pairs of historical data. In modeling, the error is generally divided into: Intrafidd error. And Interfield error and other two causes of Intrafield error, mainly from the lens system and the reticle system caused by the error source, and the cause of the Μ Γ Γ dirty error is mainly from the reticle system and the wafer Between Source caused by the difference in the Analysis of the error, (iv), respectively, to explore a secret shop

Intrafield and Interfidd errors, as shown in the first figure, the (χ, γ) coordinate system uses the wafer as the coordinate origin, and the y) lion is based on the center of the exposure field. In terms of Intrafield error, the overlap error between them mainly includes translation, rotation, expansion (Expansi〇n), reticle tilt 1299520 • (Trapezoid), wedge (Wedge) distortion, lens Distortion, and • Interfield error mainly includes: translation, rotation (Rotation), expansion (Expansion), hip (B〇w), etc., such as the first D picture for the various # pairs of errors. The lithography stack mixes and matches the effects of overlay errors designed to reduce the quotation from different lithography devices, enabling lithography devices to increase device utilization and graphics yield. Therefore, it is necessary to analyze the characteristics of the individual lithography system by stacking the error model, and then compare and search the similarity between devices, thereby avoiding the difference in the error characteristics between the devices. Causes accumulation of errors. At present, in the parameter estimation of the overlay error model, the most commonly used method is the Least Square Method (LSM). Usually, the lithography process is finished, and the wafer is sent to the measuring device to measure the overlay mark. The stacking error is then returned to the error coefficients in the mathematical model of the overlay error by the LSM method, which is used as the reference for the overlay error compensation in the next lithography. However, there are many reasons for the error caused by the whole pair. The error between the device and the device is different in the error (10) plane or the error of the correction. According to its physical meaning, it can be distinguished as the uncontrollable error term of the secret difference. The item contains the translation, rotation and scaling of the graphic. These (4) errors and _ deformations can be described by the number of miscellaneous types. Can not be sloppy, the rest can not be described by mathematical molds _ shape deformation or (4) counties, such as lens surface! 29952 〇 变形 deformation degree will directly affect the degree of scattering of the light source, resulting in the subsequent lithography The graphic is distorted. Therefore, in the process of estimating the error parameters of the stack, the LSM may not accurately estimate the size of the stack error model coefficients under the influence of these uncontrollable error term factors. SUMMARY OF THE INVENTION In the background of the invention of Li Yu, in order to meet the needs of certain interests in the industry, the present invention provides a method for mixing and matching control of lithographic overlays, which can be used to solve the above-mentioned mixing of conventional lithographic overlays. And the method of matching control failed to achieve the target. It is an object of the present invention to provide a method of mixing and matching control of lithographic overlays. Then describe. The format of this paragraph is "inventive text". [Embodiment] The direction of the present invention as discussed herein is a method of mixing and matching control of a lithographic overlay. In order to thoroughly understand the present invention, detailed steps and compositions thereof will be presented in the following description. Obviously, the practice of the present invention is not limited to the specific details familiar to those skilled in the art of mixing and matching control methods. On the other hand, well-known components or steps are not described in detail to avoid unnecessarily limiting the invention. The preferred embodiment of the present invention will be described in detail below. However, in addition to these detailed descriptions, the present invention can also be widely applied to other real estates, and the fine money of the present invention is limited to the following patent scope. Prevail. Z〇ne-Ching Lin and Wen-Jang Wu (1999) added the effects of high-order lens deformation and nonlinear term parameters in the stack error correction model, and used multiple linear regression methods to analyze the overlay error and adjust it. The mathematical model for the Intrafield stack-to-error is defined as follows: β fic = Tfic- Rj,y + Μβχ + Τ^χ2 ^ Tyxxy + w^y2 (11) + ^3χΧ(χ2 + y2) + ϋ5χΧ{χ2 + y2) 2 or fy ^ TJy ^ Rfy^ M fyy + T^y^Txyxy + WfyX2 {χ 2) + D3yy(x2 + y2)+ D5yy(x2 + y2f λ; in the above formula; y represents the position of the exposure field, The subscript Guangχ^ respectively indicates the influence of the Intrafield error source in the X and y directions, and the heart represents the total stack error caused by the axis. ^^ and 7^ represent the exposure field on the χ and J axes, respectively. The translation coefficient is usually caused by the positioning accuracy of the reticle stage. ^ and \ represent the rotation angle coefficient of the exposure field, which is caused by the rotation of the reticle. Μ and μ Y1 fy represent the amplification factor of the exposure field, and the distance between the reticle and the lens. The resulting r, ^ and the tilt coefficient of the reticle are composed of 1299520 which is not perpendicular to the plane of the reticle, and % represents the wedge distortion coefficient, which is caused by the deviation of the center of the optical lens. ^ ^ and D5y are the lens deformation coefficients. It is mainly caused by the symmetry of the filter symmetry. The mathematical model of the county source is defined as follows: dwx - RWXY + MmX +BmY2 dm -Tm - RmX +M + ΒψγΧ2 (1-3) (1-4) ; The meaning and F respectively represent the position of the exposure field in the wafer coordinate system, and the lower 'W, Z, F points are thin. The county source and the bribe face in the direction of the y, y direction, the U axis mteifleid _ =: represents the translation coefficient of the wafer stage, moved by the wafer platform

With the rotation angle coefficient of ~ and V as the crystal 11, caused by the rotation ~ 盥 曰 A 曰 曰 曰 载 载 载 载 曰 曰 曰 曰 曰 产 产 产 产 由 由 由 由 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台 平台The crystal_f curved platform is caused by unevenness. After considering these times =, the total error of the stacking can be obtained. · 钟绿差11 1299520 dx+χ = ^wx + and ^; =^WX ~ ^WX ^ + WX ^ + ^wx^2 + fx ^R^y + Mβχ + τχχχ2 ^Tyxxy + w^y2 + D3jcx(x2 + y2) + D5xx(x2 + y2)2 dY + y = ^WY + ^ fy = Tm - + + ^wY^2 + ^Rfyx + Mfyy + Tyyy2 ^ + wfyx2 + D3yy(x2 + y2) + D5yy(x2 + y2)2 In the actual process, in addition to the above mentioned stacking error sources, In addition, there are quite a few factors that will affect the error of the overlay, so most of the reasons that may cause the error of the overlay are sorted and classified as follows: The factors related to the source of the error variation of the stack are related to the error, vibration, and cleanliness of the individual measurement equipment. Wafer wafer thermal deformation: lithography system itself lens thermal deformation, lens group positioning error, mercury lamp intensity degradation, wafer stage positioning, reticle stage positioning 12 1299520 ' % Process, uneven distribution of the film, the CMP polishing non-uniformity, uneven photoresist coating, soft baking temperature and time

The present invention takes all of the aforementioned factors into consideration, and thus the formulas (1-5) and (1-6) are redefined as follows:

Άχ+Χ ^Tx+x^RW^Y + MmX +MΑχ + Κβ, γ + Γχ +χ ( 1.7 ) (where χ = + k )

Dr+y ^TY^y^RmX + MmY + Mfyy + Rficx + rY^y ( 1 - 8 ) (of which

In the above formula, the stacking error is divided into controllable and uncontrollable error terms according to the device parameter adjustability, wherein the controllable error term coefficients include, A", /W, listening, slave, %, and /^, It is generated by the clamping error of the wafer stage and the photomask, and the uncontrollable error term, in addition to the general random error, there are also errors in the (1_5) and (1_6) modes that are not adjustable by the shadow system itself. For example, wedge distortion and lens deformation, etc. The compensation of the controllable error can be compensated by the relevant control parameters, and the correction of the uncontrollable error can be eliminated by the error source, but also by the lithography system. The controllable parameter items adjust their effects to a minimum. Using the overlay error model to create a 13 1299520 • lithography system feature will facilitate the mixing and analysis of subsequent lithography system groups. There are many factors in the lithography equipment that generate stacking errors, and most of them will affect the geometric relationship of the circuit pattern. For example, the unevenness of the wafer surface will cause the difference in thickness when the photoresist is applied, and the light will arrive when the lithography is caused. The post-refraction of the wafer surface produces pattern distortion. In the process of describing these uncontrollable error terms through the overlay error model, it is often difficult to have a clear and consistent mathematical model to describe the geometric relationship. Out, thus causing the LSM to make a partial geometric relationship omission in the stacking error estimation, so that the estimated stacking error is poor, and another new stacking error is generated in the subsequent compensation process, which is called “ Model error. Therefore, in order to avoid the above model error, one embodiment of the present invention, as shown in the second figure, can obtain a linear transfer function from the position parameter 21 according to the stack error model 2〇. 3G separates the stacking errors that cannot be described in the error models of (4) and (18), and finally uses lsm 24 to obtain the estimated error parameter parameters 25 stored in the network (1_7) and (4). A stack error model that is closer to the actual lithography process can be obtained. The above __ road has the ability to be fault-tolerant, generalized, and close to each other. This concrete ____路3G_咖卿_for example, 孰希 Related collar The technician can know that there are other kinds of gods that can be applied to the present invention, and the RBF-like God network is used as an example to clearly illustrate the present invention, and 1299520 is not used to limit the present invention to March. As shown in the figure, the Na-Shen network needs to go through the basic training process, and the part of the hidden layer is mixed and matched with the regional linear function % and the Gaussian function 33 as the transfer function. The unemployed county. Partial neural domain squamous transfer function 32-instar 4 _ _ + known controllable error squeak, and the aforementioned uncontrollable error term that cannot be fully described in the overlay error model will be wrong - the rest of the neurons with the same function as the transfer function, so the overlapping errors that cannot be described in the (u) and (iv))-stacked error models can be separated from them. On the mixed and matchable man-pair control, In addition to the above-mentioned stacking error estimation method, the φ of the stacking noise is estimated first, and then the error is estimated, and the correct control can not be corrected in the stacking error model. In the middle of a description or an unknown error exists, Such as: the amount of thermal deformation of the lens on the lithography equipment, the measurement error of the measuring equipment and the amount of thermal deformation of the wafer, etc., and the characteristics of these errors will be different from the shipboard and the environment, so the compensation is spent. The compensation can be mitigated by the error caused by these errors and can be matched by the error of the overlay error. The fourth picture shows the secondary points in the micro-age system, and the + includes the light source system 4, the lens system 42, the alignment system 43, the mask stage system 44, and the wafer stage system 45, etc. After the assembly is completed, the lens system seems to be in addition to the fault repair of the light source system 41, and in order to maintain its accuracy, no adjustment or modification is made at will. As for the crystal® stage system 45 and the reticle stage system 44, in addition to the transmission of the wafer and the reticle, in addition to the 15 1299520, it is possible to overlap the errors in the process of moving, shame and mix and __Differential parameters are selected from (1-7) and (1-8), using the wafer stage 45 which is commonly used in the general lithography process, translation, rotation and lifting and translation of the reticle stage μ , rotation and lifting, etc. as a mix and match control parameters. Another embodiment of the present invention is a method of mixing and matching matching of lithographic overlays. The flow design is as shown in the fifth figure. First, after the wafer 510 is subjected to the lithography process 510, it will be subjected to the overlay mark measurement 52 〇 to generate the overlay error measurement data 530 according to the stack of the error measurement data 53〇 and the standard graphic/standard device 52. The error measurement data is subjected to an estimation 540 of the overlay error model parameters, after which the mixing and matching of the overlay control 55〇 can be performed, and then the next lithography process 510 is performed. In the process of mixing and matching the selection of control parameter values, it is necessary to have a standard pattern (Golden Pattern) or a standard equipment (Golden Equipment) as the basis for the adjustment of the device, thereby reducing the overlay of the graphics produced by different devices. After the error. Mixing and matching the production of standard graphics (GoldenPattem), usually the shape and position of the lithography of the previous lithography as the basis for the next overlay of the graphics, so that the front and rear structural graphics have the best overlay effect, In the process of repetition, the influence of the overlay error on the lithography results can be effectively improved. The production of standard equipment (Golden Equipment) is usually selected from the noise group of the same structure layer pattern, and the equipment with the minimum stacking error production capability 1299520 is selected as the adjustment time of other equipment. Therefore, after the stack error model is obtained, the stack error model can be established accordingly. The above-mentioned hybrid and matchable pair control 550 is selected via the optimal control parameters and then controlled by the most determinate parameters. The process of the most healthy parameters of the county can be based on the previous _ specific implementation outline, after the separation of the controllable parameters and the uncontrollable parameter items, the correction of the number of ageable items can be used to compensate for the uncontrollable parameters. . The optimal control parameters of the towel are selected as shown in the sixth figure. First, the 551 (Tiandakou) gene filament produces a timidity of the controllable parameter item to estimate the control parameter value. Next step 552 brings the control parameter values into the overlapped block difference model and maps to the desired function in step 553 to compare the differences between the two devices in the stacking error scale, +, the desire function value is generated. The Taguchi gene-resolving algorithm is used to calculate the adaptive control parameter adaptation value, and the adaptation value is the Taguchi-style fine-grain algorithm. In the process of optimizing the control parameter selection, whether the silk touch has optimized the shaft mechanism. Next, the adaptive value is calculated in step 5M, and it is judged whether the control parameter selection has been optimized. If the optimization has not been achieved, the process returns to step 551 to perform the same process. The eager function can be touched by S-Saki, so the standard graphics or standard equipment is used to collect the thirsty shots on the stacking difference. After that, the control process in the ^Toukou-style algorithm is used to estimate the control. After the parameter value is brought into the error model, it is mapped to the desired function to compare the difference between the two devices. Based on the judgment of the fitness value, the maximum craving value of 17 1299520 and the minimum overlap are found out. The optimal control parameter solution for the error. The details of the process are detailed below. First, the method of evaluating the optimization effect will be explained. Hybrid and Matching Pair Control The criteria followed in the global overlap error compensation can be divided into the following two categories: (1) Standard Pattern (Golden Pattern) In the mixing and matching of the stack error adjustment, in order to reduce the double stack The error is generated, so it is also necessary to consider the directionality of the adjustment of the global overlap error. The stacking error of the previous layer is the reference for the adjustment. This is called the Golden Pattern. (2) Golden Equipment is found by the same model or group device performing the same step. The device with the smallest overlap error is used as the imitation device, and the output of the device is used as the adjustment reference for other devices. This is called “standard device ((}〇1如11 Equipment)”. The value of the control parameter value is selected in the stack. In the process of optimization, it is necessary to have an optimal judgment rule to assist the correctness of the entire selection process, so as to effectively obtain the desired optimization result, so in the judgment rule of the optimization effect, the desire can be The flexible design of the function, the lithography equipment and the _ or the standard equipment on the stack of _ she is eager to record +, that is, the standard graphics or fresh equipment _ to the error system #_ as a desire letter 18 1299520 The definition of the boundary line range in the equation, as a result of the selection of the subsequent optimization parameters on the selection. This will help to maintain good mixing and matching conditions. The circuit pattern overlay effect, and increase the group device's replacement and scheduling freedom. Next, introduce the principle of the desire function, the concept of the desire function was proposed by Harrington in 1965, the main concept is through mathematics The conversion converts the multiresponse problem originally required to be converted into a single response problem, and Harrington defines the individual desire value d (Individual Desirability) according to the quality of the manufactured product. The mathematical equation for the desired value is given by equation (24):

r package two [USL + LSL] \ η \ USL - LSL y (2-1) where } represents the i-th actual response value, [AS, L, LS, L represent the upper and lower limits of the tolerance of the specification, respectively, In this case, the overlay error measurement data obtained from standard graphics or standard equipment is used as the definition of the upper and lower bounds of the desired function tolerance, which is called Deviation Importance, and then geometric for each individual desired value. On average, the overall desired value is Z) (Overaii 19 1299520

Desirability), which converts the optimal parameter problem for multiple system responses into the problem of finding the maximum overall desired value.

In Harrington's concept of eager function design, the response value is far from the specification. When the center point (Midpoint) moves above the upper and lower tolerances, the rate of change of the desired value remains the same. However, from the application considerations of the industry, this kind of financial method is relatively incapable of being close to the actual needs of the application. Therefore, the desire function is re-corrected by the material and the test:

η η 丫

V^LSL

< J 卜~收丫 fUSL<y

?LSL<y<T

J<y<USL y < LSL (2-2) ., (2~2) towel, 4 represents the value of the craving function, w the output of the stack of errors, and the M-table should be reduced by _june, Γ The size of the miscellaneous dW, £ can be == design bribe design, its large table * when the r, the rate of decline of the desired value will be greater, ^ see and ^ respectively represent the upper and lower limits of tolerance error. 1299520 Therefore, in the application of the eager function on the hybrid and matchable overlay control (under the consideration of mixing and matching conditions), the optimal control parameter selection of the device should be adjusted after considering the actual import into the lithography device. The global overlap error that may be caused is changed so as not to cause another stack error. In order to be closer to the actual use requirements, the selection of the control parameters of the lithography device is only through the translation, rotation, lifting and blister loading of the wafer stage in (1-7) and (1-8). The adjustment of the basic controllable items such as translation, rotation, and lifting of the table can improve the mixing and the problem of error. Generally, in the process of parameter optimization, an optimization evaluation method is often used to guide the selection direction of the controllable item. Therefore, based on the need to mix and match the error of stacking errors, from the design of the desired function, the optimization of the parameter selection process is guided to the tolerant mixture and can be matched within the range of the error. The additional eager function can also be targeted for individual variations. Large exposure field areas are subject to special requirements, such as a certain under-wafer, in the thermal oxidation process, when the wafer area is thermally deformed, the special limitation of the desired parameter selection in the specific area can be borrowed from the desired function. This reduces the effect of the wafer on the mixing and matching stacking errors due to thermal deformation. The design of the desired function in the mixed and matchable condition is as shown in the seventh figure. The direction of the overlay error on the standard equipment (Golden Equipment) or the standard pattern (Golden Pattern) is divided into four quadrants, in the center of the quadrant. The position where the two axes intersect is the origin of the exposure field coordinate. In application, the overlapping error vector of each position in the standard pattern is placed on the origin of the exposure field coordinate, and then stacked on each measurement position. In the direction of the error change, determine which limit to use the 21 l2g952 parameter function to be used as the limit number of the device to be selected in the optimal control parameters, so that the sieve can have the best mix and match the effect. The best control participation, salty low adjustment equipment and standard $ equipment or standard graphics on the stack, the front and rear map open / stack „ Wu difference. The optimal parameter value selection results are judged as follows: 9P ^ Max(d) (2-3) 〇-style towel (10) represents the best control parameter ^ represents the individual eager value, as shown by equation (2-3) 'if the village control item is difficult to select, the better _ response output Will be closer to the value, and the desired value will also be Increase, from the financial _ size can determine whether the selection of controllable integer has been optimized. The following describes the Taguchi gene algorithm applied in the optimal control parameter value selection process. Limited by the global domain of the overlay error The geometric relationship makes the adjustment of the controllable items difficult to estimate, and the Taguchi gene algorithm (Taguehi-Genetic)

Alg〇rithm’TGA) has the ability to evolve from generation to generation, so it can find the best parameter solution from its evolution. The biggest difference between TGA and traditional gene algorithm (SGA) is that it incorporates the concept of straight parental test and method and factor analysis, so that the chromosomes can be shouted at 22 1299520. This increases the speed of the TGA in the global optimization search process and uses the characteristics of biological evolution, from the mass-produced chromosomes in the way of /Hungarian; too, to screen out the best mix and match the control parameter solution Generally, in the definition of the search range of the parameter solution, it is usually sighed by some empirical rules. In the optimal demand, the parameter search area is usually set in a wider range, but Too wide shot, relatively optimized in the process towel, cost too (four) and difficult to get the best parameter solution for the financial efficiency 'by RBF series Long Lu Laiyuan (four) Lin County parameters, as the best parameters The reference to the locale setting can increase the speed at which the entire optimization process converges' while ensuring the correctness of the optimization process. The numerical solution in the TGA is usually generated by random random numbers in the computer. Through the coding technique, the random random number is converted into a series of genes, and finally the chromosomes in the imitation organism are formed, and the chromosomes can be used for message transmission, mating, The work of mutation and evolution, TGA's calculation process is shown in Figure 8, and these functions will be detailed later. First, step 810 and step 82 are performed to generate a chromosome population. The demand for optimization effect can be determined by the design of the number of initial chromosome groups. Each gene in each chromosome is a binary code that is converted into a random number after computer. The chromosome group is generated as follows: (2 4) C = random{POP, NXG ] cn} 23 1299520 (2 4) where (^ represents a chromosomal group, ❿ ― - represents a group of randomly generated chromosomes, where qing represents the desire to produce (4) the number of chromosomes 1 represents the number of parameters, % represents the base _ segment ugly, in the process of calculus, will use each group of groups to turn into a number of sharks, reduce the material for subsequent mating, mutation and other steps, then bring this silk color with The message is restored to the best recorded number by the action of the code. The decoding method is as follows:

Rr - L + ψ I kSi2

CckU(UL) 2 bit-1 (2-5) & represents the numerical solution of the reduction of each parameter after the decoding of the chromosome, so that the n & 体 仃 卩 卩 卩 L L L L L L L L L L L L L L L L The minimum value, the size of μ, represents the resolution of the parameter encapsulation process. Next, the chromosome adaptation value is evaluated by step 83, and the outline is obtained. In the 4_ miscellaneous, the corresponding optimization effect evaluation method will be displayed with different needs. Generally, these measures can be used to cover the quantity (4) (4), and _ (4). The dyed flaps are evolved for generations, and the stacking and measuring of the coffee and the sampling 24 1299520 are mostly carried out by means of symmetrical and multiple exposure field measurement, thereby hoping to know the offset characteristics of the overlay error over the whole domain, so In the optimization process of the stack error correction parameters, the similarity of the two lithography equipments on the hybrid and matchable stacking errors must be considered, and the control parameters in the lithography system also have a geometric relationship 'so the control parameters When adjusting, if the correction is made only for a certain measurement point or a specific _light_field, the variation of the scale field area will increase. Through the design of the eager function, chromosomes that meet the optimization requirements can be selected from the genomic group, and the adaptive function design of TGA(10) can be used to judge the correct chromosomal face and the shirt is correct. In the middle, you can select (4) the optimal control parameter adjustment amount, and do not change the stacking error of the machine to be adjusted _ direction off _ standard bribe (Golden Equipment) or fresh graphics (GGldenPattem) _ job. For the reason of the threat, the design of the adaptive function is as follows: (2-6) F represents the fitness value, used for the merits of the lining effect, 4 represents the amount, the desired value of the ij, and w represents the number of the desired function. Depending on the number of measuring points, the weight of the 'W is the weight of the weight. The larger the value of the axis function is, the higher the value of the axis function is, the better the performance of the selected parameter value is. The desired value product ensures the optimal parameter selection in the hybrid and 25 1299520 matching conditions. The global overlap error and the graphical geometry are not related to other errors caused by parameter correction errors. Next, step 850 performs a straight-through method of mating. Chromosomes exchange information with each other through mating actions. On the one hand, it avoids the goal of entropy in the optimization process. The TGA uses the orthogonal table to select better parameters and overcome the Wei. Too many parameters _ color body is too long

The main meaning of the straight father is to balance the relationship between the parameter factors, and in the case of no mixing, it is known that the _ factor has the meaning, that is, in each of the orthogonal tables, each level is present. The number will be _. In the data analysis, the direct analysis table can be used as the main effect of each-fine element and then the degree of influence of each element on the experimental result. The experimental behavior of straight silk is in fact an experimental of the trauma factor in the complete factor experiment, thus saving a lot of time in performing the full factor experiment. The orthogonal table experiment itself has the hiding of systematic reasoning, so the approximate solution of the county factor is obtained. Straight and ® sub-analysis is a typical quality control method, available 2 = good dyed _ Wei, (four) _ more than the rate. Straight wire is defined as having N factors and 2 levels, then there will be a total of y combinations, when four],)'(1'2), (2,1)'(2,2) in all experiments When the average occurs, the factor in the column will be in a right angle state. Usually - the second mating requires two chromosomes to carry out 26 f 1299520 so you can look straight ~

A variable can be obtained - each positive _ _ table does not. The right-segment chromosome contains N 矣 矣 矣 . The integer β=2ί1〇82(Ν+1)1, so the established orthogonal table can be expressed as Ln(2n-1), and φ white 7 ^ ^ 3 11 columns and Hanhang. Assuming that a chromosome contains 7 variables, you can catch # + 7 ^ , and the orthogonal table of ^(2> is shown in Table 4_1. The steps for forming the straight parent table are as follows: Step 1: Put a Xu into the first row. , the level of the a factor! > Level 2 consecutive occurrences of the purchase, that is, the emergence of * level and then continuously appear 4 levels 2. Step 2: put the b factor into the second line position, where factor b The number of consecutive occurrences on level and level 2 is 8/(2*2)=2. Step 3: The composition of line 3 is axb, ie if a==b=1 or liver^2 shell 1J a X b =1 , No Bellow J mad X b = 2. Step 4: Put the c factor into the 4th row position, where the level of the c factor level 3 appears consecutively 8/(2*2*2) =1. 5: 5, 6, and 7 row steps 3 analogy. Table 4-1 Straight table Le(27) Experiment Factor — '— Adaptation number 1 2 3 4 5 6 7 Value component abb c axe bxc axbxc 1 1 1 1 1 1 1 1 γΓ 27 1299520 2 ΓΓ 1 1 2 ΓΓ 2 2 y2 3 1 2 2 1 1 2 2 y3 4 1 2 2 2 2 One----- 1 1 y4 5 2 1 2 1 2 —~-— _ 2 y5 6 2 1 2 2 1 ———. 1 y6 7 2 2 1 1 2 1 y7 ill 2 2 1 1 2 y8 Based on the above discussion In the application, the interaction between the parameters must be small. If the interaction between the parameters can be neglected, then each line of the orthogonal table can be regarded as a parameter. The main effect evaluation of the experiment through the orthogonal table is used as a parameter selection. The main effect is the value of the evaluation function of the fth experiment in the orthogonal experiment of Μ27) indicated by Table 4_1. Then define the main effect of the j-th factor:

Sjk " Ji yt X Fk (2-7) where = 1 indicates that the level j factor in the i-th test is moxibustion; otherwise, Qiao = 〇. Obviously, the main effect can show the performance of each individual in the factor, and the factor J with the greatest influence has the largest difference in main effect (Maineffect 28 1299520 differ· 'MED) I. If the value of the adaptive function is to be maximized, The main record, indicating that the effect hbM 2 mevel is better, so that the overall combination can be closer to the best solution, otherwise it would be better to re-elect the second ievei. During the mating process, two chromosomes are randomly selected from the ethnic group. The color body is coded and each parameter is the basic unit. As the experimental factor of straight wire, and the two-level orthogonal table is used, the level 〇 and level i of each factor correspond to the parent chromosome and the mother chromosome.丨 parameters. Through the orthogonal test, the excellent parameters of the parental coloring can be effectively extracted, and the new chromosome obtained after the combination of the mice will have excellent fitness. The TGA mating process is as follows: Step 1: Select the front (6) 亍 from the orthogonal table, where η—ι)1 (each variable is the each factor in the orthogonal table). Step 2: The yth factor in level i and level 2 represents the first variable in parent chromosome 1 and parent chromosome 2. Step 3 • Calculate the % fitness value of the chromosomes of each column combination in the orthogonal table, where ί = 1, 2, ···, η. Step 4: Calculate the main effect, where y=1, 2,...,Ν;_,2. Step 5: Determine the best level by each variable. When you know %, select Level 1 from the 7th variable, otherwise choose Level 2 of the second variable. Step 6: The first progeny chromosome is found in the best of the orthogonal table - 29 1299520 combination. Step 7. By level! The most influential factors are classified into rank N, and large MED factors can be classified to higher ranks. Step 8: The second chromosome is roughly the same as the first chromosome except that the lowest categorical variable takes another level. In summary, the traditional GA can search a large number of parameters in the defined parameter space, but when the parameters of the foreign country need to be solved, the relative transformation will be extremely large, so that the GA will converge to the best shot in the region. And it takes a lot of time to calculate the time. Therefore, by using the orthogonal and unified reasoning mechanism, the best combination of factors can be found in a very short time _ ’, and the search time and regional optimal solution can be improved. After the orthogonal method of step 850, the mutation of step 860 is still required. In the process of color body tilting, in addition to the highest fitness value of the chromosome, it will also be subject to the calculation of the performance, and the mutation will be mutated by the poor adaptation of the _ _ 'from the unit base _ reorganization process, try to evolve more Good chromosomes, but because of the mutation of the silk pass # is difficult to control by the number of electric turbid machines, so the adaptation value of the chromosome may become better, but it may become worse, - in the mutation The rate is usually designed to be less than 0.1 U. The stomach will be used to mutate the stalk chromosome, which is less than 10% of the chromosomal population. 30 1299520 After the completion of step 880, the evolution of a generation is completed. In order to achieve optimal results, it is necessary to carry out evolution of sufficient generations. Therefore, it is determined by step 870 whether the evolutionary generation is sufficient. If more generations are needed, return to step 810 to regenerate the next generation. The evolutionary behavior of TGA is to preserve the best chromosomes in the previous generation and compete with the next generation, in order to maintain the optimal results, and at the same time achieve the purpose of the chemistry, usually in the process of eugenics, usually At least one of the best chromosomes will not be mated and mutated, and the number of eugenic chromosomes will usually affect the speed of evolution. Therefore, the selection of appropriate eugenic chromosomes can not only help TGA optimize. In the process, the computation time is reduced, and the convergence of the optimized parameter solution in the solution can be effectively increased. From the various processes described above, in the process of mixing and matching the selection of the control parameters, the compensation of the parameter values It must be able to achieve the principle that the overlapping direction and the stacking error of each measuring position can be minimized, and the mixing and matching errors of the overlapping direction directly affect the cumulative stacking of the structural figures in other directions. Causes the electronic components to be unable to produce the Wei, and the stacking error becomes larger. If it can be maintained in the design principle, then Accept. Therefore, through the design of the eager function, the first thing to do is to limit the selection of the optimal control parameter value to the device. The alignment error of the quasi-graphics can have the same direction, thereby reducing the morphing after the morphing. The electronic components make a good concern, and with the adaptation function, all the thirsty _ operations can be effective _ each exposure field 31 1299520 in the mixing and can be matched to the correctness of the error direction.

Accordingly, one embodiment of the present invention is a method of mixing and matching (4) a lithographic overlay. First, the m-image bribery--the object carries out a first-micro-shadow' the first-郷余乡_卿__ to the object to produce - map the object in the object, and its towel corresponds to the standard figure of the first lithography device _ Den Pat (four) / standard equipment in the stacking error „ on the characteristics of the record-craving function. Next to the Qiqiqing one stack of the difference between the measured amount of difference between the amount of the raw material _ no reflection between the error The lion Φ pairs * difference data generation _ 帛 郷 郷 equipment controllable items and uncontrollable items; miscellaneous control items and material controllable items generate "frequency error mode", according to optimization method to generate a set of optimal control a parameter, according to the optimal control parameter, the second lithography device is carried out; and the second lithography process is performed. The optimization method comprises: generating a complex array control parameter; each plant control parameter is replaced #对误的册- 叠职差和_to the eager function to generate - financial value, · Saki is expected to be straight _ out of the best _ parameters, the best control difficulty has the largest accounting minimum overlap error. The other method is related to the If method (4) type fine material field, and the present invention is not limited thereto, and the above controllable Membership can be tied by the - _ turn the private road, the surface of fine _ Finance Office paid his way, the present side 3 -_ neural networks separated fine lithography controllable item and uncontrollable Yun, 32 reuse.

Claims (1)

1299520 X. Patent application scope: 1. A method for mixing and matching control of a lithography pair, comprising: performing a first lithography process on an object by a first lithography device, the first lithography process projecting a first pattern to the object, the first lithography device and the structural pattern are mixed as a lithographic overlay and matched with a standard pattern of control (Golden Pattern)/^;Golden Equipment Measuring the first-structure _ stack-to-error measurement (4), the _ structure pattern of the stack-to-error measurement data is an error between the first pattern and the first structure pattern, wherein the first structure graphic system According to the first pattern, the definition of the boundary line in the eager function according to the first structure 形 — ; ; ; ; ; 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望 渴望The second lithography process projects a second pattern to the object; 'measuring - the second structure _ stacking error measurement (4), the second structure pattern stacking error measurement data is the second pattern and The error between the second structural figure, ^ The second structural pattern is generated according to the second graphic; according to the (4) error system data of the second structure, the (four) error characteristic value of the controllable item and the uncontrollable item of the error model is obtained; The method generates a set of the most robust parameters, wherein the optimization method comprises: generating a complex array control parameter; 39 1299520 Substituting each of the parameters into the stack error model to estimate a stack error and mapping to the desired function generation - craving a value; and judging 4 the desired value until the optimal control parameter is selected, wherein the optimal control parameter reduces the desired value to a maximum and the corresponding overlay error is minimal; and adjusts according to the optimal control parameter The second lithography device. 2. The method of mixing and matching control of the lithographic overlay according to the scope of the patent application, further comprising: the controllable item of the stack-to-fold error model according to the stacking measurement data of the first-structure figure The stack of error control features of the control; generating an optimal set of control parameters by an optimization method; and adjusting the first lithography device according to the optimal control parameter. 3. According to the method of mixing and matching control of the lithographic overlay of the first claim of the patent scope, the method further comprises: measuring the stack-to-error measurement data of the third structural figure, and the stacking error amount of the third structural figure The measurement data is an error between a third graphic and the third structural graphic, wherein the third structural graphic is generated according to the third graphic; and the stacking error is generated according to the stacking of the third structural graphic a stacking error eigenvalue of the controllable item and the uncontrollable item of the model; defining a boundary line range in the craving function according to the stacking error measurement data of the third structural figure; generating a set according to the optimization method Optimal control parameters; and 1299520. The second lithography apparatus is adjusted based on the optimal control parameters. 4. According to the method of mixing and matching control of the lithographic stack pair according to the scope of the patent application scope, the method further comprises: measuring the stacking error measurement data of the second structural figure, and the stacking of the third structural figure is different The measurement data is an error between a third graphic and the third structural graphic, wherein the third structural graphic is generated according to the third graphic; and the stack of the first structural graphic generates a stack of errors according to the error measurement data. The error characteristic value of the controllable item and the uncontrollable item of the error model; generating a set of optimal control parameters according to the optimization method; and adjusting the second lithography apparatus according to the optimal control parameter. 5. The method of mixing and matching control according to the lithographic overlay of item 1 of the patent application scope wherein the above optimization method adopts a Taguchi gene algorithm. 6. According to the method of mixing and matching control of lithographic overlays according to the scope of patent application ,5, wherein the generation of the above control parameters comprises: generating a set of controllable parameters by an optimization method; and The gene algorithm generates an adjustment amount of the controllable parameters of the group; the number of the control lengths is a combination of the controllable parameters and the amount of the control. 7. The root shot should be mixed with the lithography stack of the third item and matched with the control method, wherein the above controllable parameters are generated by the least square method. 8. The method of mixing and matching control according to the lithographic stacking of the third item of the Chinese Patent (4), wherein the above controllable uncontrollable item is generated by a type of neural network. 41 1299520 9· According to the patent application scope 帛 8 items of lithographic overlay pairs and matching control to the above-mentioned neural network package, a first neuron group and a second neural group, The first neuron group and the second neuron group respectively generate the controllable item and the uncontrollable item. 10. A method of mixing and matching control according to ninth aspect of the patent application scope, wherein the first group of neurons is a linear function as a transfer function to generate the controllable item. 11. The method of mixing and matching control of lithographic overlays according to item 9 of the patent application scope wherein the second group of neurons described above uses a Gaussian function as a transfer function to generate the uncontrollable term. 12. A method of mixing and matching control of lithographic overlays according to item 8 of the scope of the patent application, wherein the neural network is a radial base-like neural network. 13. The method of mixing and matching control of a lithographic overlay according to the scope of the patent application, wherein the lithography apparatus has an object stage and a reticle stage, wherein the controllable item comprises the following group Assembly: translation of the wafer stage, rotation of the wafer stage, lifting of the wafer stage and translation of the reticle stage, rotation of the reticle stage, and lifting of the reticle stage. 14. The method of mixing and matching control of a lithographic overlay according to claim 13 of the patent application scope, wherein the image forming apparatus has an object stage and a mask stage, wherein the uncontrollable item does not include the following group The set of groups: the translation of the wafer stage, the rotation of the wafer stage, the lifting of the wafer stage and the translation of the reticle stage, the rotation of the reticle stage 42 1299520 • The lifting of the reticle and the reticle stage. 15. A method of mixing and matching control according to the lithographic overlay of claim </ RTI> wherein the above-mentioned craving function includes a horizontally oriented stacking error error function and its boundary line range. 16. The method of mixing and matching control of lithographic overlays according to item 1 of the patent application scope wherein the above-mentioned craving function includes the vertical direction of the craving function of the error and its boundary line range. 17. A method of mixing and matching control of a lithographic overlay according to item 1 of the scope of the patent application, wherein the controllable item comprises a plurality of operands. 18. A method of mixing and matching control according to the lithographic overlay of the scope of the patent application, wherein the uncontrollable item is an operand. I9. According to the application, the lithography of the first item is superimposed on the method of mixing and matching control, wherein the object is a wafer. 43 1299520 , XI, Schema: 44 1299520 IBIBIBI _ slippery || ft ft — Ξ Ξ rr — hd Η ΓΓΓ — Ξ — three Η / /, area alignment - stack pair measurement - global alignment mmm crystal The circle is aligned with the first A picture (a) No offset coverage error flag (b) Overlap error flag when offset First B diagram