A method of manufacturing a semiconductor device special-purpose apparatus

Abstract

The application discloses a semiconductor device special equipment manufacturing method, and relates to the technical field of semiconductor device special equipment manufacturing, and comprises the following steps: collecting light path wavefront distortion data and double workpiece table vibration and positioning deviation data of a semiconductor device special equipment in situ, obtaining an integral machine actual measurement distortion vector in combination with a digital twin model, performing reverse solving on the integral machine actual measurement distortion vector based on a forward mapping function, positioning a position coordinate and an amount of stress of a residual stress node, synchronously applying high-frequency vibration excitation and a local thermal gradient field to the stress node by using an end effector before final fastening, actively releasing residual stress at the node, adjusting a control voltage of a piezoelectric micro-motion pad array according to a prediction result of the digital twin model, adjusting equivalent local stiffness of the installation node to target stiffness, completing final fastening, and performing integral machine performance re-measurement and acceptance.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device manufacturing technology, specifically a method for manufacturing semiconductor device manufacturing equipment. Background Technology

[0002] As the feature size of semiconductor devices continues to evolve towards advanced processes, the manufacturing and assembly precision requirements of semiconductor device-specific equipment have reached extremely high levels. Such equipment typically consists of core subsystems such as large-size precision optical components, precision dual-stage systems, precision bases, and vibration isolation systems. These subsystems are connected and fastened through high-precision mechanical mating surfaces. During the overall manufacturing stage, components need to undergo a series of processes such as roughing, finishing, cleaning, handling, pre-assembly, fine-tuning, and final fastening. Each process may introduce new residual assembly stresses at the mating surfaces. These residual assembly stresses refer to the undesirable stresses generated and locked inside the structure at the mating surfaces and mounting nodes during the component processing, handling, assembly, and fastening processes due to factors such as the release of mechanical stress in the base, gravity deformation, and mismatch in the thermal expansion coefficients of materials.

[0003] Existing whole machine assembly and manufacturing solutions mainly follow the design concept of positive tolerance accumulation and distribution. That is, according to the whole machine index requirements, the allowable error range is decomposed and distributed to the machining tolerance and assembly tolerance of each component. Then, the whole machine index is achieved by improving the machining accuracy of individual components. On this basis, some solutions will also introduce post-assembly software compensation methods. That is, after the whole machine is assembled, residual errors are collected by optical wavefront sensors and workpiece stage position sensors, and adaptive optical elements or workpiece stage servo control algorithms are used to perform real-time or static compensation of errors at the software level.

[0004] The closest existing technical solution to this invention is as follows: In the process of assembling a semiconductor equipment, the mechanical docking and bolt fastening of the projection lens group and the dual workpiece stage base are completed first in accordance with the pre-set assembly process sequence. Then, the wavefront of the optical path is detected by an interferometer. If the detection result exceeds the index range, the thickness of the mechanical pad of the optical element is adjusted or the wavefront distortion is compensated by software by using a deformable mirror.

[0005] For the positioning deviation of the dual workpiece stage, the main approach is to improve the machining and assembly tolerances of key components such as guide rails, grating rulers, and air bearings, and to suppress errors by combining the feedforward compensation model in the servo control algorithm. The assembly steps of this solution can be summarized as follows: precision machining and inspection of components, pre-assembly of the base, cleaning and initial fastening of the mating surfaces, wavefront detection of the optical path, adjustment of shims or software compensation, final fastening, and retesting and acceptance. The above solution includes the following structural components: projection lens group, mechanical mounting base, mating surface shims, dual workpiece stage base, guide rails and air bearings, interferometer detection system, deformable mirror and its drive controller, and workpiece stage servo control system. The components are connected by bolts and precision positioning pins, and the detection system and controller form a closed-loop feedback relationship through signal cables.

[0006] The aforementioned existing technologies have the following drawbacks: First, existing technologies rely on extreme improvements in the machining tolerances of parts to control assembly errors. However, domestically produced parts cannot fully achieve the ideal consistency level of imported parts in terms of material grain structure, surface residual stress, and micro-geometric tolerances in the short term. After assembly according to the traditional positive tolerance allocation approach, the assembly will have a non-uniform residual stress distribution that is difficult to pre-quantify due to differences in material properties. This leads to significant dispersion in wavefront distortion and positioning errors between different batches of equipment, ultimately resulting in poor mass production consistency and low assembly yield. Second, the post-production software compensation methods in existing technologies only compensate for static or quasi-static error signals that have already occurred. However, the release of assembly residual stress often has random time delay characteristics and a wide frequency range. In terms of spectral distribution, the software compensation system is limited by the travel range of the deformable mirror, the servo bandwidth, and the real-time performance of the control algorithm. It is difficult to effectively suppress the dynamic uncertainties in the high-frequency band, and it cannot solve the physical deformation limit already reached by the hardware structure itself. Therefore, even if software compensation is introduced, wavefront aberration drift and workpiece stage trajectory tracking error drift caused by assembly stress release will still remain. Thirdly, since the existing technology lacks physical means to release residual stress in situ, quantitatively, and controllably during the assembly and manufacturing stage, the assembly stress can only be released passively and disorderly over time after the equipment is fully assembled. This passive release process cannot be completely exposed and eliminated during the factory inspection stage, resulting in the risk of performance drift after the equipment is delivered. Moreover, the stress release path and release amount differ between different equipment, further aggravating the problem of overall machine consistency. Summary of the Invention

[0007] The technical problem to be solved by this invention is that during the manufacturing and assembly of semiconductor device-specific equipment, residual assembly stress caused by factors such as the release of mechanical stress of the base, gravity deformation, and mismatch of the thermal expansion coefficients of materials will be gradually released as the equipment runs, resulting in the accumulation of wavefront distortion in the core optical path and the drift of nanometer-level trajectory tracking error of the dual workpiece stage.

[0008] Existing processes rely solely on two methods: pre-processing to improve component tolerances and post-processing software compensation. These methods are insufficient to address the consistency issues arising from differences in material properties and geometric tolerances between domestically produced and imported components. Furthermore, they fail to eliminate high-frequency dynamic uncertainties and the physical deformation limits of the hardware itself at the physical hardware level. This results in poor mass production consistency and low assembly yield. The present invention aims to provide a physical manufacturing method that can eliminate assembly stress in situ, in real-time, and in a closed-loop manner during the equipment hardware manufacturing and assembly stages.

[0009] To solve the above technical problems, the present invention provides a method for manufacturing a special equipment for semiconductor devices, including the following technical solution: in the state before final fastening after pre-assembly, the optical path wavefront distortion data and the vibration and positioning deviation data of the dual workpiece stage of the special equipment for semiconductor devices are collected in situ using an in-situ data acquisition unit, and combined with a pre-established whole machine digital twin model, the measured distortion vector of the whole machine is obtained.

[0010] Based on the measured distortion vector of the whole machine and the pre-established forward mapping function, the estimated value of the residual stress distribution that causes the measured distortion vector of the whole machine is obtained by inverse solution, and the position coordinates and stress values ​​of the stress nodes are determined according to the estimated value of the residual stress distribution.

[0011] Before final tightening, a high-frequency vibration excitation and a local thermal gradient field are simultaneously applied to the stress node using an end effector to form a thermal-vibration composite excitation. Through the synergistic effect of thermally induced material softening and vibration-induced micro-dislocation slip, the residual stress at the stress node is actively released until the residual stress at the stress node drops to within the preset stress release target threshold.

[0012] Based on the prediction results of the overall dynamic characteristics of the digital twin model, the target stiffness of each mounting node is determined, and the equivalent local stiffness of the mounting node is adjusted to the target stiffness by adjusting the control voltage of the piezoelectric micro-motion pad array arranged at the mounting node.

[0013] After completing the stress relief and stiffness adjustment, the semiconductor device-specific equipment is finally tightened, and the overall performance of the machine after final tightening is retested and accepted.

[0014] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0015] 1. This invention uses the in-situ multi-physics coupling error reverse propagation method, taking the measured wavefront distortion and positioning deviation data of the whole machine as input, and reversely solves the residual stress node position and stress value that cause the distortion, thereby realizing the precise positioning of assembly stress nodes and overcoming the defect that the forward tolerance allocation method cannot predict the actual stress distribution.

[0016] 2. This invention eliminates the potential for wavefront aberration drift caused by assembly stress at the physical hardware level by applying thermal vibration composite excitation to the stress nodes located before final tightening during the assembly and manufacturing process. This overcomes the shortcomings of pure software compensation methods in dealing with high-frequency dynamic uncertainties and the physical deformation limits of hardware.

[0017] 3. The present invention achieves stiffness topology adaptive compensation through piezoelectric micro-motion pad array. Based on the dynamic error distribution predicted by the digital twin model, the local stiffness of the installation node is reconfigured in a purposeful manner. To a certain extent, the dependence on extreme machining tolerances of individual parts is relaxed, and the overall assembly consistency and yield of the machine after the replacement of domestic parts are improved.

[0018] 4. This invention forms a closed-loop overall assembly and manufacturing process chain through the cooperation of the above four links: data acquisition, reverse stress positioning, active stress release, and stiffness topology compensation. This enables residual stress in the assembly to be actively and thoroughly exposed and eliminated before the equipment leaves the factory, avoiding the risk of performance drift caused by passive stress release after the equipment is delivered. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall machine manufacturing system architecture in an embodiment of the present invention;

[0020] Figure 2 This is a flowchart illustrating the in-situ multiphysics coupling error backpropagation algorithm in an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of the overall assembly and manufacturing process in an embodiment of the present invention. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1-3 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0023] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. It should be noted that the following embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention. Equivalent substitutions or modifications made by those skilled in the art based on the technical solution of the present invention without creative effort should all fall within the scope of protection of the present invention.

[0024] System Overall Architecture

[0025] This embodiment takes the assembly and manufacturing of a certain type of immersion lithography equipment as an example. The main body of the equipment to be assembled mainly includes a projection lens group, a mechanical mounting base, a mating surface between the projection lens group and the mechanical mounting base, a dual-workpiece stage base, and a dual-workpiece stage motion system. In order to achieve in-situ, real-time, and closed-loop elimination of assembly residual stress, the present invention sets up an in-situ data acquisition unit, a digital twin modeling and reverse propagation calculation unit, a thermal shock composite active stress release execution unit, a piezoelectric micro-motion pad stiffness topology compensation unit, and a final fastening and retesting acceptance unit in the whole machine manufacturing system. The units are connected in a closed-loop feedback relationship through a data bus and control signal cables.

[0026] The in-situ data acquisition unit includes a multi-wavelength interferometer arranged at key nodes of the optical path and a multi-dimensional accelerometer array arranged at key nodes of the dual workpiece stage movement. These are used to acquire optical path wavefront distortion data and vibration and positioning deviation data during the workpiece stage movement in real time, respectively. The digital twin modeling and reverse propagation calculation unit receives the measured data output by the in-situ data acquisition unit, performs reverse calculations in combination with the pre-established whole-machine finite element digital twin model, and outputs the position coordinates and stress values ​​of the stress nodes. The thermal vibration composite active stress release execution unit receives the reverse calculation results and applies thermal vibration composite excitation to the stress nodes through the end effector. The piezoelectric micro-motion pad stiffness topology compensation unit is arranged at each installation node, receives the dynamic error prediction results output by the digital twin model, and adaptively adjusts the local stiffness. The final tightening and retesting acceptance unit is used to complete the final bolt tightening and perform overall machine performance retesting.

[0027] Data Acquisition and Digital Twin Modeling

[0028] In the assembly and manufacturing process, a digital twin model is first established for the core optical path system and dual workpiece stage system of the semiconductor equipment to be assembled. The digital twin model refers to a digital mirror model constructed in a virtual computing environment that is highly consistent with the physical entity semiconductor equipment in terms of geometric structure, material constitutive model, boundary constraints, and dynamic characteristics. It is used to simulate, predict, and invert the operating state of the physical entity. The model includes the geometric model, material constitutive model, joint surface contact model, and assembly boundary condition model of each component. The material constitutive model includes parameters such as the elastic modulus, thermal expansion coefficient, and yield strength of each component material. The joint surface contact model is used to describe the contact stiffness and friction characteristics at the joint surface. The assembly boundary condition model is used to describe the constraint relationship between each component.

[0029] Before final fastening after pre-assembly, a multi-wavelength interferometer is used to perform in-situ detection of wavefront distortion in the optical path. In-situ refers to the method of directly collecting, analyzing, and processing data on in-process or assembled components without disassembling or transferring the workpieces at the semiconductor equipment assembly and manufacturing site. At the same time, a multi-dimensional accelerometer array arranged on the key moving parts of the dual workpiece stage is used to collect vibration and positioning deviation data. In this embodiment, the wavefront detection accuracy of the multi-wavelength interferometer reaches the sub-nanometer level, and the sampling frequency of the multi-dimensional accelerometer array is not less than 10 kHz to ensure that high-frequency vibration components during the movement of the dual workpiece stage can be captured. The above-mentioned in-situ measured data is used as the input boundary conditions of the digital twin model for subsequent reverse calculation.

[0030] In-situ multiphysics coupling error backpropagation algorithm

[0031] Unlike the traditional forward tolerance accumulation and allocation method, the error backward propagation algorithm proposed in this invention takes the measured wavefront distortion and positioning deviation data of the whole machine as input, and solves in reverse the residual stress node location and stress value that cause the distortion. The error backward propagation refers to the process of calculating the stress distribution node and stress value that cause the distortion, which is different from the traditional analysis method of deriving the whole machine error from the forward accumulation of component tolerances. In this invention, it specifically refers to the calculation process of starting from the measured wavefront distortion and positioning deviation data of the whole machine and solving in reverse the distortion. The specific implementation method is as follows.

[0032] First, a forward transfer model of the whole machine digital twin is established. This model describes the mapping relationship between the residual stress distribution of the bonding surface and the optical path wavefront distortion and the workpiece stage positioning deviation based on the finite element method. It is denoted as the forward mapping function. Its input is the stress distribution vector to be solved and its output is the predicted distortion vector.

[0033] Then, a multi-wavelength interferometer is used to collect the measured wavefront distortion data of the optical path, and a multi-dimensional accelerometer array is used to collect the measured positioning deviation data of the dual workpiece stage. The two types of data are then fused into a unified whole-machine measured distortion vector according to preset weights.

[0034] Next, the objective function for inverse solution is constructed. By minimizing the sum of squared residuals between predicted and measured distortions, and by introducing a regularization term to suppress the non-uniqueness of the solution, the following objective function is obtained:

[0035] ;

[0036] in, This represents the estimated optimal residual stress distribution obtained from the solution;

[0037] This represents the measured distortion vector of the entire machine obtained by fusing the aforementioned steps;

[0038] This indicates that the positive mapping function established by the aforementioned steps is applied to the stress distribution. Predicted distortion output;

[0039] It represents the square of the Euclidean norm, used to measure the deviation between predicted and measured values;

[0040] This represents a regularization constraint term applied to the stress distribution, used to ensure the physical rationality and numerical stability of the solution;

[0041] This represents the regularization coefficient, used to balance the accuracy of data fitting with the smoothness of the solution.

[0042] Subsequently, an iterative optimization method based on gradient descent and sensitivity matrix is ​​used to solve the above objective function. In each iteration, the residual between the predicted distortion and the measured distortion corresponding to the current stress distribution estimate is first calculated. Then, the stress distribution estimate is updated in reverse according to the Jacobian sensitivity matrix of the forward mapping function. The iteration is repeated until the residual converges to within the preset threshold, thereby obtaining the converged stress distribution estimate.

[0043] Finally, based on the converged stress distribution estimation results, and combined with the pre-calibrated coordinate library of the joint surface and mounting base nodes, the specific distribution nodes of residual stress that cause wavefront distortion and positioning deviation, as well as the stress values ​​at each node, are determined. The node coordinates and stress value information are then output to the subsequent thermal vibration composite active stress release execution unit and piezoelectric micro-motion pad stiffness topology compensation unit. In this embodiment, through the above reverse calculation, three main stress nodes at the joint surface between the projection lens group and the mechanical mounting base, and two main stress nodes at the joint surface between the dual workpiece stage base and the vibration isolation base are located.

[0044] Localized thermal vibration combined with active stress relief assembly process

[0045] For the stress nodes located in the aforementioned steps, before final tightening, this invention uses a dedicated end effector to apply a composite excitation of nanoscale micro-amplitude high-frequency vibration and local non-uniform thermal gradient to the node. The thermal-vibration composite active stress release refers to the simultaneous application of a local non-uniform thermal gradient field and nanoscale micro-amplitude high-frequency vibration. Through the coupling effect of thermally induced material softening and vibration-induced micro-dislocation slip effect, the stress locking state at the joint surface is actively broken, thereby realizing a process method for controllable release of assembly stress. The specific implementation steps are as follows.

[0046] First, based on the stress node coordinates output by the aforementioned algorithm, the thermal-vibration composite end effector is positioned and attached to the target node surface. The end effector integrates a piezoelectric ceramic resonator module and a local laser heating module. The end effector also includes an actuator body shell, a flexible contact end attached to the node surface, and a stress monitoring sensor integrated near the end. The piezoelectric ceramic resonator module and the local laser heating module are fixedly installed inside the actuator body shell. The output excitation of the two modules acts synchronously on the target stress node surface through the flexible contact end. The stress monitoring sensor feeds back the real-time monitoring signal to the drive controller, forming a closed-loop control relationship.

[0047] Then, the piezoelectric ceramic oscillator module applies high-frequency micro-amplitude vibration excitation with adjustable frequency and amplitude to the target node, while the local laser heating module applies a non-uniform thermal gradient field to the same node area. The two types of excitation are applied synchronously in time to form a thermal-vibration composite excitation field. In this embodiment, the frequency adjustment range of the vibration excitation is set to the range of 2 kHz to 50 kHz, the vibration amplitude adjustment range is set to the range of 1 nanometer to 100 nanometer, and the temperature gradient applied by the local laser heating module is controlled within a safe range that does not cause phase change or permanent plastic deformation of the component material.

[0048] Next, during the application of the composite excitation, stress monitoring sensors placed near the node are used to collect the dynamic process of stress release at the node in real time. The stress decay law over time can be expressed as:

[0049] ;

[0050] in, The duration of the composite excitation is... The real-time value of the residual stress at that node at that moment;

[0051] This represents the initial residual stress value at the node before the application of the composite excitation, obtained from the aforementioned reverse calculation steps;

[0052] Indicates the applied vibration amplitude;

[0053] Indicates the applied vibration frequency;

[0054] This represents the applied local thermal gradient value;

[0055] The coupling coefficient representing the vibration excitation to the stress release rate is predetermined by the end effector calibration test;

[0056] The coupling coefficient between thermal gradient excitation and stress release rate is also determined in advance by calibration tests;

[0057] This indicates the duration of the combined excitation.

[0058] Subsequently, based on the residual stress value fed back in real time by the stress monitoring sensor, it is determined whether it has dropped to within the preset stress release target threshold. If the threshold has not been reached, at least one of the vibration amplitude, vibration frequency or thermal gradient parameters is adjusted according to the preset step size, and the thermal vibration composite excitation is applied repeatedly until the residual stress at the node drops to within the target threshold.

[0059] Finally, the above-mentioned excitation and monitoring process is executed sequentially or in parallel on all stress nodes located in the aforementioned steps to complete the active release of all key stress nodes of the whole machine, and then the final tightening process is entered. In this embodiment, after the above-mentioned process, the residual stress value of each stress node at the joint surface of the projection lens group and the mechanical mounting base is reduced to within the preset target threshold compared with the previous process, which lays the foundation for subsequent stiffness topology compensation and final tightening.

[0060] Hardware-level stiffness topology adaptive compensation

[0061] After local stress relief is completed, in order to further improve the overall dynamic consistency, this invention introduces a piezoelectric micro-motion pad array at each mounting node. Based on the dynamic error predicted by the digital twin model, the local stiffness at the node is adaptively adjusted before final tightening to achieve hardware alignment and dynamic performance matching. The piezoelectric micro-motion pad is a functional pad device based on piezoelectric ceramic material that can achieve nanoscale thickness and local stiffness dynamic adjustment by applying voltage. It is used to achieve programmable adjustment of stiffness topology at the mounting node.

[0062] The aforementioned stiffness topology adaptive compensation refers to the purposeful reconfiguration of the local stiffness distribution at the assembly installation nodes based on the dynamic error distribution predicted by the digital twin model, in order to match the compensation method of the design dynamic objectives.

[0063] An array of piezoelectric micro-motion pads is arranged at the mating surface between the upper surface of the mounting base and the upper assembled component. Multiple piezoelectric micro-motion pad units are evenly arranged circumferentially along the mounting node, and each unit is connected to a control voltage source through an independent lead. The local stiffness adjustment relationship at this node can be expressed as:

[0064] ;

[0065] in, This indicates the equivalent local stiffness at the installation node after adjustment;

[0066] This indicates the reference stiffness of the piezoelectric micro-motion pad array at this node when no control voltage is applied;

[0067] This indicates the total number of piezoelectric micro-motion pad units arranged at this node;

[0068] Indicates the first The stiffness adjustment increment that each unit of control of a piezoelectric micro-motion pad unit can provide is obtained by pre-calibrating the piezoelectric material properties and geometric dimensions of the pad unit;

[0069] Indicates the application to the first The control voltage or corresponding displacement control quantity of each piezoelectric micro-motion pad unit.

[0070] The specific implementation method of this compensation process is as follows:

[0071] First, based on the simulation prediction results of the overall dynamic characteristics of the digital twin model, the ideal stiffness distribution required for each installation node to achieve the target dynamic indicators is determined and denoted as the target stiffness set.

[0072] Then, the reference stiffness of the piezoelectric micro-motion pad array at each installation node and the stiffness adjustment coefficient of each unit are read. Combined with the target stiffness set, the control quantity required to be applied to each piezoelectric micro-motion pad unit is solved.

[0073] The obtained control quantity is then applied to the corresponding piezoelectric micro-motion pad unit to adjust the equivalent local stiffness at the node to near the target stiffness.

[0074] Finally, the adjusted overall dynamic response is retested using the in-situ data acquisition unit, and the retest results are fed back to the digital twin model to determine whether the target dynamic index is met. If not, the above control quantity solution and application process is repeated for fine-tuning until the index requirements are met, and then the final fastening process is entered. In this embodiment, by applying different control voltages to multiple piezoelectric micro-motion pad units arranged at each installation node, the programmable topology adjustment of the local stiffness distribution at the installation node is realized, so that the dynamic characteristics of the assembly gradually approach the target dynamic index predicted by the digital twin model.

[0075] Final tightening and retesting

[0076] After completing the above-mentioned local thermal vibration composite active stress release and piezoelectric micro-motion pad stiffness topology compensation, the connecting bolts at the mating surface are finally tightened. Then, the in-situ data acquisition unit is used to re-measure the wavefront distortion data of the whole machine optical path and the positioning accuracy of the dual workpiece stage after final tightening. The re-measured data is compared with the preset whole machine factory acceptance index to determine whether the whole machine performance meets the acceptance requirements.

[0077] If the requirements are met, the entire machine assembly and manufacturing process is complete.

[0078] If the requirements are not met, the corresponding steps in digital twin modeling, error reverse positioning, active stress release, and stiffness topology compensation will be carried out again based on the retest data until the overall performance meets the acceptance requirements.

[0079] Overall process flow

[0080] In summary, the semiconductor device manufacturing method provided by this invention comprises five stages: data acquisition and modeling, error reverse propagation and positioning, local thermal vibration combined active stress release, piezoelectric micro-motion pad stiffness topology compensation, final fastening and retesting acceptance. Each stage works together to form a closed-loop assembly and manufacturing process chain, namely, a complete technical solution of in-situ data acquisition, reverse stress positioning, active stress release, stiffness topology compensation, and retesting acceptance. Through the above-mentioned closed-loop process chain, assembly residual stress can be actively and thoroughly exposed and eliminated before the equipment leaves the factory, thereby overcoming the defects of poor overall mass production consistency and low assembly yield of existing technologies that rely solely on improving component processing tolerances and subsequent software compensation.

[0081] Alternative implementation methods

[0082] In another embodiment of the present invention, in addition to using the finite element method to establish the forward mapping function, the aforementioned forward transfer model can also be constructed by using a data-driven proxy model trained based on historical assembly data to replace the finite element forward model. The data-driven proxy model includes a Gaussian process regression model or a neural network order reduction model to further improve the real-time computational efficiency of the reverse solution, which is suitable for assembly scenarios with high requirements for closed-loop response speed.

[0083] In another embodiment of the present invention, in addition to using a thermal-vibration composite end effector that combines a piezoelectric ceramic oscillator and a local laser heating module, the aforementioned local stress release excitation method can also use an ultrasonic transducer to replace the piezoelectric ceramic oscillator to provide high-frequency vibration excitation, and use a resistive local heating element to replace the local laser heating module to provide a local thermal gradient. A similar thermal-vibration composite stress release effect can be achieved through a low-cost hardware combination, which is suitable for low- to mid-range semiconductor special equipment manufacturing scenarios where equipment costs are more sensitive.

[0084] The above embodiments are merely preferred embodiments of the present invention. The semiconductor device special equipment manufacturing method of the present invention can also be applied to the manufacturing process of other electronic special equipment with high requirements for assembly accuracy and dynamic consistency, such as integrated circuit production line equipment, insulated gate bipolar transistor production line equipment, magnetron sputtering equipment, and excimer laser annealing equipment. All equivalent substitutions or improvements made within the technical concept of the present invention should be covered within the protection scope of the present invention.

[0085] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for manufacturing a semiconductor device-specific equipment, used for assembling and manufacturing a projection lens assembly and a dual-stage base for the semiconductor device-specific equipment, characterized in that... include: Before final fastening after pre-assembly, the in-situ data acquisition unit is used to collect the optical path wavefront distortion data and the vibration and positioning deviation data of the dual workpiece stage of the semiconductor device special equipment in situ. Combined with the pre-established whole machine digital twin model, the measured distortion vector of the whole machine is obtained. Based on the measured distortion vector of the whole machine and the pre-established forward mapping function, the estimated value of the residual stress distribution that causes the measured distortion vector of the whole machine is obtained by inverse solution, and the position coordinates and stress values ​​of the stress nodes are determined according to the estimated value of the residual stress distribution. Before final tightening, a high-frequency vibration excitation and a local thermal gradient field are simultaneously applied to the stress node using an end effector to form a thermal-vibration composite excitation. Through the synergistic effect of thermally induced material softening and vibration-induced micro-dislocation slip, the residual stress at the stress node is actively released until the residual stress at the stress node drops to within the preset stress release target threshold. Based on the prediction results of the overall dynamic characteristics of the digital twin model, the target stiffness of each mounting node is determined, and the equivalent local stiffness of the mounting node is adjusted to the target stiffness by adjusting the control voltage of the piezoelectric micro-motion pad array arranged at the mounting node. After stress relief and stiffness adjustment are completed, the semiconductor device-specific equipment is finally tightened, and the overall performance of the machine after final tightening is retested and accepted.

2. The method for manufacturing semiconductor device-specific equipment according to claim 1, characterized in that, The in-situ data acquisition unit includes a multi-wavelength interferometer arranged at key nodes of the optical path and a multi-dimensional acceleration sensor array arranged at key nodes of the movement of the dual workpiece stage. The multi-wavelength interferometer is used to acquire wavefront distortion data of the optical path, and the multi-dimensional acceleration sensor array is used to acquire vibration and positioning deviation data of the dual workpiece stage. The digital twin model includes geometric models of each component, material constitutive models, contact models of mating surfaces, and assembly boundary condition models.

3. The method for manufacturing semiconductor device-specific equipment according to claim 1, characterized in that, The positive mapping function is established based on the finite element method and is used to describe the mapping relationship between the residual stress distribution of the bonding surface and the optical path wavefront distortion and the positioning deviation of the dual workpiece stage. The reverse solution includes: constructing an optimization function with the objective of minimizing the sum of squared residuals between the predicted distortion and the measured distortion vector of the whole machine and superimposing a regularization constraint term; solving the optimization function using an iterative method combining gradient descent and sensitivity matrix until the residuals converge to within a preset threshold to obtain the estimated value of the residual stress distribution.

4. The method for manufacturing semiconductor device-specific equipment according to claim 1, characterized in that, The end effector includes an actuator body housing, a piezoelectric ceramic oscillator module and a local laser heating module integrated inside the actuator body housing, a flexible contact end attached to the surface of the stress node, and a stress monitoring sensor integrated near the flexible contact end. The piezoelectric ceramic oscillator module is used to generate the high-frequency vibration excitation, and the local laser heating module is used to generate the local thermal gradient field. The output excitations of the piezoelectric ceramic oscillator module and the local laser heating module are synchronously applied to the stress node surface through the flexible contact end.

5. The method for manufacturing semiconductor device-specific equipment according to claim 4, characterized in that, It also includes: using the stress monitoring sensor to collect the real-time value of the residual stress at the stress node, and determining whether the real-time value has dropped to within the stress release target threshold; If the stress does not drop below the target stress release threshold, at least one of the parameters—vibration amplitude, vibration frequency, and thermal gradient—is adjusted by a preset step size, and the combined thermal-vibration excitation is applied repeatedly until the real-time value drops below the target stress release threshold.

6. The method for manufacturing semiconductor device-specific equipment according to claim 1, characterized in that, The piezoelectric micro-motion pad array is arranged at the joint surface between the upper surface of the mounting base and the upper assembled component. Multiple piezoelectric micro-motion pad units are evenly arranged circumferentially along the mounting node. Each piezoelectric micro-motion pad unit is based on piezoelectric ceramic material and is connected to a control voltage source through an independent lead. It is used to adjust its own thickness and local stiffness after the control voltage is applied.

7. The method for manufacturing semiconductor device-specific equipment according to claim 6, characterized in that, The determination of the target stiffness and the adjustment of the control voltage include: determining the target stiffness of each of the installation nodes based on the simulation prediction results of the overall dynamic characteristics of the machine by the digital twin model; Read the reference stiffness of the piezoelectric micro-motion pad array at each of the installation nodes and the stiffness adjustment coefficient of each of the piezoelectric micro-motion pad units, combine the target stiffness to solve the control amount required to be applied to each of the piezoelectric micro-motion pad units, and apply the solved control amount to the corresponding piezoelectric micro-motion pad unit. The adjusted overall dynamic response is retested using the in-situ data acquisition unit, and the retest results are fed back to the digital twin model to determine whether the target dynamic index is met. If not, the control quantity is repeatedly adjusted until the target dynamic index is met.

8. The method for manufacturing semiconductor device-specific equipment according to claim 1, characterized in that, The retesting and acceptance of the overall performance after final tightening includes: using the in-situ data acquisition unit to retest the wavefront distortion data of the optical path of the overall machine after final tightening and the positioning accuracy of the dual workpiece stage, comparing the retested data with the preset overall machine factory acceptance indicators, and judging whether the overall performance meets the acceptance requirements. If the acceptance requirements are not met, at least one of the four steps described in claim 1—in-situ data acquisition and whole-machine measured distortion vector acquisition, stress node position coordinates and stress value determination, active stress release, and local stiffness adjustment—must be repeated until the overall machine performance meets the acceptance requirements.

9. The method for manufacturing semiconductor device-specific equipment according to claim 3, characterized in that, The forward mapping function is a data-driven proxy model trained based on historical assembly data. The data-driven proxy model is used to replace the forward mapping function established based on the finite element method for the reverse solution.

10. The method for manufacturing semiconductor device-specific equipment according to claim 4, characterized in that, The high-frequency vibration excitation is generated by an ultrasonic transducer, and the local thermal gradient field is generated by a resistive local heating element. The ultrasonic transducer and the resistive local heating element are integrated inside the end effector and act synchronously on the stress node surface through the flexible contact end.

Images