Research On Three-dimensional Distortion Self-calibration For Confocal Microscopic Measurement

Confocal microscopy is widely used in the fields of material surface structure measurement,micro-system and micro-device processing quality control due to its highly sensitive longitudinal tomography capability and better lateral resolution than ordinary optical microscopes.The confocal microscope is an optical scanning imaging system,and the three-dimensional(3D)distortion caused by the optical module and the scanning module is one of the main factors leading to the loss of its measurement accuracy.Therefore,the 3D distortion correction is a necessary prerequisite to ensure the high-precision detection of the 3D micro-scale with the confocal microscopy measurement system.Traditional distortion correction methods require high-precision calibrated templates to realize the 3D distortion correction of confocal microscopy,and the correction of lateral distortion and axial distortion is achieved through the two separate measurement processes,which leads to a complex correction process and high cost.In addition,during the correction process,the accuracy of the 3D centroid extraction of the groove structure on the grating template is a key factor affecting the effect of the 3D distortion correction.And the accuracy of the centroid extraction depends on the positioning accuracy of the groove edge.However,the traditional edge setting method ignores the influence of the groove height on the imaging pupil occlusion,which reduces edge positioning accuracy and worsens the result of 3D distortion correction.In view of the above-mentioned problems in the realization of the 3D distortion correction for confocal microscopy,the subject has carried out the research on width determination and edge setting of the deep groove for confocal microscopic measurement,and the 3D distortion self-calibration for the confocal microscopy.The main contents are as follows:Aiming at the problem of inaccurate positioning of the deep groove edge caused by the traditional edge setting method ignoring the influence of deep groove height structure on pupil occlusion,a width determination and edge setting method based on deep groove edge response imaging model is proposed.First,based on the occluded aperture point spread function and the defocus point spread function,the deep groove edge response imaging model is established to obtain the normalized intensity value at the deep groove edge as the edge judgment basis.Then,the in-focus scan image of the bottom surface of the groove is extracted from the 3D measurement of the deep groove and the normalization operation on it is performed.According to the edge judgment basis,the accurate positioning of the deep groove edge is realized.The proposed width determination and edge setting method based on the deep groove edge response imaging model provides support for the accurate extraction of the 3D centroid of grooves on the grating template in the 3D distortion correction.The experimental results show that,compared with the traditional 1/4 edge setting method,using the deep groove edge setting method can get more accurate width setting results.The width setting error is reduced by 52.2%.The standard deviation is reduced by23.2%,and the repeatability of edge positioning is better.Aiming at the problem that the existing self-calibration methods only focus on the correction of discrete feature points,and do not realize synchronous correction for measurement points other than the feature points,the self-calibration method based on a regularization neural network is proposed.Firstly,the objective function of the self-calibration method is constructed by taking the rigid body characteristics of the grating template as the constraint condition.Then,the neural network model is introduced to realize the mapping between the coordinates of the measurement points and the system errors.In order to alleviate the over-fitting problem of the neural network model for the discrete feature points,a regularization term is added to the objective function,and the stop condition is given.This method realizes the synchronous correction of any measuring point in the measurement range and lays a foundation for the realization of the 3D distortion self-calibration for confocal microscopic measurement.Finally,the influences of the translation distance at the translational pose and the number of measurement poses on the performance of the regularization neural network-based self-calibration method are analyzed.The simulation results show that choosing a translation distance of 1/2 pitch can alleviate the over-fitting phenomenon.Besides,the increase in the number of measurement poses can reduce the correction error,and the number of measured poses is recommended to be greater than or equal to 5 in practical applications.Aiming at the problems of relying on the high-precision calibration of the grating template and the discrete correction of lateral and axial distortions when the traditional distortion correction method is used to realize the 3D distortion correction for the confocal microscopic measurement system,a three-dimensional distortion self-calibration method based on regularization neural network is carried out.Firstly,base on the analysis of distortion spectrum and the limited energy loss decoupling method,a method for determining the parameters of the grating template suitable for3 D distortion correction is proposed to realize the selection of the grating template.Secondly,the objective function of the 3D distortion self-calibration method is established based on the constraint of the rigid body characteristics of the grating template.By training the 3D distortion self-calibration neural network model,the model has the ability to fit the mapping relationship between the distorted space and the undistorted space.The 3D distortion correction for confocal microscopic measurement is achieved.Finally,the simulation experiment focuses on the analysis of the influence of the distortion type and size,and the centroid extraction errors on the correction effect.The simulation experiment results show that this method can achieve nano-level correction accuracy without centroid extraction errors.Besides,the centroid extraction errors affect the accuracy of 3D distortion correction.With the increase of centroid extraction errors,3D distortion correction errors show an increasing trend,and the correction errors are of the same magnitude as the centroid extraction errors.The experimental verification of the three-dimensional distortion selfcalibration method based on the regularization neural network is carried out.The industrial confocal microscopic measurement system is used as the target to be corrected,and the grating with groove structures is used as the correction template.Firstly,the correction template is measured multiple times at different poses.Secondly,the proposed deep groove edge setting method is used to extract the 3D centroid coordinates of the grooves on the grating template.Finally,the 3D centroid coordinates are used as input data to train the three-dimensional distortion selfcalibration neural network model,and the 3D distortion correction for the industrial confocal microscopic measurement system is realized.The verification experiments of the 3D distortion correction effect are designed.The experimental results are as follows: taking the shape of the grating template measured by the atomic force microscope as the reference value,compared with the grating template shape calculated by the proposed method,the root mean square of the lateral correction errors and the axial correction errors are less than 50 nm and 6 nm,respectively.By measuring the standard step,it is found that after 3D distortion correction,the accuracy of the determination of the step height is increased to 3.6 times of that before the correction.By measuring the grating,it can be observed that the consistency of the pitches of the grating measured at different positions in the field of view is improved after 3D distortion correction.