Laser deburring and cleaning device for plastic package

Abstract

The present application relates to the technical fields of semiconductor device processing, in particular to a plastic package laser overflow removal and cleaning equipment, comprising: an image acquisition module; a region calibration module; an edge blur determination module for determining the average edge blur of the active surface area of the chip; an abnormal preliminary screening module for determining the abnormal preliminary screening category of the active surface area of the chip according to the average edge blur; a residual glue detection module; a residual glue determination module for determining whether the suspected normal area is a residual glue area according to the double-wavelength reflectivity ratio; a feature recognition module for determining the gray scale uniformity index of the suspected overflow abnormal area; an overflow classification module for determining whether the suspected overflow abnormal area is a pure overflow area or an overflow and residual glue mixture area according to the gray scale uniformity index; a position determination module; a strategy determination module; and a laser processing module, thereby improving the cleaning efficiency of the plastic package laser overflow removal and cleaning equipment.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device processing technology, specifically to a laser-based device for removing overflow and cleaning plastic encapsulations. Background Technology

[0002] Laser cleaning is an advanced surface treatment technology that uses a high-energy laser beam to irradiate the surface of an object, causing impurities, contaminants, or coatings to evaporate or peel off rapidly through photophysical and photochemical effects. In the semiconductor packaging field, laser overflow removal technology has become the mainstream solution to replace traditional chemical cleaning due to its non-contact, high-precision (capable of removing micron-level overflow) and stress-free characteristics. Initially, this technology was mainly used for removing overflow from the pins of high-end single-chip packages such as BGA and CSP to prevent short circuits between high-density pins.

[0003] With the development of advanced packaging technologies, fan-out wafer-level packaging has become mainstream. In the face-down process, the chip is mounted on a temporary carrier with its active side facing down and encapsulated with epoxy molding compound. After debonding and flipping to reconstruct the wafer, the exposed active side of the chip needs to be laser-cleaned to remove EMC overflow from high voltage penetration and transparent organic residue left by debonding, thereby ensuring the yield of subsequent redistribution layer electroplating.

[0004] However, existing conventional visual recognition solutions are difficult to handle such multi-material composite surfaces: on the one hand, conventional height gradient or grayscale threshold algorithms are easily affected by global wafer warping, making it impossible to accurately define the overflow edge, and there are serious blind spots for transparent residual adhesive without height difference, leading to missed detection and insufficient electroplating adhesion; on the other hand, existing solutions lack in-depth analysis of the micro-texture and spatial position of contaminants, making it impossible to effectively identify the mixed area where overflow and residual adhesive coexist, and they do not consider the distance risk between contaminants and sensitive pads, often forcing the use of high-energy coverage cleaning, which can easily lead to carbonization of the surface layer of the mixed area and thermal damage to the pads in the high-risk interface area. Summary of the Invention

[0005] To address this, the present invention provides a laser-based material removal and cleaning device for molded bodies, which overcomes the problems of poor cleaning efficiency in existing technologies that do not consider the differences in morphology and spectrum of overflow and residual adhesive, do not consider the spatial location risks of defects and pads, resulting in high-risk areas being prone to damage to pads, flat and transparent residual adhesive being easily missed, and mixed areas being unable to be cleaned in a targeted manner.

[0006] To achieve the above objectives, the present invention provides a laser-assisted cleaning and removal device for molded plastic seals, comprising: Image acquisition module, which is used to acquire grayscale images and height maps of the reconstructed wafer; A region calibration module is used to determine the active chip surface region of the reconstructed wafer based on the grayscale image and the height image; An edge blur determination module is used to determine the average edge blur of the active surface area of ​​the chip based on the height value of each pixel position within the active surface area of ​​the chip. An anomaly screening module is used to determine the anomaly screening category of the active surface area of ​​the chip based on the average edge ambiguity value. The anomaly screening category includes suspected normal area and suspected overflow anomaly area. A residual adhesive detection module is used to determine the dual-wavelength reflectance ratio of the suspected normal area; The residual adhesive determination module is used to determine whether the suspected normal area is an area with residual adhesive based on the dual-wavelength reflectance ratio. The feature recognition module is used to determine the gray uniformity index of the suspected overflow abnormal area based on the pixel gray values ​​of the suspected overflow abnormal area in the grayscale image. The overflow classification module is used to determine whether the suspected overflow abnormal area is a pure overflow area or a mixture of overflow and residual glue area based on the gray uniformity index. The location determination module is used to determine the location risk category of the area to be cleaned, which consists of the area with residual adhesive, the area with pure overflow, and the area with a mixture of overflow and residual adhesive. The location risk category includes safe areas and high-risk areas. The strategy determination module is used to determine the laser cleaning strategy for the area to be cleaned based on the category and location risk category of the area to be cleaned; A laser processing module for performing laser ablation on the surface of the reconstructed wafer according to the laser cleaning strategy.

[0007] Preferably, the anomaly screening module determines the active surface area of ​​the chip as a suspected overflow anomaly area when the mean edge ambiguity of the active surface area of ​​the chip is less than a preset edge ambiguity threshold. The anomaly screening module determines that the active surface area of ​​the chip is a suspected normal area if the mean edge ambiguity value is greater than or equal to a preset edge ambiguity threshold.

[0008] Preferably, the edge ambiguity determination module includes: A pixel blur calculation unit is used to calculate the edge blur of the pixel based on the height value of the pixel position in the height map, wherein the edge blur is the ratio of a first height difference to a second height difference. The first height difference is the absolute value of the height difference between adjacent pixel positions on both sides of the pixel in the height map along the image row coordinate direction, and the second height difference is the absolute value of the height difference between two pixel positions on both sides of the pixel in the height map along the image row coordinate direction, separated by one pixel. The grid mean calculation unit is used to divide the active surface area of ​​the chip into several grids and take the arithmetic mean of the edge ambiguity of all pixels in the grid, which is denoted as the edge ambiguity mean of the grid. The region mean calculation unit is used to take the arithmetic mean of the edge ambiguity of all grids in the active surface area of ​​the chip, and denoted as the edge ambiguity mean of the active surface area of ​​the chip.

[0009] Preferably, the region calibration module is used to determine the active surface region of the chip in the reconstructed wafer, wherein... The region calibration module identifies connected components that simultaneously satisfy preset reflectivity, preset height, and preset pad shape conditions from the grayscale and height maps as pad imaging regions, and extracts the centroids of these connected components, marking them as reliable pad points. After spatial clustering of the reliable pad points, each cluster is graph-matched with the chip pad layout to determine the chip-internal coordinates of each reliable pad point. Based on the chip-internal coordinates and image coordinates of the successfully matched pad points, the spatial transformation relationship is calculated, and the chip active surface design boundary is mapped to global image coordinates through the spatial transformation relationship to obtain the chip active surface region. The preset reflectivity condition is that the average gray value of the connected domain is greater than the preset pad gray value threshold. The preset height condition is that the difference between the average height of the connected region and the reference height of the pad is less than the preset height difference. The preset pad shape condition is that the ratio of the area of ​​the connected region to the area of ​​the preset pad is within a preset area ratio range, and the difference between the aspect ratio of the minimum bounding rectangle of the connected region and the aspect ratio of the preset pad is less than a preset ratio threshold.

[0010] Preferably, the residual adhesive determination module determines the suspected normal area as a normal area without residual adhesive when the ratio of the dual-wavelength reflectance of the suspected normal area is less than a preset dual-wavelength reflectance ratio. The residual adhesive determination module determines the suspected normal area as an area with residual adhesive in response to the dual-wavelength reflectivity ratio being greater than or equal to a preset dual-wavelength reflectivity ratio.

[0011] Preferably, the residual adhesive detection module is used to determine the dual-wavelength reflectance ratio of the suspected normal area, wherein... The residual adhesive detection module obtains the first reflectance of each grid under the first preset wavelength laser irradiation and the second reflectance under the second preset wavelength laser irradiation, and records the ratio of the first reflectance to the second reflectance as the dual-wavelength reflectance ratio of the suspected normal area.

[0012] Preferably, the overflow classification module determines the suspected overflow abnormal area as a pure overflow area in response to the gray uniformity index of the suspected overflow abnormal area being less than a preset uniformity threshold. The overflow classification module determines that the suspected overflow abnormal area is a mixture of overflow and residual adhesive if the grayscale uniformity index of the suspected overflow abnormal area is greater than or equal to the preset uniformity threshold.

[0013] Preferably, the feature recognition module is used to determine the grayscale uniformity index of the suspected overflow anomaly area, wherein... The feature recognition module obtains the pixel grayscale values ​​of the suspected overflow anomaly area in the grayscale image, calculates the ratio of the standard deviation to the mean of all pixel grayscale values ​​in the suspected overflow anomaly area, and determines the ratio as the grayscale uniformity index of the suspected overflow anomaly area.

[0014] Preferably, the location determination module determines the area to be cleaned as a high-risk area when the safe distance to the area to be cleaned is less than a preset safe distance threshold; the location determination module determines the area to be cleaned as a safe area when the safe distance to the area to be cleaned is greater than or equal to the preset safe distance threshold. The safe distance of the area to be cleaned is the ratio of the shortest Euclidean distance between the pixel point on the edge contour of the area to be cleaned and all the pad imaging areas to the effective radius of the preset laser spot; wherein, the area with residual adhesive, the area with pure overflow, and the area with a mixture of overflow and residual adhesive are referred to as the area to be cleaned.

[0015] Preferably, the strategy determination module responds to the fact that the area to be cleaned is an area with residual adhesive and the location risk category is a safe area, and cleans with a first preset energy until the average gray value is greater than a preset gray value threshold and then terminates. In response to the fact that the area to be cleaned is an area with residual adhesive and the location risk category is a high-risk area, the cleaning is terminated when the average gray value is greater than the preset gray value threshold using the second preset energy. In response to the fact that the area to be cleaned is a pure overflow area and the location risk category is a safe area, the cleaning is terminated when the average height is less than a preset height threshold using a third preset energy. In response to the fact that the area to be cleaned is a pure overflow area and the location risk category is a high-risk area, the cleaning is terminated when the average height is less than the preset height threshold using the fourth preset energy. In response to the fact that the area to be cleaned is a mixture of overflow and residual adhesive and the location risk category is a safe zone, the first preset energy is used for cleaning. After the average gray value is greater than the preset gray value threshold, the third preset energy is used for cleaning until the average height of the area to be cleaned is less than the preset height threshold. In response to the fact that the area to be cleaned is a mixture of overflow and residual adhesive and the location risk category is a high-risk area, the second preset energy is used to clean until the average gray value of the area to be cleaned is greater than the preset gray value threshold, and then the fourth preset energy is used to clean until the average height of the area to be cleaned is less than the preset height threshold, and then the process is terminated. Wherein, the grayscale mean is the arithmetic mean of the grayscale values ​​of all pixels in the area to be cleaned, the height mean is the arithmetic mean of the height values ​​of all pixels in the area to be cleaned, the first preset energy is less than the third preset energy, the second preset energy is determined according to the safety distance of the area to be cleaned, and the smaller the safety distance, the smaller the second preset energy; the fourth preset energy is determined according to the safety distance of the area to be cleaned, and the smaller the safety distance, the smaller the fourth preset energy; and, for the same safety distance, the second preset energy is less than the first preset energy, and the fourth preset energy is less than the third preset energy.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention acquires grayscale and height information of the reconstructed wafer through multimodal imaging, uses morphological ambiguity for initial screening and separation, uses dual-wavelength reflectivity to detect flat residual adhesive, and uses grayscale uniformity to distinguish pure overflow from mixtures. It performs progressive and fine classification of overflow and residual adhesive on the active surface of the chip, and combines positional risk assessment based on safe distance to match differentiated laser energy and cleaning termination conditions for six types of cleaning scenarios, thereby improving the cleaning efficiency of laser overflow removal and cleaning equipment for encapsulated products.

[0017] 2. This invention achieves rapid initial screening of whether there are significant morphological anomalies in the active surface area by comparing the average edge ambiguity value with a preset edge ambiguity threshold. The area is divided into suspected normal areas and suspected overflow abnormal areas, thereby providing accurate entry diversion for subsequent differentiated fine detection and avoiding blind traversal of the entire area by a single detection method.

[0018] 3. This invention identifies reliable solder pads by integrating preset conditions of grayscale, height and shape, and maps the active surface design boundary to image coordinates based on graph matching and spatial transformation, thus achieving accurate calibration of the active surface area of ​​the chip under interference from overflow and residual adhesive.

[0019] 4. This invention compares the ratio of the dual-wavelength reflectance of the suspected normal area with a preset value, thereby enabling the detection of transparent residual adhesive with a smooth surface morphology but material spectral characteristics different from the substrate. This compensates for the blind spot of pure height morphology detection for smooth films, thus effectively avoiding defects such as subsequent electroplating open circuits or insufficient adhesion caused by missed detection of residual adhesive.

[0020] 5. This invention sets differentiated cleaning termination conditions based on grayscale average or height average for areas with residual adhesive, pure overflow, and a mixture of overflow and residual adhesive. Different preset energies are configured for safe and high-risk areas. The second and fourth preset energies in the high-risk area dynamically decrease with decreasing safety distance and remain lower than the corresponding safe area energy at the same safety distance. Simultaneously, the mixture area is strictly cleaned in a step-by-step sequence of removing residual adhesive first, followed by overflow. This achieves multi-dimensional precise matching of laser cleaning process parameters with the characteristics of defective materials, removal difficulty, and spatial risk level. This ensures that high-risk areas near solder pads are always cleaned safely with lower energy, while the mixture area is cleaned step-by-step, removing two contaminants with vastly different physicochemical properties. This thoroughly removes various plastic encapsulation residues while minimizing the risk of thermal damage to solder pads, ensuring the electroplating yield and packaging reliability of subsequent redistribution layers. Attached Figure Description

[0021] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0022] Figure 1 This is a schematic diagram of the module connection of the laser descaling and cleaning equipment for the encapsulated body according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process of determining the initial screening category of anomalies in the active surface area of ​​a chip based on the mean edge ambiguity, as described in an embodiment of the present invention. Figure 3 This is a flowchart illustrating how the suspected normal area is a residual adhesive area based on the dual-wavelength reflectivity ratio, according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating how the suspected overflow abnormal area is determined as a pure overflow area or a mixture of overflow and residual adhesive based on the grayscale uniformity index, according to an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 , Figure 2 , Figure 3 as well as Figure 4The diagrams shown are: a schematic diagram of the module connection of the laser de-overflow and cleaning equipment for the encapsulated body according to an embodiment of the present invention; a flowchart of the present invention for determining the abnormal initial screening category of the active surface area of ​​the chip based on the mean edge ambiguity value; a flowchart of the present invention for determining whether the suspected normal area is a residual adhesive area based on the dual-wavelength reflectivity ratio; and a flowchart of the present invention for determining whether the suspected overflow abnormal area is a pure overflow area or a mixture of overflow and residual adhesive based on the grayscale uniformity index.

[0025] The present invention provides a laser-based device for removing excess material from plastic sealant and for cleaning it, comprising: Image acquisition module, which is used to acquire grayscale images and height maps of the reconstructed wafer; A region calibration module, which is connected to the image acquisition module, is used to determine the active chip surface region of the reconstructed wafer based on the grayscale image and the height image; An edge blur determination module, which is connected to the image acquisition module and the region calibration module respectively, is used to determine the average edge blur value of the active surface area of ​​the chip based on the height value of each pixel position within the active surface area of ​​the chip. An anomaly screening module, which is connected to the edge ambiguity determination module, is used to determine the anomaly screening category of the active surface area of ​​the chip based on the mean edge ambiguity value. The anomaly screening category includes suspected normal area and suspected overflow anomaly area. The residual adhesive detection module, which is connected to the abnormal screening module, is used to determine the dual-wavelength reflectance ratio of the suspected normal area; The residual adhesive determination module is connected to the residual adhesive detection module and is used to determine whether the suspected normal area is an area with residual adhesive based on the dual-wavelength reflectance ratio. The feature recognition module is connected to the anomaly screening module and is used to determine the gray uniformity index of the suspected overflow anomaly area based on the pixel gray values ​​of the suspected overflow anomaly area in the grayscale image. The overflow classification module is connected to the feature recognition module and is used to determine whether the suspected overflow abnormal area is a pure overflow area or a mixture of overflow and residual glue area based on the gray uniformity index. The location determination module is connected to the residual adhesive determination module, the overflow classification module and the area calibration module respectively. It is used to determine the location risk category of the area to be cleaned. The area to be cleaned consists of the area with residual adhesive, the area with pure overflow and the area with a mixture of overflow and residual adhesive. The location risk category includes safe area and high-risk area. The strategy determination module is connected to the residual adhesive determination module, the overflow classification module and the location determination module respectively. It is used to determine the laser cleaning strategy of the area to be cleaned based on the category and location risk category of the area to be cleaned. A laser processing module, connected to the strategy determination module, is used to perform laser ablation on the surface of the reconstructed wafer according to the laser cleaning strategy.

[0026] Specifically, the image acquisition module controls the 3D laser profilometer to scan the reconstructed wafer surface, simultaneously acquiring a 2D grayscale image (8-bit, 0-255 gray levels) with a resolution of 2μm / pixel and a 3D height image with a Z-axis accuracy of 0.5μm; the laser processing module uses a 355nm ultraviolet picosecond laser.

[0027] Specifically, the specific structure of the region calibration module, edge ambiguity determination module, residual adhesive detection module, residual adhesive judgment module, feature recognition module, overflow classification module, location judgment module, and strategy determination module is not limited. Each module and its units can be composed of logic components, including field-programmable components, computers, or microprocessors in computers.

[0028] Specifically, the anomaly screening module determines that the active surface area of ​​the chip is a suspected overflow anomaly area when the average edge ambiguity of the active surface area of ​​the chip is less than a preset edge ambiguity threshold of 0.55. The anomaly screening module determines that the active surface area of ​​the chip is a suspected normal area if the mean edge ambiguity value is greater than or equal to a preset edge ambiguity threshold.

[0029] Specifically, edge blur reflects the sharpness of surface height changes. As an irregular accumulation of molding compound, overflow has significant height abrupt changes at its boundaries, resulting in a small ratio between the height difference of adjacent pixels and the height difference of a pixel apart, leading to a lower average edge blur in the region. In contrast, active surface areas covered only with transparent residue or with clean surfaces have gentle height changes, with the two height differences being of similar magnitude and the ratio close to 1, thus resulting in a higher average edge blur.

[0030] In this embodiment, the preset edge ambiguity threshold is set to 0.55. One hundred reconstructed wafers that have undergone molding and debonding processes are selected as the verification sample set. Among them, 50 wafers were confirmed by microscopy and 3D topology analysis to have obvious overflow defects, 30 wafers were confirmed by Fourier transform infrared spectroscopy to have only transparent organic residue but no abnormal topology, and 20 wafers were confirmed to have clean, defect-free surfaces. The edge ambiguity determination module of this device calculates the average edge ambiguity of the active surface area of ​​the chip for each sample. Statistically, the average edge ambiguity of samples with overflow defects is distributed in the range of 0.15 to 0.35, the average edge ambiguity of samples with only residue and no overflow is distributed in the range of 0.58 to 0.82, and the average edge ambiguity of clean, defect-free samples is distributed in the range of 0. Between 0.62 and 0.88, there is a clear dividing gap between the distribution range of the mean edge ambiguity of the two groups of samples without surface abnormalities and the distribution range of the mean edge ambiguity of the samples with overflow defects, which is between 0.35 and 0.58. Based on this, the receiver operating characteristic (ROC) curve analysis method was used, with overflow defects as the positive class and samples with only residual glue and clean samples as the negative class. The optimal cutoff point was determined with the goal of maximizing the Youden index. It was calculated that when the preset edge ambiguity threshold was 0.55, the Youden index reached its maximum value, and the classification accuracy was 96.7%. Among them, the detection sensitivity of overflow defects was 96.0% and the specificity was 97.5%. The preset edge ambiguity threshold is exactly located in the middle of the distribution gap between the two types of samples, which takes into account both high sensitivity and low false detection rate. The above experimental data were obtained under the conditions of a 3D laser profilometer (XY resolution 2μm / pixel, Z-axis accuracy 0.5μm) and a 64×64 pixel grid division used in this embodiment. For implementation methods with different hardware configurations or grid sizes, those skilled in the art can recalibrate the value of the preset edge blur threshold using the same experimental method described above.

[0031] Specifically, the edge ambiguity determination module includes: The pixel blur calculation unit obtains the height values ​​H(i-1,j) and H(i+1,j) of the adjacent pixel positions on both sides of the pixel P(i,j) along the image row coordinate direction in the height map, and calculates the first height difference: Δ1=|H(i-1,j)-H(i+1,j)|.

[0032] Obtain the height values ​​H(i-2,j) and H(i+2,j) of pixel P(i,j) at two positions one pixel apart along the row coordinate direction of the image in the height map, and calculate the second height difference: Δ2=|H(i-2,j)-H(i+2,j)|; the edge blur of pixel P(i,j) is F(i,j)=Δ1 / Δ2; The grid mean calculation unit is used to divide the active surface area of ​​the chip into several grids (in this embodiment, the active surface area of ​​the chip is uniformly divided into 64×64 pixel grids) and take the arithmetic mean of the edge ambiguity of all pixels in the grid, which is recorded as the edge ambiguity mean of the grid. The region mean calculation unit is used to take the arithmetic mean of the edge ambiguity of all grids in the active surface area of ​​the chip, and denoted as the edge ambiguity mean of the active surface area of ​​the chip.

[0033] Specifically, the region calibration module is used to determine the active surface region of the chip in the reconstructed wafer, wherein, The region calibration module identifies connected components that simultaneously satisfy preset reflectivity, height, and pad shape conditions from the grayscale and height maps as pad imaging regions, and extracts the centroids of these connected components, marking them as reliable pad points. It then precisely matches the actual pad points in these images with the theoretical pad points in the chip design drawings to calculate the spatial transformation relationship. To achieve this, this embodiment employs a two-dimensional point set affine transformation registration algorithm based on RANSAC (Random Sample Consensus). The RANSAC-based 2D point set affine transformation registration algorithm takes as input an image pad point set (i.e., the set of 2D image coordinates of all reliable pad points extracted from the grayscale and height maps) and a standard pad point set (i.e., the set of 2D coordinates of theoretical pads obtained from the chip design CAD drawings); its output is an optimal affine transformation matrix (containing translation, rotation, and scaling parameters) and successfully matched interior point pad pairs (i.e., valid matching point pairs between image pad points and standard pad points that satisfy the transformation relationship); after obtaining the optimal affine transformation matrix, the region calibration module maps the chip active surface design boundary (i.e., the active area polygon outline in the CAD drawings) to global image coordinates through the optimal affine transformation matrix, thereby obtaining an accurate chip active surface region.

[0034] Specifically, the execution process of the RANSAC-based two-dimensional point set affine transformation registration algorithm is as follows: Within a preset maximum number of iterations (1000), three point pairs are randomly selected from the input point set each time, and an initial affine transformation matrix is ​​calculated. Then, the reprojection error of all input point pairs under this matrix is ​​calculated, and point pairs with errors less than a preset reprojection error threshold of 5 pixels are recorded as inliers. Finally, the affine transformation matrix corresponding to the iteration with the most inliers is output as the optimal affine transformation matrix. The above-described RANSAC-based point set registration and matrix solving process is a common technique used by those skilled in the art and will not be elaborated upon further.

[0035] The preset reflectivity condition is that the average gray value of the connected domain is greater than the preset pad gray value threshold of 180. In this embodiment, a 3D laser profilometer with a built-in calibration light source is used for coaxial illumination. The light source conditions are fixed. The reflectivity of copper or aluminum pads in the visible light band is significantly higher than that of the molding compound EMC substrate. According to the gray value statistics of known pad samples, the average gray value of the pad area is concentrated in the range of 190 to 230, and that of the EMC substrate is concentrated in the range of 40 to 90. Taking 180 can form a clear boundary between the two and does not change with the ambient light. The preset height condition is that the difference between the average height of the connected region and the pad reference height of 5μm is less than the preset height difference of 3μm. It can be understood that for different chip products, the pad reference height can be directly read from the corresponding parameters in its design file. The preset height difference is 3μm, which is based on the fact that the standard deviation of the height fluctuation on the surface of the clean pad calibrated by a white light interferometer is about 1μm. Taking 3 times the standard deviation as the tolerance range can effectively distinguish the clean pad from the raised area whose height is significantly increased after being covered by overflow material. The preset pad shape condition is that the ratio of the area of ​​the connected region to the area of ​​the preset pad (400 pixels, calculated from the typical pad size of 20μm×20μm in the chip design file and converted to the XY resolution of 2μm / pixel of this device) is within the preset area ratio range of 0.8 to 1.2, and the difference between the aspect ratio of the minimum bounding rectangle of the connected region and the aspect ratio of the preset pad (1.0, since the pad is usually designed as a square, the value of the aspect ratio of the preset pad is directly taken from the chip design file) is less than the preset ratio threshold of 0.15 (under the resolution conditions of this device, the aspect ratio measurement fluctuation of the reliable imaging pad does not exceed 0.12, and 0.15 is taken to leave a margin and effectively exclude non-pad connected regions with severe shape distortion).

[0036] Specifically, the residual adhesive determination module determines the suspected normal area as a normal area without residual adhesive when the ratio of the dual wavelength reflectance of the suspected normal area is less than the preset dual wavelength reflectance ratio of 1.35. The residual adhesive determination module determines the suspected normal area as an area with residual adhesive in response to the dual-wavelength reflectivity ratio being greater than or equal to a preset dual-wavelength reflectivity ratio.

[0037] Specifically, the preset dual-wavelength reflectance ratio was 1.35. Thirty wafer samples from the anomaly screening module's verification sample set, confirmed by Fourier transform infrared spectroscopy to contain only transparent organic residue but without any morphological anomalies (referred to as the "residue-containing positive class"), and 20 wafer samples confirmed to have clean, defect-free surfaces (referred to as the "residue-free negative class") were selected for dual-wavelength reflectance ratio testing. The results showed that the dual-wavelength reflectance ratios of the clean, defect-free samples ranged from 1.05 to 1.22 (mean 1.12), while the dual-wavelength reflectance ratios of the samples containing only transparent organic residue ranged from 1.45 to 2. The mean value was 1.95. The two groups of samples had a clear dividing gap between the distribution intervals of 1.22 and 1.45. Based on this, the receiver operating characteristic (ROC) curve analysis method was used, with samples containing only residual adhesive as the positive class and clean samples as the negative class. The optimal cutoff point was determined with the goal of maximizing the Youden index. It was calculated that the Youden index reached its maximum value when the preset dual-wavelength reflectance ratio was 1.35, with a classification accuracy of 98.0%, of which the sensitivity for detecting residual adhesive was 96.7% and the specificity was 100%.

[0038] Specifically, when the surface of the suspected normal area is clean and free of adhesive residue, the substrate material is directly exposed. Its reflectivity at 850nm and 405nm is entirely determined by the substrate's own optical constants, and the two maintain a fixed intrinsic material ratio. The calculated dual-wavelength reflectivity ratio is stable near a low reference constant and is necessarily less than the preset dual-wavelength reflectivity ratio. However, when the surface is covered with a transparent adhesive film, the 850nm near-infrared light, as the molecule, has extremely strong penetrating power and penetrates the adhesive residue almost without damage, keeping the first reflectivity basically unchanged. However, the 405nm violet light, as the denominator, is significantly attenuated due to the intrinsic absorption of the organic material and the destructive interference effect occurring on the upper and lower surfaces of the adhesive film, thus becoming greater than or equal to the preset dual-wavelength reflectivity ratio.

[0039] Specifically, the residual adhesive detection module is used to determine the dual-wavelength reflectance ratio of the suspected normal area, wherein, The residual adhesive detection module acquires the first reflectance of each grid under the first preset wavelength of 850nm laser irradiation and the second reflectance under the second preset wavelength of 405nm laser irradiation. During measurement, for each grid in the suspected normal area, the central area of ​​the grid is vertically irradiated with 850nm and 405nm lasers in sequence. The CMOS camera of the image acquisition module acquires the reflected light image respectively, extracts the average gray value of the 3×3 pixel area in the center of the grid, and uses the ratio of the average gray value to the reference gray value measured in advance on a standard white board under the same conditions as the reflectance at that wavelength, which is recorded as the first reflectance and the second reflectance respectively.

[0040] Specifically, the ratio of the first reflectance to the second reflectance is denoted as the dual-wavelength reflectance ratio of the suspected normal region. The first preset wavelength is assigned to 850nm near-infrared light, and the second preset wavelength is assigned to 405nm violet light. In this embodiment, 50 wafer samples with transparent organic residues of varying thicknesses (50 nm to 2 μm) confirmed by Fourier transform infrared spectroscopy were selected, along with 50 wafer samples with clean surfaces, for reflectance spectroscopy testing. The test data showed that organic residues commonly used in semiconductor packaging (such as epoxy resin) have extremely high transmittance and almost no intrinsic absorption in the 850 nm near-infrared band. 850 nm light waves can directly penetrate the extremely thin residue film, and its reflectance mainly reflects the true reflectance characteristics of the underlying substrate, with minimal influence from surface residues. Therefore, it is used as a reference signal. Organic polymers, on the other hand, have strong intrinsic absorption in the 405 nm violet band. When the residue thickness is in the range of hundreds of nanometers to micrometers, the short-wavelength 405 nm light will produce a strong thin-film interference effect on the upper and lower surfaces of the residue film, causing its reflectance to fluctuate drastically or decrease significantly. Therefore, it is used as a sensitive detection signal.

[0041] In this embodiment, the detection laser with a first preset wavelength of 850nm and a second preset wavelength of 405nm is provided by a low-power dual-wavelength semiconductor laser independent of the laser processing module. Its power is controlled below 1mW and is used only for reflectivity measurement, without participating in laser ablation processing.

[0042] Specifically, the overflow classification module determines that the suspected overflow abnormal area is a pure overflow area when the gray uniformity index of the suspected overflow abnormal area is less than the preset uniformity threshold of 0.12. The overflow classification module determines that the suspected overflow abnormal area is a mixture of overflow and residual adhesive if the grayscale uniformity index of the suspected overflow abnormal area is greater than or equal to the preset uniformity threshold.

[0043] In this embodiment, the preset uniformity threshold is set to 0.12. Fifty wafer samples with obvious overflow defects from the previously verified sample set (of which, as confirmed by microscopy and energy dispersive spectroscopy, 30 were pure overflow defects and 20 were mixtures of overflow and residual adhesive defects) were selected for grayscale uniformity index testing. Statistical analysis showed that the grayscale uniformity index of the pure overflow samples ranged from 0.04 to 0.09, while that of the overflow and residual adhesive mixture samples ranged from 0.15 to 0.28. There was a clear dividing gap between the two groups of samples in the range of 0.09 to 0.15. Based on this, receiver operating characteristic (ROC) curve analysis was used, with overflow and residual adhesive mixtures as the positive class and pure overflow as the negative class. The optimal cutoff point was determined with maximizing the Youden index as the optimization objective. Calculations showed that the Youden index reached its maximum value when the preset uniformity threshold was 0.12, with a classification accuracy of 98.0%. This preset uniformity threshold was precisely located in the middle of the gap between the two sample distributions.

[0044] Specifically, pure overflow mainly consists of epoxy molding compound with uniform composition. Its surface texture and optical scattering characteristics are highly consistent. In the grayscale image, it is characterized by concentrated grayscale distribution and small standard deviation. Therefore, the grayscale uniformity index is low and less than the preset uniformity threshold. However, when transparent or semi-transparent residual glue is mixed in the overflow, the film interference effect caused by the uneven thickness of the residual glue, as well as the exposure of the underlying morphology due to the light transmission of the residual glue, will cause drastic fluctuations in grayscale in local areas, resulting in an increase in standard deviation. Consequently, the grayscale uniformity index increases and becomes greater than or equal to the preset uniformity threshold.

[0045] Specifically, the feature recognition module is used to determine the grayscale uniformity index of the suspected overflow anomaly area, wherein, The feature recognition module obtains the pixel grayscale values ​​of the suspected overflow anomaly area in the grayscale image, calculates the ratio of the standard deviation to the mean of all pixel grayscale values ​​in the suspected overflow anomaly area, and determines the ratio as the grayscale uniformity index of the suspected overflow anomaly area.

[0046] Specifically, the location determination module determines the area to be cleaned as a high-risk area when the safe distance to the area to be cleaned is less than a preset safe distance threshold of 1.5; the location determination module determines the area to be cleaned as a safe area when the safe distance to the area to be cleaned is greater than or equal to the preset safe distance threshold.

[0047] In this embodiment, the preset safety distance threshold is set to 1.5. Thirty reconstructed wafer samples with known cleaning area locations are selected. The shortest Euclidean distance from the edge of the cleaning area to the edge of the nearest pad imaging area covers a range of 0 to 50 μm. These samples are divided into 5 groups of 6 wafers each. A fixed laser energy of 0.6 J / cm² is used. 2Under the condition of an effective spot radius of 10 μm, a single-scan cleaning was performed on each group of samples. After cleaning, the surface morphology of the pads was examined under a microscope, and any visible ablation marks or oxidation discoloration on the pad surface was considered damage. The experimental results showed that the pad damage rate of the sample group with a distance greater than 15 μm (i.e., a safe distance greater than 1.5) was zero, the damage rate of the sample group with a distance between 10 μm and 15 μm (safe distance 1.0 to 1.5) was 8.3%, and the damage rate of the sample group with a distance less than 10 μm (safe distance less than 1.0) reached 41.7%. Therefore, the safe distance threshold was determined to be 1.5.

[0048] Specifically, the edge contour of the area to be cleaned is obtained. The input to this step is a binary mask image of the area to be cleaned, and the output is a set of edge contour pixels of the area to be cleaned. This is implemented using the OpenCV machine vision library. Secondly, all pad imaging areas are acquired. This step uses the YOLOv5 object detection model. The input of YOLOv5 is a standard reference image of the wafer surface, and the output is the pad imaging area image with the ROI bounded. The training data of YOLOv5 is the image of the pad area in the standard reference image annotated with rectangular boxes using the labeling tool LabelImg. Next, the shortest Euclidean distance is calculated. The input to this step is the set of edge contour pixels and the binarized mask image of the pad imaging region, and the output is the shortest Euclidean distance between them (in pixels), which is achieved by calling the OpenCV library; Finally, the ratio of the shortest Euclidean distance value to the preset effective radius of the laser spot of 10μm is recorded as the safe distance; wherein, the area with residual glue, the area with pure overflow, and the area with a mixture of overflow and residual glue are recorded as the areas to be cleaned.

[0049] Specifically, the strategy determination module responds to the condition that the area to be cleaned is an area with residual adhesive and the location risk category is a safe zone, using a first preset energy of 0.6 J / cm². 2 ; The cleaning process will terminate when the average grayscale value exceeds the preset grayscale threshold of 175. In response to the fact that the area to be cleaned is an area with residual adhesive and the location risk category is a high-risk area, the cleaning is terminated when the average gray value is greater than the preset gray value threshold using the second preset energy. In response to the fact that the area to be cleaned is a pure overflow area and the location risk category is a safe area, the cleaning is terminated when the average height is less than the preset height threshold of 2.0 μm using a third preset energy of 1.0 J / cm². In response to the fact that the area to be cleaned is a pure overflow area and the location risk category is a high-risk area, the cleaning is terminated when the average height is less than the preset height threshold using the fourth preset energy. In response to the fact that the area to be cleaned is a mixture of overflow and residual adhesive and the location risk category is a safe zone, the first preset energy is used for cleaning. After the average gray value is greater than the preset gray value threshold, the third preset energy is used for cleaning until the average height of the area to be cleaned is less than the preset height threshold. In response to the fact that the area to be cleaned is a mixture of overflow and residual adhesive and the location risk category is a high-risk area, the second preset energy is used to clean until the average gray value of the area to be cleaned is greater than the preset gray value threshold, and then the fourth preset energy is used to clean until the average height of the area to be cleaned is less than the preset height threshold, and then the process is terminated. Wherein, the grayscale mean is the arithmetic mean of the grayscale values ​​of all pixels in the area to be cleaned, the height mean is the arithmetic mean of the height values ​​of all pixels in the area to be cleaned, the first preset energy is less than the third preset energy, the second preset energy is determined according to the safety distance of the area to be cleaned, and the smaller the safety distance, the smaller the second preset energy; the fourth preset energy is determined according to the safety distance of the area to be cleaned, and the smaller the safety distance, the smaller the fourth preset energy; and, for the same safety distance, the second preset energy is less than the first preset energy, and the fourth preset energy is less than the third preset energy.

[0050] In this embodiment, the preset grayscale threshold is set to 175. The average grayscale value of the clean EMC substrate under 355nm coaxial illumination is distributed in the range of 180 to 195. The grayscale value of the area covered by residual adhesive drops to the range of 90 to 140 due to the absorption of ultraviolet light. When the average grayscale value rises back to above 175, it indicates that the residual adhesive has been basically removed and the substrate has been exposed. The preset height threshold is set to 2μm, based on the fact that the standard deviation of the height fluctuation of the active surface area of ​​the clean chip calibrated by a three-dimensional profilometer is about 0.5μm. Four times the standard deviation is taken as the allowable deviation for overflow removal to meet the surface flatness requirements of the subsequent redistribution layer.

[0051] In this embodiment, the residual adhesive is usually an extremely thin organic residue, and the energy density of the first preset energy of 0.6 J / cm² is sufficient to cause it to break photochemical bonds and vaporize without damaging the substrate; while the pure overflow is a thicker layer of cured epoxy resin, which requires a higher energy density of the third preset energy of 1.0 J / cm² to generate sufficient thermal stress to peel it off layer by layer.

[0052] Specifically, within the set of areas identified as high-risk zones, the specific values ​​of the safety distance reflect the differences in the degree of danger. The second and fourth preset energies are not fixed values, but rather continuous variables dynamically determined based on the actual safety distance of the area to be cleaned; the smaller the safety distance, the smaller the corresponding preset energy.

[0053] Specifically, this embodiment uses a linear decay model for dynamic calculation: The second preset energy E2 = E1 × (D / D0); the fourth preset energy E4 = E3 × (D / D0); where D is the actual safe distance of the current high-risk area, D0 is the preset safe distance threshold of 1.5, E1 is the first preset energy, and E3 is the third preset energy.

[0054] All the above strategies are executed at a preset scanning speed of 1000 mm / s. After each full-area scan, the image acquisition module simultaneously acquires the grayscale and height maps of the area to be cleaned, and the strategy determination module updates the average grayscale and height values ​​accordingly. For residual adhesive cleaning, the process terminates early when the average grayscale value is greater than 175; for overflow cleaning, the process terminates early when the average height value is less than 2 μm. If the criteria are not met, the next scan continues, with a maximum preset number of scans of 10. The process is forcibly terminated when the maximum number of scans is reached to prevent over-processing. The two-step cleaning of the mixture area is executed sequentially. After the first stage meets the grayscale termination condition, the process automatically switches to the second stage, with a maximum total of 20 scans for both stages.

[0055] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A laser-based device for removing excess material and cleaning plastic sealant, characterized in that, include: Image acquisition module, which is used to acquire grayscale images and height maps of the reconstructed wafer; A region calibration module is used to determine the active chip surface region of the reconstructed wafer based on the grayscale image and the height image; An edge blur determination module is used to determine the average edge blur of the active surface area of ​​the chip based on the height value of each pixel position within the active surface area of ​​the chip. An anomaly screening module is used to determine the anomaly screening category of the active surface area of ​​the chip based on the average edge ambiguity value. The anomaly screening category includes suspected normal area and suspected overflow anomaly area. A residual adhesive detection module is used to determine the dual-wavelength reflectance ratio of the suspected normal area; The residual adhesive determination module is used to determine whether the suspected normal area is an area with residual adhesive based on the dual-wavelength reflectance ratio. The feature recognition module is used to determine the gray uniformity index of the suspected overflow abnormal area based on the pixel gray values ​​of the suspected overflow abnormal area in the grayscale image. The overflow classification module is used to determine whether the suspected overflow abnormal area is a pure overflow area or a mixture of overflow and residual glue area based on the gray uniformity index. The location determination module is used to determine the location risk category of the area to be cleaned, which consists of the area with residual adhesive, the area with pure overflow, and the area with a mixture of overflow and residual adhesive. The location risk category includes safe areas and high-risk areas. The strategy determination module is used to determine the laser cleaning strategy for the area to be cleaned based on the category and location risk category of the area to be cleaned; A laser processing module for performing laser ablation on the surface of the reconstructed wafer according to the laser cleaning strategy.

2. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 1, characterized in that, The anomaly screening module determines that the active surface area of ​​the chip is a suspected overflow anomaly area when the mean edge ambiguity value of the active surface area of ​​the chip is less than a preset edge ambiguity threshold. The anomaly screening module determines that the active surface area of ​​the chip is a suspected normal area if the mean edge ambiguity value is greater than or equal to a preset edge ambiguity threshold.

3. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 2, characterized in that, The edge ambiguity determination module includes: A pixel blur calculation unit is used to calculate the edge blur of the pixel based on the height value of the pixel position in the height map, wherein the edge blur is the ratio of a first height difference to a second height difference. The first height difference is the absolute value of the height difference between adjacent pixel positions on both sides of the pixel in the height map along the image row coordinate direction, and the second height difference is the absolute value of the height difference between two pixel positions on both sides of the pixel in the height map along the image row coordinate direction, separated by one pixel. The grid mean calculation unit is used to divide the active surface area of ​​the chip into several grids and take the arithmetic mean of the edge ambiguity of all pixels in the grid, which is denoted as the edge ambiguity mean of the grid. The region mean calculation unit is used to take the arithmetic mean of the edge ambiguity of all grids in the active surface area of ​​the chip, and denoted as the edge ambiguity mean of the active surface area of ​​the chip.

4. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 3, characterized in that, The region calibration module is used to determine the active surface region of the chip on the reconstructed wafer, wherein, The region calibration module identifies connected components that simultaneously satisfy preset reflectivity, height, and pad shape conditions from the grayscale and height maps as pad imaging regions, and extracts the centroids of these connected components, marking them as reliable pad points. After spatial clustering of these reliable pad points, each cluster is graph-matched with the chip pad layout to determine the chip-internal coordinates of each reliable pad point. Based on the chip-internal coordinates and image coordinates of the successfully matched pad points, a spatial transformation relationship is calculated, and the chip active surface design boundary is mapped to global image coordinates through this spatial transformation relationship to obtain the chip active surface region. The preset reflectivity condition is that the average gray value of the connected domain is greater than the preset pad gray value threshold. The preset height condition is that the difference between the average height of the connected domain and the reference height of the pad is less than the preset height difference. The preset pad shape condition is that the ratio of the area of ​​the connected region to the area of ​​the preset pad is within a preset area ratio range, and the difference between the aspect ratio of the minimum bounding rectangle of the connected region and the aspect ratio of the preset pad is less than a preset ratio threshold.

5. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 4, characterized in that, The residual adhesive determination module determines that the suspected normal area is a normal area without residual adhesive when the ratio of the dual wavelength reflectance of the suspected normal area is less than the preset ratio of the dual wavelength reflectance. The residual adhesive determination module determines the suspected normal area as an area with residual adhesive in response to the dual-wavelength reflectivity ratio being greater than or equal to a preset dual-wavelength reflectivity ratio.

6. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 5, characterized in that, The residual adhesive detection module is used to determine the dual-wavelength reflectance ratio of the suspected normal area, wherein... The residual adhesive detection module obtains the first reflectance of each grid under the first preset wavelength laser irradiation and the second reflectance under the second preset wavelength laser irradiation, and records the ratio of the first reflectance to the second reflectance as the dual-wavelength reflectance ratio of the suspected normal area.

7. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 6, characterized in that, The overflow classification module determines that the suspected overflow abnormal area is a pure overflow area when the gray uniformity index of the suspected overflow abnormal area is less than a preset uniformity threshold. The overflow classification module determines that the suspected overflow abnormal area is a mixture of overflow and residual adhesive if the grayscale uniformity index of the suspected overflow abnormal area is greater than or equal to the preset uniformity threshold.

8. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 7, characterized in that, The feature recognition module is used to determine the grayscale uniformity index of the suspected overflow anomaly area, wherein... The feature recognition module obtains the pixel grayscale values ​​of the suspected overflow anomaly area in the grayscale image, calculates the ratio of the standard deviation to the mean of all pixel grayscale values ​​in the suspected overflow anomaly area, and determines the ratio as the grayscale uniformity index of the suspected overflow anomaly area.

9. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 8, characterized in that, The location determination module determines the area to be cleaned as a high-risk area when the safe distance to the area to be cleaned is less than a preset safe distance threshold; the location determination module determines the area to be cleaned as a safe area when the safe distance to the area to be cleaned is greater than or equal to the preset safe distance threshold. The safe distance of the area to be cleaned is the ratio of the shortest Euclidean distance between the pixel point on the edge contour of the area to be cleaned and all the pad imaging areas to the effective radius of the preset laser spot; wherein, the area with residual adhesive, the area with pure overflow, and the area with a mixture of overflow and residual adhesive are referred to as the area to be cleaned.

10. The laser-assisted descaling and cleaning equipment for molded bodies according to claim 9, characterized in that, The strategy determination module responds to the fact that the area to be cleaned is an area with residual adhesive and the location risk category is a safe area, and cleans with a first preset energy until the average gray value is greater than a preset gray value threshold and then terminates. In response to the fact that the area to be cleaned is an area with residual adhesive and the location risk category is a high-risk area, the cleaning is terminated when the average gray value is greater than the preset gray value threshold using the second preset energy. In response to the fact that the area to be cleaned is a pure overflow area and the location risk category is a safe area, the cleaning is terminated when the average height is less than a preset height threshold using a third preset energy. In response to the fact that the area to be cleaned is a pure overflow area and the location risk category is a high-risk area, the cleaning is terminated when the average height is less than the preset height threshold using the fourth preset energy. In response to the fact that the area to be cleaned is a mixture of overflow and residual adhesive and the location risk category is a safe zone, the first preset energy is used for cleaning. After the average gray value is greater than the preset gray value threshold, the third preset energy is used for cleaning until the average height of the area to be cleaned is less than the preset height threshold. In response to the fact that the area to be cleaned is a mixture of overflow and residual adhesive and the location risk category is a high-risk area, the second preset energy is used to clean until the average gray value of the area to be cleaned is greater than the preset gray value threshold, and then the fourth preset energy is used to clean until the average height of the area to be cleaned is less than the preset height threshold, and then the process is terminated. Wherein, the grayscale mean is the arithmetic mean of the grayscale values ​​of all pixels in the area to be cleaned, the height mean is the arithmetic mean of the height values ​​of all pixels in the area to be cleaned, the first preset energy is less than the third preset energy, the second preset energy is determined according to the safety distance of the area to be cleaned, and the smaller the safety distance, the smaller the second preset energy; the fourth preset energy is determined according to the safety distance of the area to be cleaned, and the smaller the safety distance, the smaller the fourth preset energy; and, for the same safety distance, the second preset energy is less than the first preset energy, and the fourth preset energy is less than the third preset energy.

Images