Processing method for solving problems of residual glue on copper surface at position of Cavity product and small size of bound PAD

Abstract

The invention relates to the technical field of HDI printed circuit board processing, in particular to a processing method for solving the problems of residual glue on a copper surface at the position of a Cavity product and small size of a bound PAD, after laser Cavity processing, feature vectors are input into a trained gradient boosting regression tree (GBRT) model, adaptive Plasma parameters are output, and the adaptive Plasma parameters are adopted to carry out Plasma processing on the Cavity copper surface; according to the method, the Plasma parameters are output in a self-adaptive mode through the GBRT model, and dynamic matching of the residual glue characteristics and the processing parameters is achieved in combination with the sand blasting path planned through the A * algorithm; wherein the prediction error of the GBRT model on the Plasma parameter is less than or equal to 5%, the preferential coverage of a to-be-optimized area is ensured, the path overlapping rate is less than or equal to 5%, the final residual glue removal rate is improved to 98% or above, and the problem that the traditional fixed parameter processing is not thorough is solved.

Description

TECHNICAL FIELD

[0001] The application relates to the HDI printed circuit board processing technical field, in particular to a processing method for solving the copper surface residual glue and small size of a bound PAD of a Cavity product. BACKGROUND

[0002] Under the environment of continuous iteration of technology and deep reform of industry, the Cavity product, as a key component in multiple fields, is becoming increasingly important. Whether it is the optimization of chip performance in semiconductor manufacturing, the provision of support for accurate diagnosis in medical equipment, or the assistance in efficient signal transmission in communication systems, the Cavity product plays an indispensable role.

[0003] With the continuous advancement of Moore's Law, chip manufacturing processes continue to move towards smaller sizes. In this process, Cavity products are widely used in the semiconductor packaging link. To meet the heat dissipation needs generated by high-speed operation of chips, Cavity structures with good thermal conductivity are designed for chip packaging. By building Cavity in the packaging material to accommodate heat sinks, the heat generated by the chip can be quickly dissipated, ensuring stable operation of the chip. At the same time, in advanced 3D packaging technology, Cavity is used to realize vertical interconnection between chips, such as through-silicon via (TSV) technology, which manufactures Cavity on a silicon wafer and fills it with metal to achieve electrical connection between different chip layers, greatly improving data transmission rate and reducing packaging size, promoting the development of semiconductor devices towards miniaturization and high performance.

[0004] However, the existing processing technology has the following key problems: traditional Plasma processing uses fixed parameters (such as carbon tetrafluoride flow, radio frequency power), which cannot adapt to the differences in residual glue type, thickness and distribution density, resulting in local residual glue residue often exceeding 3 μm, affecting the subsequent plating layer adhesion. Sandblasting is a supplementary means, but due to the lack of precise path planning, it is easy to cause the contradiction between insufficient processing of the area to be optimized and excessive damage to the non-target area. The existing process has the following problems: when the dry film is pressed for the second time, the Cavity area is prone to dry film floating due to complex terrain, resulting in pattern transfer deviation; and residual glue residue will directly cause the actual size of the PAD after etching to be small, which does not meet the chip bonding requirements. In the traditional process, residual glue detection, parameter adjustment and path planning are independent of each other, and there is a lack of data-driven adaptive mechanism. SUMMARY

[0005] The application aims to provide a processing method for solving the copper surface residual glue and small size of the binding PAD of the Cavity product position, and solve the above technical defects; in order to solve the above technical problems, the carbon powder residue and the resin residue which are not burned clean by the laser during the laser processing of the Cavity product can be completely removed by adopting the Plasma and sand blasting processes after the Cavity product is processed by the laser; and due to the uneven characteristics of the Cavity position plate surface, a vacuum film pressing machine process needs to be used for film pressing, so that the dry film of the binding PAD at the edge of the Cavity position can not be separated, and after the development and etching of the following chemical solution, the size of the binding PAD will not be attacked, and the problem of small size of the binding PAD will not occur; after the plate is etched, the AOI scanning is confirmed again; and the processing quality of the Cavity product can be further ensured.

[0006] In order to achieve the above effects, the technical scheme adopted by the application is as follows: a processing method for solving the copper surface residual glue and small size of the binding PAD of the Cavity product position, comprising the following core steps: Step 1: after the laser Cavity processing, the residual glue characteristics of the Cavity position copper surface are collected, the average thickness, maximum thickness, type and distribution density of the residual glue are obtained, the feature vector is input into the trained gradient boosting regression tree (GBRT) model, the adaptive Plasma parameters are output, and the Cavity copper surface is processed by the adaptive Plasma parameters; Step 2: after the Plasma processing, the residual glue of the Cavity copper surface is rechecked by the laser thickness gauge, the areas to be optimized with the residual glue thickness greater than 3 microns are screened, the spray pressure, sand-water ratio and nozzle moving speed are calculated based on the average residual glue thickness of the areas to be optimized, the nozzle path is planned combined with the A* algorithm, and the mixture of corundum and water is used for sand blasting treatment; Step 3: after the sand blasting treatment, the vacuum film pressing machine is used for secondary dry film pressing to ensure that the dry film of the binding PAD at the edge of the Cavity has no floating, and the outer layer is sequentially exposed, etched and removed; Step 4: after the film is removed, the size of the binding PAD around the Cavity is detected by the secondary AOI, and the chemical nickel-palladium-gold, IR drying, plate warping, electric milling, electric measurement and OQC inspection are finally completed.

[0007] Preferably, the Plasma processing in step 1 adopts carbon tetrafluoride gas.

[0008] Preferably, the residual glue feature acquisition in step 1 adopts a high-resolution line array camera with a resolution of 2048x1080 pixels, and combines with an on-axis light source to collect images at a rate of 10 frames per second; wherein the residual glue thickness is calculated by the formula T=0.2x(G-Gb), G is the gray value of the residual glue area in the collected image, Gb is the reference gray value of the copper surface without residual glue, and the residual glue type is distinguished by the texture entropy value E: when E is less than 5, it is determined that carbon powder is left, and at this time the residual glue type C=0 is recorded; when E is greater than 8, it is determined that resin is left, and at this time the residual glue type C=1 is recorded.

[0009] Preferably, the training process of the GBRT model in step 1 includes: collecting 1000 groups of historical process data of feature input vectors containing residual glue and optimal Plasma output parameters, and dividing them into training set and validation set according to 8:2; setting the number of regression trees to 300, the learning rate to 0.1, the maximum depth of a single tree to 5, and the subsampling ratio to 0.8; the optimal Plasma parameters include carbon tetrafluoride gas flow 20-50sccm, radio frequency power 100-300W, and processing time 30-120s.

[0010] Preferably, in the mixture of corundum and water in step 2, the sand-water ratio is calculated as R=0.2xHavg+1 when the average thickness of residual glue Havg is less than 5μm, and is fixed as 1:2 when Havg is greater than or equal to 5μm.

[0011] Preferably, the calculation method of the sandblasting parameters in step 2 is: the spraying pressure L=0.1xHavg+0.2; the nozzle moving speed V is calculated as V=-0.5xHavg+5 when Havg is less than 8μm, and is fixed as 1mm / s when Havg is greater than or equal to 8μm, wherein Havg is the arithmetic mean of the residual glue thickness in the region to be optimized.

[0012] Preferably, the application of A* algorithm in step 2 includes: discretizing the Cavity copper surface into a two-dimensional grid map with a size of 1mmx1mm, defining the evaluation function , as the actual sandblasting cost from the starting point to the current node n, the region to be optimized, =1, the non-optimized region, =5; as the Manhattan distance from the current node n to the nearest region to be optimized, the region to be optimized is preferentially covered, and the non-optimized region is treated with a spraying pressure of 0.2MPa and a nozzle speed of 5mm / s.

[0013] Preferably, in step 3, the vacuum degree of the vacuum laminator ranges from -0.095 to -0.085MPa, and the dry film floating rate after lamination is ≤0.1%.

[0014] Preferably, the secondary AOI detection in step 4 uses a 5 million pixel industrial camera to detect the size deviation of the bound PAD ≤ ± 0.005 mm.

[0015] Preferably, it further includes secondary Plasma processing after secondary AOI detection, and then sequentially performing the steps of chemical nickel-palladium-gold, IR infrared drying, plate warping and baking, electrical milling profile processing and electrical measurement; the IR infrared drying temperature is 110-130℃; the pressure of plate warping and baking is 0.2-0.3MPa.

[0016] Compared with the prior art, the beneficial effects of the present application are: 1. By using the GBRT model to adaptively output the Plasma parameters, combined with the sandblasting path planned by the A* algorithm, the dynamic matching of the residual glue characteristics and the processing parameters is realized; wherein the prediction error of the GBRT model for the Plasma parameters is ≤ 5%, which ensures that the to-be-optimized area is preferentially covered, the path overlap rate is ≤ 5%, and the final residual glue removal rate is improved to more than 98%, solving the problem of incomplete processing of traditional fixed parameters.

[0017] 2. The secondary dry film non-floating pressing is realized by the vacuum film pressing machine, and the accurate monitoring of the PAD size is realized by the secondary AOI detection, so that the size deviation of the bound PAD is controlled within ± 0.005 mm, meeting the stringent requirements of chip bonding.

[0018] 3. The sandblasting efficiency is improved by 30% through the path planning of the A* algorithm, and the copper surface scratching rate is reduced to below 0.5% by using a low-pressure high-speed processing strategy in the non-to-be-optimized area; the residual glue feature data is linked in each process, and the rework rate is reduced to below 3%; the improvement of the completeness of residual glue removal and the PAD size precision improves the chemical nickel-palladium-gold plating layer adhesion by 20%; the plate warping and baking process further ensures the flatness of the substrate, meeting the high and low temperature cycle reliability requirements from -40℃ to 125℃. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only represent some embodiments of the present application, and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0020] Figure 1 is a flowchart of a processing method for solving the position copper surface residual glue and small size bonding PAD of the Cavity product according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] The present application will be further explained in conjunction with the drawings and specific embodiments. Embodiment 1

[0022] After the cavity product is processed by laser, there will be carbon powder residues and resin residues that are not burned clean by laser during laser processing on the bottom copper surface of the cavity position. By using the plasma and sand blasting processes, the carbon powder residues and resin residues that are not burned clean by laser during laser processing of the cavity can be completely removed. Moreover, due to the uneven characteristics of the cavity position plate surface, a vacuum film pressing machine process needs to be used for film pressing, so that the edge binding PAD dry film of the cavity position can not float away, and after the subsequent development and etching of the chemical solution, the binding PAD size will not be attacked, and the problem of small binding PAD size will not occur. After the plate is etched, an AOI scanning confirmation is added again, which can further ensure the quality requirements of the processed cavity product.

[0023] Please refer to Figure 1 As shown in the drawings, the embodiment discloses a processing method for solving the problems of residue glue on the copper surface of the cavity product position and small binding PAD size, which comprises the following processing procedures: Outer layer lamination: laminate multi-layer core board, prepreg, etc. into one, forming a multi-layer substrate structure; → Laser brown: process copper surface, increase roughness and oxidation resistance, improve interlayer adhesion; → Laser: use laser for fine processing, prepare for subsequent processes; → Drill through-hole: drill holes through the substrate for interlayer electrical connection; → Deslagging: remove the sludge generated by drilling to ensure clean hole wall; → Blind hole AOI: automatically detect the size, position, etc. of blind holes, screen out defective blind holes; → PTH: deposit copper on the hole wall to make the hole wall conductive and realize interlayer connection; → Hole filling and plating: fill and thicken the hole by electroplating to ensure full filling and meet the copper layer thickness standard; → Outer layer pretreatment: clean and roughen the outer layer copper surface to improve the adhesion with dry film / ink; → Outer layer exposure: transfer the outer layer circuit pattern to the dry film to protect the copper surface that needs to be retained; → Outer layer etching: etch the copper layer not protected by the dry film to form the outer layer circuit; → Impedance measurement: measure the characteristic impedance of the circuit to ensure it meets the design requirements; → Outer layer AOI: automatically optically detect the outer layer circuit to check for defects such as open circuit, short circuit, and abnormal line width; → Anti-solder plug hole: fill holes that do not need to be soldered to prevent solder paste from entering and protect the inner circuit; → Anti-solder printing: print anti-solder ink to protect the copper surface in non-soldering areas from oxidation and short circuit; → Anti-solder exposure: transfer the anti-solder pattern to determine the area where the anti-solder layer needs to be retained; → Finished product impedance: measure the finished product impedance again to ensure that the final performance meets the specifications; → Text printing: print text such as model number and production information for easy product identification; → Laser Cavity: use laser to process Cavity structure to accommodate chips and other devices; → Plasma treatment: plasma treat the Cavity copper surface to remove residue and improve cleanliness; → Sandblasting processing: further clean the Cavity area of impurities to optimize the copper surface topography; → Second dry film: laminate the dry film again to prepare for subsequent pattern transfer for PAD bonding; → Second outer layer exposure: perform second exposure on fine structures such as bonded PAD to accurately transfer the pattern; → Second outer layer acid etching: acid etch the copper layer not protected by the dry film to form fine structures such as bonded PAD; → Film removal: remove the dry film to expose the copper surface for subsequent processing; → Second outer layer AOI: detect the size, topography, etc. of the bonded PAD after secondary processing to screen out defective products; → Second plasma treatment: plasma treat again to optimize copper surface performance and prepare for chemical nickel-palladium-gold; → Chemical nickel-palladium-gold: deposit nickel, palladium, and gold plating layer on the copper surface to improve solderability, oxidation resistance, and reliability; → IR heat treatment: infrared heating to dry and cure the plating layer or coating to ensure stable performance; → Plate warping and baking: laminate and bake to correct plate warping to ensure substrate flatness; → Electrical milling: accurately process the substrate shape to meet the final size and shape requirements; → Electrical measurement: electrically test the conductivity and insulation of the circuit to ensure no electrical defects; → Visual inspection: manual visual inspection to supplement automatic detection and screen out products with appearance or minor defects; → OQC: final shipment quality inspection to ensure that the products meet the quality standards;→ Packaging: The qualified products are packaged for storage, transportation and delivery.

[0024] It should be noted that by adopting the Plasma and sand blasting processes, the residual carbon powder and the residual resin that are not completely removed during laser processing of the Cavity can be completely removed, and the reliability of the product can be ensured. Among them: Plasma: The carbon tetrafluoride gas of the Plasma equipment can effectively remove the residual glue of the large copper skin at the bottom of the Cavity position, ensuring the reliability of the problem. It is necessary to ensure that the uniformity of Plasma glue removal reaches more than 80%. Sand blasting: The mixture of corundum and water is sprayed onto the board surface, which can further effectively remove the residual glue of the large copper skin at the bottom of the Cavity position and the carbon powder during laser PP burning.

[0025] Among them, the processing process of Plasma processing is: residual glue feature collection→Plasma parameter adaptive algorithm→Plasma processing; Specifically, the residual glue feature collection process is as follows: a high-resolution line array camera is combined with an on-axis light source to collect images of the copper surface after laser Cavity, and the gray value, texture feature and coordinate information of the residual glue area are output synchronously; among them, the high-resolution line array camera adopts a resolution of 2048*1080 pixels, and the image acquisition rate is 10 frames / second; the gray value of the residual glue area reflects the residual glue thickness: every 10 increase in gray value corresponds to 2μm increase in residual glue thickness; the texture feature reflects the residual glue type: the texture entropy value of the carbon powder residue is <5, and the texture entropy value of the resin residue is >8.

[0026] At the same time, the collected residual glue features are preprocessed, the collected images are subjected to Gaussian filtering to remove noise, the residual glue area is extracted through Otsu threshold segmentation, and the edge profile is optimized through morphological operation, and finally the standardized feature vector is output; among them, the standardized feature vector includes: residual glue average thickness Tavg, residual glue maximum thickness Tmax, residual glue type C and residual glue distribution density D.

[0027] Further, the residual glue area thickness is calculated according to the formula T=0.2* (G-Gb), wherein G is the gray value of the residual glue area in the collected image, Gb is the reference gray value of the copper surface without residual glue, which is calibrated by pre-collecting standard samples without residual glue, and 0.2 is the gray-thickness conversion coefficient based on experimental data fitting, every 10 gray values correspond to 2μm thickness, the arithmetic average value of all residual glue area thicknesses is calculated to obtain the residual glue average thickness Tavg, which ranges from 5-20μm; at the same time, the maximum thickness value in the residual glue area is obtained and recorded as the residual glue maximum thickness Tmax; the ratio of the residual glue area to the total area of the Cavity copper surface is recorded as the residual glue distribution density D.

[0028] Further, the residue type C is 1 and 0 respectively, wherein C=0 is carbon powder and C=1 is resin; a texture entropy value E is calculated by a gray level co-occurrence matrix to reflect the complexity of the residue surface texture, when E is less than 5, it is determined that carbon powder is left, at this time, the residue type C=0 is recorded, the carbon powder left texture is regular and the entropy value is low; when E is greater than 8, it is determined that resin is left, at this time, the residue type C=1 is recorded, the resin left texture is messy and the entropy value is high.

[0029] Specifically, the application process of the Plasma parameter adaptive algorithm is as follows: a trained gradient boosting regression tree (GBRT) model is used, the input vector is: [Tavg, Tmax, C, D], and the output parameter is: carbon tetrafluoride gas flow F, radio frequency power P and processing time t; The GBRT model dynamically adjusts the parameters according to the following rules: when the value of Tavg rises, the values of F, P and t are all raised to enhance the etching intensity; when C=1, F increases by 10-15sccm compared with C=0; when the value of D rises, t increases by 20-30s; through the dynamically adjusted output parameters, the Plasma device controller is sent through the communication interface to realize the adaptive adjustment of the Plasma parameters; the next step of Plasma processing is carried out according to the adaptive adjustment of the Plasma parameters.

[0030] Further, the training method of the GBRT model is as follows: a1: 1000 groups of historical process data are collected, each group of data contains input features and output parameters, wherein the input features include residue average thickness Tavg, residue maximum thickness Tmax, residue type C and residue distribution density D; the output parameters are the optimal Plasma process parameters verified by experiments, including carbon tetrafluoride gas flow F, radio frequency power P and processing time t, and these optimal Plasma process parameters need to meet the process requirements: Plasma degreasing uniformity ≥95%, residue left amount ≤3μm, copper surface damage depth ≤0.5μm.

[0031] a2: the key hyperparameter configuration of the GBRT model is set: the number of regression trees is 300, which is used to balance the training accuracy and efficiency; the learning rate is 0.1, which is used to control the contribution weight of each tree to the final prediction; the maximum depth of a single tree is 5, which is used to limit the complexity of a single tree to avoid overfitting; the subsampling ratio is 0.8, which is used to enhance the generalization ability by randomly selecting 80% of the samples during the training of each tree.

[0032] a3: 1000 groups of data are divided into 800 groups for model learning training set and 200 groups for monitoring training process validation set according to the ratio of 8:2; a step-by-step fitting residual strategy is used for training: the initial prediction value is the mean value of the training set of output parameters; a4: For each regression tree, first calculate the residual error between the current model prediction value and the true value, that is, the difference between the true value and the current prediction value; use the input features of the training set and the residual error to train a regression tree to learn how to fit the residual error, multiply the prediction result of the tree by the learning rate, and add it to the prediction result of the current model to update the model; a5: Repeat step a4 until 300 trees are trained, or the prediction error on the validation set no longer decreases; in addition, to ensure the generalization ability of the model, 1000 groups of data are randomly divided into 5 equal-sized subsets, and each time 4 subsets are used for training and 1 subset is used for validation, repeated 5 times; through cross-validation, adjust the hyperparameters, and finally make the prediction error ≤5%, that is, the deviation between the predicted value and the true optimal value ≤5%.

[0033] a6: Deploy the trained GBRT model to the industrial computer of the algorithm control module. When the residual glue feature vector [Tavg, Tmax, C, D] is collected in real time, the model can quickly load and output the corresponding optimal Plasma parameters F, P, t, which are sent to the Plasma device controller through the communication interface to realize adaptive adjustment of parameters.

[0034] It should be noted that the algorithm model breaks the limitations of traditional Plasma fixed parameters: on the one hand, it accurately captures dynamic characteristics such as residual glue thickness, type, and distribution, allowing parameters to be optimized in real time according to product residual glue differences; on the other hand, relying on the precise prediction and closed-loop control of the GBRT model, Plasma glue removal uniformity is significantly improved, with residual glue residue controlled at ≤5 μm and copper surface damage depth limited to ≤0.5 μm, ensuring Cavity copper surface cleanliness while avoiding excessive etching of the copper surface, laying a high-quality foundation for subsequent processes and effectively solving the problem of product defects caused by uneven residual glue removal, thereby improving the processing yield and quality stability of Cavity products.

[0035] Among them, the processing process of sandblasting processing is: residual glue re-inspection before sandblasting → dynamic optimization algorithm of sandblasting parameters → sandblasting treatment; Specifically, the method of residual glue re-inspection before sandblasting is: using a laser thickness gauge to measure the copper surface after Plasma treatment, with a measurement point density of 1000 points / mm2 2 One measurement point, a two-dimensional distribution of residual glue thickness is obtained , m is the number of measurement points, n is the number of measurement points, and each element in the matrix represents the residual glue thickness at coordinate . Then, the area that satisfies > 3 μm is selected as the area to be optimized.

[0036] Specifically, the analysis process of the sand blasting parameter dynamic optimization algorithm is: for the average residual glue thickness Havg of the to-be-optimized area, a dynamic calculation model of the spray pressure L, the sand-water ratio R, and the nozzle moving speed V is established: The calculation formula of the spray pressure L is: L=0.1*Havg+0.2, the thicker the residual glue, the greater the pressure required to enhance the grinding capacity of sand blasting.

[0037] The calculation formula of the sand-water ratio R is: When the residual glue thickness Havg is less than 5 microns, the sand ratio is increased in a linear relationship; when the residual glue thickness Havg is greater than or equal to 5 microns, the sand-water ratio is fixed at 1:2 to avoid excessive sand particles scratching the copper surface.

[0038] The calculation formula of the nozzle moving speed V is The thicker the residual glue, the lower the nozzle moving speed needs to be to extend the grinding time; when the residual glue thickness Havg is greater than or equal to 8 microns, the speed is fixed at 1 mm / s to ensure that thick residual glue is fully removed.

[0039] Wherein, the average residual glue thickness Havg of the to-be-optimized area is the arithmetic mean of all in the to-be-optimized area.

[0040] Further, the A* algorithm is used to plan the motion path of the nozzle, and the core logic is: to preferentially cover the to-be-optimized area, and the path overlap rate is less than or equal to 5%; for non-to-be-optimized areas, low pressure and high speed are used to process, which can ensure that there is no residual glue on the copper surface while avoiding scratching the copper surface.

[0041] Specifically, the specific application steps of the A* algorithm in the sand blasting nozzle path planning are: b1: discretize the to-be-processed area of the Cavity copper surface into a two-dimensional grid map, and each grid is used as a node in the algorithm; the grid size can be set according to the process precision requirement to ensure that the residual glue area can be accurately covered and the calculation complexity can be controlled. Each node contains coordinate information and a residual glue thickness label of the position, the residual glue thickness label includes a to-be-optimized area H>3 microns or a non-to-be-optimized area H≤3 microns.

[0042] b2: define the evaluation function , wherein is the actual sand blasting cost from the starting point to the current node n; the cost is defined as the "comprehensive measure of the potential damage risk of sand blasting to the copper surface and the process efficiency": if node n belongs to the to-be-optimized area, set to 1; if node n belongs to the non-to-be-optimized area, set to 5; The Manhattan distance from the current node n to the nearest node of the region to be optimized is calculated by a heuristic function, which satisfies that the estimated distance is not greater than the actual distance, ensuring that the A* algorithm can find the optimal path.

[0043] b3: Open list: stores nodes to be checked, sorted by value from small to large; each time the smallest node is selected from the open list for expansion, ensuring that the most promising path is searched first; closed list: stores nodes that have been checked to avoid repeated processing of the same node and improve algorithm efficiency.

[0044] b4: path search process: add the initial position node of the nozzle to the open list, calculate its , take the smallest node n from the open list, add it to the closed list, and check whether node n belongs to the region to be optimized. If it belongs to the region to be optimized, mark the node as having been covered first, then generate its 8-neighborhood child nodes. For each child node: if the child node is already in the closed list, skip it; if the child node is not in the open list, calculate its , = the Manhattan distance from the child node to the nearest region to be optimized, and then get , add the child node to the open list. If it belongs to the non-optimized region, generate 8-neighborhood child nodes as well, but when calculating

[0045] , if the child node still belongs to the non-optimized region, then ; if the child node enters the optimized region, then .

[0046] b5: the termination condition is that when the open list is empty, or the coverage times of all nodes of the region to be optimized satisfy that the path overlap rate ≤ 5%, i.e., the proportion of the number of times that the same node of the region to be optimized is covered by the path to the total search times ≤ 5%, to avoid local over-sanding, the search is stopped, and the final nozzle path is obtained.

[0047] When the nozzle passes through the region to be optimized along the planned path, the dynamic optimization algorithm of sanding parameters is triggered, and the spray pressure L, sand-water ratio R, and nozzle moving speed V are output in real time according to the average thickness of the residual glue Havg in the region; when the nozzle passes through the non-optimized region, it automatically switches to fixed parameters, the spray pressure L = 0.2 MPa, the nozzle moving speed V = 5 mm / s, and the sand-water ratio remains at a conventional low ratio, to ensure that the copper surface is free of residual glue while avoiding scratches on the copper surface.

[0048] ​​It should be noted that through the heuristic search of the A* algorithm and the design of the cost function, the sand blasting path can preferentially and sufficiently cover the serious residual glue area to be optimized, and quickly and gently process the non-optimized area with slight residual glue, so as to finally realize the process target of a residual glue removal rate of 98% and a copper surface scratch rate of 0.5%, thereby providing a guarantee for the subsequent binding PAD size accuracy.

[0049] The above formulas are dimensionless values, the formulas are obtained by collecting a large amount of data to simulate the most recent real situation, the size of the coefficient is a specific value obtained by quantifying each parameter, and the size of the coefficient can be as long as it does not affect the proportional relationship between the parameter and the quantized value. Meanwhile, the contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0050] The application is not limited to the above optional embodiments, and anyone can derive other various forms of products under the inspiration of the application. The above specific embodiments should not be understood as limiting the protection scope of the application, and the protection scope of the application should be defined by the claims, and the specification can be used to explain the claims.

Claims

1. A processing method for resolving residual adhesive on the copper surface and undersized bonding pads in Cavity products, characterized in that, Includes the following steps: Step 1: After laser Cavity processing, residual adhesive features are collected on the copper surface at the Cavity location to obtain feature vectors of average thickness, maximum thickness, type and distribution density of residual adhesive. The feature vectors are input into the trained gradient boosting regression tree model to output adaptive Plasma parameters. The adaptive Plasma parameters are then used to perform Plasma processing on the copper surface of the Cavity. Step 2: After Plasma treatment, the Cavity copper surface is re-inspected for residual adhesive using a laser thickness gauge. Areas with residual adhesive thickness greater than 3μm are selected for optimization. Based on the average residual adhesive thickness of the areas to be optimized, the spraying pressure, sand-to-water ratio, and nozzle movement speed are calculated. The nozzle path is planned using the A* algorithm, and a mixture of diamond abrasive and water is used for sandblasting. Step 3: After sandblasting, a vacuum laminator is used for secondary dry film lamination, followed by secondary exposure of the outer layer, secondary acid etching of the outer layer, and film stripping. Step 4: After film removal, the dimensions of the PADs around the Cavity are confirmed by secondary AOI inspection. Finally, the electroless nickel palladium gold process, IR drying, plate warping and baking, electric milling, electrical testing and OQC inspection are completed.

2. The processing method for solving the problems of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The Plasma treatment in step 1 uses carbon tetrafluoride gas.

3. The processing method for solving the problems of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The residual adhesive feature acquisition in step 1 uses a high-resolution linear scan camera with a resolution of 2048×1080 pixels, combined with a coaxial light source, to acquire images at a rate of 10 frames / second. The residual adhesive thickness is calculated using the formula T=0.2×(G-Gb), where G is the gray value of the residual adhesive area in the acquired image, and Gb is the reference gray value of the copper surface without residual adhesive. The type of residual adhesive is distinguished by the texture entropy value E: when E<5, it is determined to be toner residue, and the residual adhesive type is recorded as C=0; when E>8, it is determined to be resin residue, and the residual adhesive type is recorded as C=1.

4. The processing method for solving the problem of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The training process of the gradient boosting regression tree model in step 1 includes: collecting 1000 sets of historical process data containing residual adhesive feature input vectors and optimal Plasma output parameters, and dividing them into training and validation sets in an 8:2 ratio; setting the number of regression trees to 300, the learning rate to 0.1, the maximum depth of a single tree to 5, and the subsampling ratio to 0.8; the optimal Plasma parameters include a carbon tetrafluoride gas flow rate of 20-50 sccm, an RF power of 100-300W, and a processing time of 30-120s.

5. The processing method for solving the problem of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, In the mixture of corundum and water described in step 2, the ratio of corundum to water is calculated as R=0.2×Havg+1 when the average thickness of residual adhesive Havg is less than 5μm, and is fixed at 1:2 when Havg is greater than or equal to 5μm.

6. The processing method for solving the problem of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The sandblasting parameters mentioned in step 2 are calculated as follows: spray pressure L = 0.1 × Havg + 0.2; nozzle moving speed V is calculated as V = -0.5 × Havg + 5 when Havg < 8 μm, and fixed speed is 1 mm / s when Havg ≥ 8 μm, where Havg is the arithmetic mean of the residual adhesive thickness in the area to be optimized.

7. The processing method for solving the problems of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The application of the A* algorithm in step 2 includes: discretizing the Cavity copper surface into a 1mm × 1mm two-dimensional grid map, and defining the evaluation function. , The actual sandblasting cost from the starting point to the current node n is the region to be optimized. =1, not a region to be optimized. =5; The Manhattan distance from the current node n to the nearest node in the region to be optimized is used. The region to be optimized is covered first, while the non-optimized regions are treated with a spray pressure of 0.2 MPa and a nozzle speed of 5 mm / s.

8. The processing method for solving the problem of residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The vacuum pressure of the vacuum laminator in step 3 is in the range of -0.095 to -0.085 MPa, and the dry film float rate after lamination is ≤0.1%.

9. A processing method for resolving residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, The secondary AOI inspection in step 4 uses a 5-megapixel industrial camera, and the size deviation of the detected PAD is ≤ ±0.005mm.

10. A processing method for resolving residual adhesive on the copper surface and undersized bonding pads in Cavity products according to claim 1, characterized in that, It also includes a second Plasma treatment after the second AOI inspection, followed by chemical nickel-palladium-gold plating, IR infrared drying, plate warping and pressing, electric milling of the shape and electrical testing. The IR infrared drying temperature is 110-130℃; the plate warping and pressing pressure is 0.2-0.3MPa.

Images