Processing method for solving cavity product position copper surface residual glue and small size binding pad

Abstract

The present application relates to the technical field of HDI printed circuit board processing, in particular to a processing method for solving the problems of copper surface residual glue and small size of binding PAD in cavity product position, after laser cavity processing, the characteristic vector is input into the trained gradient boosting regression tree (GBRT) model, and the adaptive plasma parameter is output, and the cavity copper surface is treated by adopting the adaptive plasma parameter; the present application realizes the dynamic matching of residual glue characteristics and treatment parameters by adopting the GBRT model to adaptively output the plasma parameter and combining with the sand blasting path planned by the A* algorithm; wherein the prediction error of the GBRT model to the plasma parameter is less than or equal to 5%, the priority coverage of the region to be optimized is ensured, the path overlap rate is less than or equal to 5%, the final residual glue removal rate is improved to more than 98%, and the problem of incomplete treatment of traditional fixed parameters is solved.

Description

Technical Field

[0001] This invention relates to the field of HDI printed circuit board processing technology, specifically a processing method to solve the problems of residual adhesive on the copper surface of Cavity products and the small size of the bonding pads. Background Technology

[0002] In an environment of continuous technological iteration and profound industrial transformation, Cavity products are becoming increasingly important as a key component in multiple fields. Whether it's optimizing chip performance in semiconductor manufacturing, supporting accurate diagnosis in medical devices, or facilitating efficient signal transmission in communication systems, Cavity products play an indispensable role.

[0003] As Moore's Law continues to advance, chip manufacturing processes are constantly shrinking in size. In this process, cavity products are widely used in semiconductor packaging. To meet the heat dissipation demands of high-speed chip operation, cavity structures with excellent thermal conductivity are designed for chip packaging. By constructing cavities within the packaging material to house heat sinks, the heat generated by the chip can be quickly dissipated, ensuring stable chip operation. Simultaneously, in advanced 3D packaging technologies, cavities are used to achieve vertical interconnects between chips. For example, through-silicon vias (TSVs) technology fabricates cavities on silicon wafers and fills them with metal, thereby achieving electrical connections between different chip layers. This significantly improves data transmission rates, reduces package size, and drives the miniaturization and high performance of semiconductor devices.

[0004] However, existing processing techniques suffer from the following key problems: Traditional plasma processing uses fixed parameters (such as carbon tetrafluoride flow rate and RF power), which cannot adapt to differences in residual adhesive type, thickness, and distribution density. This often results in localized residual adhesive exceeding 3μm, affecting the adhesion of subsequent coatings. While sandblasting is a supplementary method, the lack of precise path planning easily leads to a contradiction between insufficient treatment of the target area and excessive damage to non-target areas. During the secondary dry film lamination process, the complex terrain of the cavity area makes it prone to dry film detachment, causing pattern transfer deviations; and residual adhesive directly results in the actual size of the etched PAD being smaller than required for chip bonding. In traditional processes, residual adhesive detection, parameter adjustment, and path planning are independent of each other, lacking a data-driven adaptive mechanism. Summary of the Invention

[0005] The purpose of this invention is to provide a processing method to address the issues of residual adhesive on the copper surface and undersized bonding pads in the cavity area of ​​a product, thus resolving the aforementioned technical defects. To address these problems, after laser processing, the bottom copper surface of the cavity area often retains carbon powder residue and resin residue that was not completely burned off by the laser. By employing a combination of plasma and sandblasting processes, this method effectively removes these residues. Furthermore, due to the uneven surface of the cavity area, a vacuum lamination process is used to ensure that the dry film on the bonding pads at the cavity edge does not detach. This prevents damage to the bonding pad dimensions after subsequent development and etching processes, thus avoiding the problem of undersized bonding pads. An additional AOI scan is performed after etching to further ensure the production quality requirements of the cavity product processing.

[0006] To achieve the above effects, the technical solution adopted by this invention is as follows: a processing method for solving the problems of residual adhesive on the copper surface and undersized bonding pads in Cavity products, comprising the following core steps:

[0007] 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 (GBRT) model to output adaptive Plasma parameters. The adaptive Plasma parameters are then used to perform Plasma processing on the copper surface of the Cavity.

[0008] Step 2: After Plasma processing, the Cavity copper surface is re-inspected for residual adhesive using a laser thickness gauge. Areas with residual adhesive thickness > 3 μm 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.

[0009] Step 3: After sandblasting, a vacuum laminator is used for secondary dry film lamination to ensure that the dry film of the PAD is bonded to the edge of the Cavity without floating off. Then, the outer layer is exposed twice, the outer layer is acid-etched twice, and the film is removed.

[0010] 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.

[0011] Preferably, the Plasma treatment in step 1 uses carbon tetrafluoride gas.

[0012] Preferably, in step 1, the residual adhesive feature acquisition 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 grayscale value of the residual adhesive area in the acquired image, and Gb is the reference grayscale value of the copper surface without residual adhesive. The type of residual adhesive is distinguished by the texture entropy value E: when E is less than 5, it is determined to be carbon powder residue, and the residual adhesive type C=0 is recorded; when E is greater than 8, it is determined to be resin residue, and the residual adhesive type C=1 is recorded.

[0013] Preferably, the training process of the GBRT model in step 1 includes: collecting 1000 sets of historical process data containing residual adhesive feature input vectors and optimal Plasma output parameters, 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; and setting the optimal Plasma parameters to include a carbon tetrafluoride gas flow rate of 20-50 sccm, an RF power of 100-300 W, and a processing time of 30-120 s.

[0014] Preferably, in step 2, the sand-to-water ratio in the mixture of corundum and water is calculated as R=0.2×Havg+1 when the average residual adhesive thickness Havg<5μm, and is fixed at 1:2 when Havg≥5μm.

[0015] Preferably, the sandblasting parameters 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.

[0016] Preferably, 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 an 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.

[0017] Preferably, the vacuum degree 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%.

[0018] Preferably, in step 4, the secondary AOI inspection uses a 5-megapixel industrial camera, and the size deviation of the inspection-bound PAD is ≤ ±0.005mm.

[0019] Preferably, the process further includes a second Plasma treatment after the second AOI inspection, followed by sequential steps of electroless nickel-palladium-gold plating, IR infrared drying, plate warping and pressing, electric milling of the shape, and electrical testing; wherein the IR infrared drying temperature is 110-130℃; and the plate warping and pressing pressure is 0.2-0.3MPa.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. By adopting the GBRT model to adaptively output Plasma parameters and combining the sandblasting path planned by the A* algorithm, dynamic matching between residual adhesive characteristics and processing parameters is achieved. The prediction error of the GBRT model for Plasma parameters is ≤5%, ensuring priority coverage of the area to be optimized, and the path overlap rate is ≤5%. Ultimately, the residual adhesive removal rate is increased to over 98%, solving the problem of incomplete processing by traditional fixed parameters.

[0022] 2. A vacuum laminator is used to achieve secondary dry film bonding without floating, and secondary AOI inspection is used to accurately monitor the PAD size, so that the PAD size deviation is controlled within ±0.005mm, which meets the stringent requirements of chip bonding.

[0023] 3. The A* algorithm improves sandblasting efficiency by 30% through path planning, while the low-pressure, high-speed processing strategy in non-optimized areas reduces the copper surface scratch rate to below 0.5%; the linkage of residual adhesive feature data among various processes reduces the rework rate to below 3%; the improvement in the thoroughness of residual adhesive removal and PAD dimensional accuracy increases the adhesion of the electroless nickel palladium gold plating by 20%; the board warping and pressing baking process further ensures the flatness of the substrate and meets the high and low temperature cycling reliability requirements from -40℃ to 125℃. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a processing method for solving the problems of residual adhesive on the copper surface and undersized bonding pads in Cavity products, according to an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments. Example 1

[0027] After laser processing, the bottom copper surface of the cavity area may retain carbon powder residue and resin residue from the laser process. By employing a combination of plasma and sandblasting processes, these residues can be completely removed. Furthermore, due to the uneven surface of the cavity area, a vacuum lamination process is used to ensure the dry film of the bonding pads at the cavity edge does not detach. This prevents damage to the bonding pad dimensions during subsequent development and etching processes, avoiding issues such as undersized bonding pads. An additional AOI scan after etching further ensures the high-quality production of the cavity products.

[0028] Please see Figure 1 As shown, this embodiment discloses a processing method to solve the problems of residual adhesive on the copper surface and undersized bonding pads in Cavity products, including the following processing steps:

[0029] Outer Layer Lamination: Laminating multilayer core boards, prepregs, etc., into a single multilayer substrate structure; → Laser Browning: Treating the copper surface to increase roughness and oxidation resistance, improving interlayer adhesion; → Laser: Using lasers for fine processing, preparing for subsequent processes; → Through-Hole Drilling: Drilling holes through the substrate for interlayer electrical connections; → Removal of Adhesive Sludge: Removing adhesive sludge generated during drilling to ensure clean hole walls; → Blind Via AOI: Automated optical inspection of blind via size, location, etc., to screen for defective blind vias; → PTH: Depositing metallic copper on the hole walls to make them conductive, achieving interlayer bonding; → Hole Filling Plating: Filling holes and electroplating for thickness, ensuring full filling and adequate copper layer thickness; → Outer Layer Pretreatment: Cleaning and roughening the outer copper surface to improve adhesion to dry film / ink. The process involves several steps: → Bonding strength; → Outer layer exposure: Transferring the outer layer circuit pattern onto the dry film, protecting the copper surface to be retained; → Outer layer etching: Etching the copper layer not protected by the dry film to form the outer layer circuit; → Impedance measurement: Measuring the characteristic impedance of the circuit to ensure it meets design requirements; → Outer layer AOI: Automated optical inspection of the outer layer circuit to check for defects such as open circuits, short circuits, and abnormal line widths; → Solder mask plugging: Filling holes that do not need to be soldered to prevent solder paste from entering and protect the circuit inside the holes; → Solder mask printing: Printing solder mask ink to protect the copper surface in non-soldered areas from oxidation and short circuits; → Solder mask exposure: Transferring the solder mask pattern to determine the areas where the solder mask layer needs to be retained; → Finished product impedance: Re-measuring the finished product impedance to ensure the final performance meets specifications; → Text printing: Printing text such as model number and production information. The process includes: → Marking for easy product identification; → Laser Cavity: Laser processing creates the cavity structure to house chips and other components; → Plasma Treatment: Plasma treatment of the copper surface of the cavity removes residual adhesive and improves cleanliness; → Sandblasting: Further cleaning of impurities in the cavity area and optimization of the copper surface morphology; → Secondary Dry Film: Re-lamination of the dry film prepares for subsequent pattern transfer of bonded PADs; → Secondary Outer Layer Exposure: Secondary exposure of the bonded PADs and other fine structures for precise pattern transfer; → Secondary Outer Layer Acid Etching: Acid etching of the copper layer not protected by the dry film to form fine bonded PADs and other structures; → Film Removal: Removal of the dry film exposes the copper surface requiring further processing; → Secondary Outer Layer AOI: Inspection of the bonded PADs after secondary processing. 1. Screening for defective products based on size, shape, etc.; 2. Secondary Plasma Treatment: Secondary plasma treatment to optimize copper surface performance and prepare for electroless nickel-palladium-gold plating; 3. Electroless Nickel-Palladium-Gold Plating: Depositing nickel, palladium, and gold plating on the copper surface to improve solderability, oxidation resistance, and reliability; 4. Infrared Heat Treatment (IR): Infrared heating to dry and cure the plating or coating, ensuring stable performance; 5. Board Warping Correction: Pressing and baking to correct board warping and ensure substrate flatness; 6. Electro-milling: Precisely machining the substrate shape to meet final size and shape requirements; 7. Electrical Testing: Testing the continuity and insulation of electrical circuits to ensure no electrical defects; 8. Visual Inspection: Manual visual inspection supplementing automatic detection to screen products with appearance or minor defects; 9. Outgoing Quality Control (OQC): Final outgoing quality inspection to ensure products meet quality standards.→Packaging: Packaging qualified products to facilitate storage, transportation, and delivery.

[0030] It should be noted that by employing both Plasma and sandblasting processes, it is entirely possible to completely remove residual carbon powder from the laser-processed cavity and any resin residue that could not be completely burned off by the laser, ensuring product reliability. Specifically: Plasma: The tetrafluoroethylene gas used in the Plasma equipment effectively removes residual adhesive from the large copper foil at the bottom of the cavity, ensuring reliability. The uniformity of adhesive removal using Plasma must reach over 80%. Sandblasting: A mixture of diamond abrasive and water is sprayed onto the board surface, further effectively removing residual adhesive from the large copper foil at the bottom of the cavity and carbon powder from the laser-burned PP.

[0031] The Plasma processing procedure is as follows: residual adhesive feature acquisition → Plasma parameter adaptive algorithm → Plasma processing;

[0032] Specifically, the residual adhesive feature acquisition process is as follows: A high-resolution line scan camera combined with a coaxial light source is used to acquire images of the copper surface after laser cavity treatment, and the grayscale value, texture features, and coordinate information of the residual adhesive area are output simultaneously; the high-resolution line scan camera has a resolution of 2048×1080 pixels, and the image acquisition rate is 10 frames / second; the grayscale value of the residual adhesive area reflects the thickness of the residual adhesive: every 10 increase in grayscale value corresponds to an increase of 2μm in the thickness of the residual adhesive; the texture features reflect the type of residual adhesive: the texture entropy value of carbon powder residue is <5, and the texture entropy value of resin residue is >8.

[0033] Simultaneously, the collected residual adhesive features are preprocessed, Gaussian filtering is applied to the collected images to remove noise, Otsu threshold segmentation is used to extract the residual adhesive region, and morphological operations are used to optimize the edge contour, finally outputting a standardized feature vector; the standardized feature vector includes: average residual adhesive thickness Tavg, maximum residual adhesive thickness Tmax, residual adhesive type C, and residual adhesive distribution density D.

[0034] Furthermore, the thickness of the residual adhesive area is calculated using the formula T=0.2×(G-Gb), where G is the grayscale value of the residual adhesive area in the acquired image, Gb is the reference grayscale value of the copper surface without residual adhesive (calibrated by pre-collecting standard samples without residual adhesive), and 0.2 is the grayscale-thickness conversion coefficient based on experimental data fitting. Every 10 grayscale values ​​correspond to a thickness of 2μm. The average thickness of the residual adhesive, Tavg, is calculated based on the arithmetic mean of the thicknesses of all residual adhesive areas, ranging from 5 to 20μm. At the same time, the maximum thickness value in the residual adhesive area is recorded as the maximum residual adhesive thickness Tmax. The ratio of the residual adhesive area to the total area of ​​the copper surface with Cavity is recorded as the residual adhesive distribution density D.

[0035] Furthermore, the residue type C is 1 and 0, where C=0 represents toner and C=1 represents resin. The texture entropy value E is calculated through the gray-level co-occurrence matrix to reflect the complexity of the surface texture of the residue. When E is less than 5, it is determined to be toner residue, and the residue type C=0 is recorded. The toner residue texture is regular and the entropy value is low. When E is greater than 8, it is determined to be resin residue, and the residue type C=1 is recorded. The resin residue texture is messy and the entropy value is high.

[0036] Specifically, the application process of the Plasma parameter adaptive algorithm is as follows: a trained gradient boosting regression tree (GBRT) model is used, with input vector: [Tavg,Tmax,C,D], and output parameters: carbon tetrafluoride gas flow rate F, radio frequency power P, and processing time t;

[0037] The GBRT model dynamically adjusts its parameters according to the following rules: when the value of Tavg increases, the values ​​of F, P, and t are all increased to enhance the etching intensity; when C=1, compared to C=0, F increases by 10-15 sccm; when the value of D increases, t increases by 20-30 s; the dynamically adjusted output parameters are sent to the Plasma device controller through the communication interface to achieve adaptive adjustment of Plasma parameters; and the next Plasma processing is then performed according to the adaptively adjusted Plasma parameters.

[0038] Furthermore, the GBRT model is trained as follows:

[0039] a1: Collect 1000 sets of historical process data. Each set of data includes input features and output parameters. The input features include the average residual adhesive thickness Tavg, the maximum residual adhesive thickness Tmax, the residual adhesive type C, and the residual adhesive distribution density D. The output parameters are the experimentally verified optimal Plasma process parameters, including the carbon tetrafluoride gas flow rate F, the RF power P, and the processing time t. These optimal Plasma process parameters must meet the following process requirements: Plasma adhesive removal uniformity ≥95%, residual adhesive content ≤3μm, and copper surface damage depth ≤0.5μm.

[0040] a2: Configure the key hyperparameters of the GBRT model: 300 regression trees to balance training accuracy and efficiency; learning rate of 0.1 to control the contribution weight of each tree to the final prediction; maximum depth of a single tree of 5 to limit the complexity of a single tree and avoid overfitting; subsampling ratio of 0.8 to enhance generalization ability by randomly selecting 80% of the samples during training of each tree.

[0041] a3: Divide the 1000 sets of data into 800 sets for training and 200 sets for monitoring the training process in an 8:2 ratio; use a stepwise fitting residual strategy for training: initialize the predicted values ​​as the mean of the training set of output parameters;

[0042] a4: For each regression tree, first calculate the residual between the current model's predicted value and the true value, which is the difference between the true value and the current predicted value; train a regression tree using the input features of the training set and the residual, learn how to fit the residual, 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.

[0043] a5: Repeat step a4 above until 300 trees have been trained, or the prediction error on the validation set no longer decreases; in addition, to ensure the generalization ability of the model, the 1000 sets of data are randomly divided into 5 equal-sized subsets, and 4 subsets are used for training and 1 subset for validation each time, which is repeated 5 times; through cross-validation and adjustment of hyperparameters, the prediction error is finally made ≤5%, that is, the deviation between the predicted value and the true optimal value is ≤5%.

[0044] a6: The trained GBRT model is deployed 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, and send them to the Plasma device controller through the communication interface to achieve adaptive parameter adjustment.

[0045] It should be noted that the algorithm model breaks through the limitations of fixed parameters in traditional Plasma: on the one hand, it accurately captures dynamic features such as residual adhesive thickness, type, and distribution, allowing parameters to be optimized in real time according to the differences in residual adhesive in the product; on the other hand, relying on the accurate prediction and closed-loop control of the GBRT model, it significantly improves the uniformity of Plasma adhesive removal, keeping the amount of residual adhesive within limits, while limiting the depth of copper surface damage. This ensures the cleanliness of the copper surface in Cavity and avoids excessive etching of the copper surface, laying a high-quality foundation for subsequent processes. It effectively solves the product defects caused by uneven removal of residual adhesive and improves the processing yield and quality stability of Cavity products.

[0046] The sandblasting process is as follows: pre-sandblasting residual adhesive re-inspection → sandblasting parameter dynamic optimization algorithm → sandblasting treatment;

[0047] Specifically, the method for re-inspecting residual adhesive before sandblasting is as follows: a laser thickness gauge is used to perform spot measurements on the copper surface after Plasma treatment, with a measurement point density of 1 mm. 2 Evidence of a two-dimensional distribution of the remaining adhesive thickness at a single measuring point. m is the number of rows of measurement points, n is the number of columns of measurement points, and each element in the matrix... Representing coordinates The residual adhesive thickness at the location was then used to screen for those that met the requirements. Regions larger than 3μm are defined as regions to be optimized.

[0048] Specifically, the analysis process of the dynamic optimization algorithm for sandblasting parameters is as follows: For the average residual adhesive thickness Havg in the area to be optimized, a dynamic calculation model is established for the spraying pressure L, sand-to-water ratio R, and nozzle moving speed V.

[0049] The formula for calculating the spray pressure L is: L = 0.1 × Havg + 0.2. The thicker the residual adhesive, the greater the pressure required to enhance the sandblasting's grinding ability.

[0050] The formula for calculating the sand-to-water ratio R is: When the residual adhesive thickness Havg < 5μm, the sand ratio is increased linearly; when the residual adhesive thickness Havg ≥ 5μm, the sand-to-water ratio is fixed at 1:2 to avoid excessive sand particles scratching the copper surface.

[0051] The formula for calculating the nozzle movement speed V is: The thicker the residual adhesive, the slower the nozzle movement speed needs to be to extend the grinding time; when the residual adhesive thickness Havg≥8μm, the fixed speed is 1mm / s to ensure that the thick residual adhesive is fully removed.

[0052] Wherein, the average residual adhesive thickness Havg of the region to be optimized is the total residual adhesive thickness within the region to be optimized. The arithmetic mean.

[0053] Furthermore, the A* algorithm is used to plan the movement path of the nozzle. Its core logic is: prioritize covering the area to be optimized, and the path overlap rate is ≤5%; for the non-optimized area, low pressure and high speed are used to ensure that there is no residual adhesive on the copper surface while avoiding scratches on the copper surface.

[0054] Specifically, the application steps of the A* algorithm in sandblasting nozzle path planning are as follows:

[0055] b1: The area to be processed on the Cavity copper surface is discretized into a two-dimensional mesh map, with each mesh serving as a node in the algorithm. The mesh size can be set according to the process accuracy requirements to ensure accurate coverage of the residual adhesive area while controlling computational complexity. Each node contains coordinate information. And the residual adhesive thickness label at that location, the residual adhesive thickness label includes the area to be optimized H>3μm or the non-optimized area H≤3μm.

[0056] b2: Define the evaluation function ,in Let n be the actual sandblasting cost from the starting point to the current node n; the cost is defined as "a comprehensive measure of the potential damage risk of sandblasting to the copper surface and the process efficiency": if node n belongs to the region to be optimized, Set to 1; if node n belongs to a region not to be optimized, Set it to 5; The Manhattan distance from the current node n to the nearest node in the region to be optimized is calculated using a heuristic function. This heuristic function ensures that the estimated distance is no greater than the actual distance, thus guaranteeing that the A* algorithm can find the optimal path.

[0057] b3: Open list: Stores nodes to be inspected, categorized by... Sort values ​​from smallest to largest; select from the open list each time. Expand the smallest node to ensure that the most promising path is searched first; close the list: store the nodes that have been checked to avoid processing the same node repeatedly and improve the efficiency of the algorithm.

[0058] b4: Path search process: Add the initial position node of the nozzle to the open list, calculate its... Take from the open list The smallest node n is added to the closed list, and the successor node n is checked to see if it belongs to the region to be optimized.

[0059] If the node belongs to the region to be optimized, mark it as prioritized for coverage, then generate its 8 neighboring 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 thus obtain Add the child nodes to the open list.

[0060] If it belongs to a region that is not to be optimized, it also generates 8 neighboring child nodes, but the calculation... If the child node still belongs to the non-optimized region, then If a child node enters the region to be optimized, then .

[0061] b5: Termination condition: When the open list is empty, or the number of times all nodes in the region to be optimized are covered meets the path overlap rate ≤ 5%, that is, the number of times the same node in the region to be optimized is covered by the path accounts for ≤ 5% of the total number of searches, to avoid local over-blasting, then stop the search and obtain the final nozzle path.

[0062] When the nozzle passes through the area to be optimized along the planned path, the dynamic optimization algorithm for sandblasting parameters is triggered. Based on the average thickness of residual adhesive in that area (Havg), the spraying pressure (L), sand-to-water ratio (R), and nozzle moving speed (V) are output in real time. When the nozzle passes through a non-optimized area, it automatically switches to fixed parameters: spraying pressure (L=0.2MPa), nozzle moving speed (V=5mm / s), and sand-to-water ratio is kept at a normal low ratio to ensure that there is no residual adhesive on the copper surface while avoiding scratches on the copper surface.

[0063] It should be noted that, through the heuristic search and cost function design of the A* algorithm, the sandblasting path can not only prioritize and fully cover the areas to be optimized with heavy adhesive residue, but also quickly and gently handle the non-optimized areas with slight adhesive residue. Ultimately, the process target of adhesive residue removal rate ≥98% and copper surface scratch rate ≤0.5% is achieved, which provides a guarantee for the subsequent bonding of PAD dimensional accuracy.

[0064] The above formulas are all dimensionless numerical calculations. These formulas are derived from software simulations using collected data to obtain the most recent real-world results. The coefficients are used to quantify each parameter to obtain a specific numerical value. Regarding the coefficients, the size only needs to be sufficient to maintain the proportional relationship between the parameter and the quantized value. Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0065] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret 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. 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 grayscale value of the residual adhesive area in the acquired image, Gb is the reference grayscale value of the copper surface without residual adhesive (calibrated by pre-acquiring standard samples without residual adhesive), and 0.2 is the grayscale-thickness conversion coefficient based on experimental data fitting, with each 10 grayscale values ​​corresponding to a thickness of 2μm. 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 C=0; when E>8, it is determined to be resin residue, and the residual adhesive type C=1. The training process of the gradient boosting regression tree model described in step 1 includes: collecting 1000 sets of historical process data containing residual adhesive feature input vectors and optimal Plasma output parameters, 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; 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. 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 <5μm, and is fixed at 1:2 when Havg ≥5μm. 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 < 8μm, and fixed speed is 1mm / s when Havg ≥ 8μm, where Havg is the arithmetic mean of the residual adhesive thickness in the area to be optimized. 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 region is treated with a spray pressure of 0.2MPa and a nozzle speed of 5mm / s. 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 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%.

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 secondary AOI inspection in step 4 uses a 5-megapixel industrial camera, and the size deviation of the detected PAD is ≤ ±0.005mm.

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, 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