A reduction plating method and system for optimizing the uniformity of a thick gold layer of a packaging substrate

Abstract

The application provides a kind of optimization package substrate thick gold layer uniformity reduction electroplating method and system, belong to semiconductor packaging technical field.For the problem that existing global regulation cannot adapt to the difference of pattern density on the surface of substrate, the thickness uniformity of thick gold layer is insufficient, the application adopts partition closed-loop control architecture: based on the adaptive division control sub-region of line geometry characteristics, combined with copper thick sidewall effect and line width edge effect calculation three-dimensional pattern density factor;Three-stage process of nucleation feedforward, fast deposition and densification feedback is executed, according to the thickness deviation and density factor, the difference of multi-channel liquid supply system is differentiated to supplement reducing agent, synchronous correction hardware disturbance and sub-region cross interference, maintain the global concentration stability of plating solution.The application significantly improves the uniformity of thick gold layer, the process stability and compensation accuracy are high, meet the requirements of high-end packaging gold wire bonding.

Description

Technical Field

[0001] This invention relates to the field of semiconductor packaging technology, and in particular to a reduction electroplating method and system for optimizing the uniformity of thick gold layers on packaging substrates. Background Technology

[0002] When using electroless nickel-palladium-gold (ENEPIG) or electroless nickel-gold (ENIG) plating processes on packaging substrates, the top thick gold layer is primarily used for wire bonding. The bonding process demands strict uniformity in gold layer thickness; areas that are too thin will result in insufficient bonding strength, while areas that are too thick will waste gold material and may cause bonding parameter drift. Currently, the industry mainly relies on overall control methods to improve the uniformity of the thick gold layer. Common practices include uniformly increasing the plating bath temperature, increasing the reducing agent concentration, or strengthening the circulation and agitation within the plating bath. A common characteristic of these methods is that they apply a uniform, global effect across the entire plating bath.

[0003] The core problem with existing technologies lies in the heterogeneity of the substrate surface itself: the density of circuit patterns varies significantly across different regions of the actual substrate. The difference between densely packed circuit areas and open copper areas causes asynchronous flow patterns in the plating solution, mass transfer boundary layer thickness, and local reactant consumption rates. This leads to intrinsic differences in the diffusion of reducing agents to the reaction interface, ultimately resulting in inconsistent deposition rates. Although existing production lines have achieved the detection and adjustment of bath temperature, pH value, and reducing agent concentration, this type of control targets global process parameters and cannot make independent, differentiated interventions based on the differences in pattern density in different local areas of the substrate surface. Essentially, it is a feedback-based allocation strategy without spatial resolution. Therefore, how to implement differentiated and quantifiable compensation adjustments for different pattern density regions on the packaging substrate surface remains a long-standing technical bottleneck in this field.

[0004] This invention proposes a reduction electroplating method and system for optimizing the uniformity of thick gold layers on packaging substrates, in order to solve the aforementioned technical problems. Summary of the Invention

[0005] This invention provides a reduction electroplating method and system for optimizing the uniformity of thick gold layers on packaging substrates, in order to solve the aforementioned technical problems.

[0006] This invention provides a reduction electroplating method for optimizing the uniformity of a thick gold layer on a packaging substrate, comprising: Step 1: Obtain the surface circuit pattern data of the substrate to be plated, divide the substrate surface into several control sub-regions, and calculate the pattern density factor of each sub-region based on the circuit area ratio. Step 2: Immerse the substrate to be plated into the electroless gold plating solution and perform a phased reduction electroplating, which includes at least a nucleation stage, a rapid deposition stage and a densification stage performed sequentially. Each stage corresponds to a different set of basic process parameters. The nucleation stage performs differentiated feedforward supply of reducing agent based on the pattern density factor, and the rapid deposition stage and the densification stage perform thickness feedback control. Step 3: Starting from the rapid deposition stage, the gold plating thickness of each sub-region is collected in real time at a fixed sampling period. The average value of multiple sampling points in each sub-region is used as the gold plating thickness. Step 4: Based on the thickness of the gold plating layer, calculate the instantaneous thickness deviation of each sub-region; Step 5: Based on the graphic density factor and instantaneous thickness deviation, use the compensation model to generate the adjustment value of reducing agent replenishment for each sub-region; Step 6: Using a multi-channel liquid supply system, the adjustment value is first cross-corrected based on the cross-interference coupling coefficient of adjacent sub-regions, and then a reducing agent is quantitatively added to the near-wall boundary layer of the plating solution in each sub-region according to the corrected adjustment value. Step 7: Repeat steps 3 to 6 until the gold plating thickness of all sub-regions is within the allowable tolerance zone of the preset target thickness for m0 consecutive sampling periods.

[0007] Preferably, the graph density factor for each sub-region is calculated, including: Obtain the line area ratio of each control sub-region and calculate the initial plane density factor of each control sub-region; Meanwhile, based on the copper thickness data of the circuit of the substrate to be plated, a depth topography constraint map of each sub-region is constructed. The depth topography constraint map is used to characterize the range of reducing agent dissipation influence corresponding to the three-dimensional topography of the circuit and to delineate the effective area of ​​the sidewall effect. Within the effective area defined by the depth topography constraint map, the effective reaction area of ​​the sidewall is compensated for the corresponding initial planar density factor to generate a constraint correction density factor. Based on the local gradient change of the constraint correction density factor, each sub-region is adaptively divided into multiple mass transfer micro-domains, and the consumption weight within each mass transfer micro-domain is assigned according to the specific surface area of ​​the exposed lines in each micro-domain. The constraint correction density factor of each mass transfer micro-domain within the same sub-region is normalized and weighted by the corresponding consumption weight to obtain the graphic density factor of the corresponding sub-region.

[0008] Preferably, the depth topography constraint map of each sub-region is constructed based on the copper thickness data of the circuit of the substrate to be plated, including: Obtain the copper thickness distribution of the line within the corresponding sub-region, and construct a digital elevation model of the line body based on the plane coordinates of the sub-region and the copper thickness value of each point as the height value. On the digital elevation model, a gradient operator is used to extract the edge gradient vector field at the edge of the line, wherein the edge gradient vector field includes the maximum slope direction and the corresponding gradient magnitude of each edge point; Based on the edge gradient vector field, the gradient amplitude is used as the modulation coefficient and combined with the preset sidewall dissipation coefficient to construct the sidewall consumption flux boundary condition, where the consumption flux is the outflow flux of the liquid phase solution domain; the liquid phase plating solution region within the sub-region is used as the solution domain, the outer boundary of the sub-region is used as the first boundary of the plating solution body concentration, and the line sidewall is used as the second boundary of the consumption flux. The two-dimensional steady-state diffusion equation is solved to obtain the reducing agent concentration distribution field covering the entire liquid phase domain of the sub-region. Map the reducing agent concentration distribution field as a scalar field of dissipative effects; The equipotential surface of the dissipation influence scalar field is extracted, and the equipotential surface corresponding to the preset dissipation influence threshold is selected as the dissipation boundary surface. The dissipation boundary surface serves as the depth topography constraint map of the corresponding sub-region to characterize the effective boundary range of the reaction dissipation influence caused by the three-dimensional topography of the line.

[0009] Preferably, the surface circuit pattern data of the substrate to be plated is obtained, and the substrate surface is divided into several control sub-regions, including: Line entity recognition is performed on the surface line graphic data, and geometric feature parameters of each line entity are extracted, wherein the geometric feature parameters include at least line width, line spacing and line centerline curvature; Using the geometric feature distribution clustering of line entities as the primary classification criterion, line entities with geometric feature similarity higher than a preset threshold are grouped into the same primary partition, generating several geometrically homogeneous regions. Within each geometrically homogeneous region, a spatial adjacency graph of the line entities is constructed, wherein the spatial adjacency graph is constructed with each line entity as a node and the minimum distance between two line entities as an edge connection relationship based on whether it is less than a preset diffusion coupling threshold. Under the constraints of the spatial adjacency graph, a region partitioning cost function is constructed with the dual optimization objectives of local density equilibrium of line area and equilibrium of equivalent deposition rate distribution of line within the sub-region. The spatial adjacency graph is iteratively segmented using a graph cut algorithm. Each segmentation selects the edge set that maximizes the decrease in the region partitioning cost function for cutting, until the coefficient of variation of the local density of the line area and the coefficient of variation of the equivalent deposition rate distribution of the line in the spatial region corresponding to each segmented subgraph are reduced to below the preset equilibrium threshold, and the total number of control sub-regions obtained by segmentation does not exceed the preset maximum number of partitions. Each spatial region obtained from the final segmentation is taken as the control sub-region.

[0010] Preferably, the region partitioning cost function is constructed, including: The line entities within the geometrically homogeneous region are classified into at least two line width levels according to their line width dimensions, and a corresponding area conversion weight is assigned to each line width level. The area conversion weight corresponding to narrow lines is greater than that corresponding to wide lines. For any candidate sub-region, the equivalent deposition area is obtained by multiplying the projected area of ​​each line entity by the area conversion weight of the corresponding line width level of the line entity, and the weighted equivalent density of the candidate sub-region is calculated based on the equivalent deposition area. The maximum value of the coefficient of variation of the weighted equivalent density within each sub-region is taken as the first cost term, and the maximum value of the coefficient of variation of the linewidth equivalent deposition rate within each sub-region is taken as the second cost term. The region partitioning cost function is constructed by the weighted sum of the first and second cost terms.

[0011] Preferably, the reducing agent is replenished quantitatively via a multi-channel liquid supply system, including: Acquire the operation status monitoring data of the multi-channel liquid supply system during the current process cycle. The operation status monitoring data includes at least the opening and closing response delay of each channel nozzle, the amplitude of pipeline residual pressure fluctuation, and the near-wall temperature drift of the sub-region. Normalize the deviation between the current value of each disturbance in the operation status monitoring data and the historical benchmark value of the corresponding disturbance to construct a multi-source disturbance deviation vector. Based on the preset perturbation coupling coefficient matrix, the multi-source perturbation deviation vector is mapped to the hardware perturbation correction factor of each sub-region. At the same time, based on the preset adjacent sub-region cross-interference coupling coefficient matrix, the concentration contribution caused by the diffusion of the supplementary amount in adjacent sub-regions is calculated, and the cross-interference correction factor is generated. The adjustment value is multiplicatively corrected by the product of the hardware disturbance correction factor and the cross-interference correction factor to obtain the actual replenishment amount, and quantitative replenishment is performed according to the actual replenishment amount, wherein the actual replenishment amount is constrained between preset upper and lower limits of flow.

[0012] Preferably, constructing a multi-source perturbation bias vector includes: The deviations between the current values ​​of each disturbance in the operation status monitoring data and the corresponding historical benchmark values ​​are mapped to the flow attenuation rate of the corresponding channel, the increase in pipeline flow resistance, and the cumulative near-wall temperature drift of each sub-region in the current process cycle through a preset empirical conversion relationship, and are used as the loss input parameters for the current cycle. The uniformity deviation of the gold plating thickness in the corresponding sub-region of multiple substrates that have completed plating under historical process cycles is recorded, as well as the flow rate attenuation rate, pipeline flow resistance increment, and near-wall temperature drift accumulation in the multiple completed plating cycles. Using the uniformity deviation record as the output reference, and taking the flow rate attenuation rate and pipeline flow resistance increment as the first input dimension, and the flow rate attenuation rate and near-wall temperature drift accumulation as the second input dimension, a first-order loss feature mapping table and a second-order coupling loss feature mapping table are established for each sub-region. The first-order loss feature mapping table records the first-order system deviation value of the supplementary amount corresponding to a single input dimension combination, and the second-order coupling loss feature mapping table records the coupling compensation term that exceeds the linear superposition of the corresponding first-order system deviation value under the cross combination of two input dimensions. Based on the loss input parameters of the current period, determine the current first-order loss input combination and second-order coupling loss input combination of each sub-region, and obtain the first-order system deviation value and second-order coupling compensation term from the first-order loss feature mapping table and the second-order coupling loss feature mapping table, respectively. The first-order system deviation value is added to the second-order coupling compensation term to obtain the composite system deviation value, and the deviation of each disturbance quantity is biased and corrected using the composite system deviation value to obtain the biased disturbance quantity deviation. The bias values ​​of each disturbance quantity after bias correction are combined to form the multi-source disturbance bias vector.

[0013] This invention provides a reduction electroplating system for optimizing the uniformity of thick gold layers on packaging substrates, comprising: The factor acquisition module is used to acquire the surface circuit pattern data of the substrate to be plated, divide the substrate surface into several control sub-regions, and calculate the pattern density factor of each sub-region based on the proportion of the circuit area. The gold plating module is used to immerse the substrate to be plated into a chemical gold plating solution and perform a phased reduction electroplating, which includes at least a nucleation stage, a rapid deposition stage and a densification stage performed sequentially. Each stage corresponds to a different set of basic process parameters. In the nucleation stage, the reducing agent is fed forward in a differentiated manner based on the pattern density factor. In the rapid deposition stage and the densification stage, thickness feedback control is performed. The thickness acquisition module is used to collect the gold plating thickness of each sub-region in real time at a fixed sampling period starting from the rapid deposition stage. The average value of multiple sampling points in each sub-region is used as the gold plating thickness. The deviation calculation module is used to calculate the instantaneous thickness deviation of each sub-region based on the thickness of the gold plating layer; The adjustment determination module is used to generate adjustment values ​​for the amount of reducing agent replenishment for each sub-region based on the graphic density factor and instantaneous thickness deviation using a compensation model; The quantitative replenishment module is used to first cross-correct the adjustment value based on the cross-interference coupling coefficient of adjacent sub-regions through a multi-channel liquid supply system, and then quantitatively replenish the reducing agent to the near-wall boundary layer of the plating solution in each sub-region according to the corrected adjustment value. The cyclic adjustment module is used to repeatedly execute the thickness acquisition module, deviation calculation module, adjustment determination module, and quantitative supplementation module until the gold plating thickness of all sub-regions is within the allowable tolerance zone of the preset target thickness within m0 consecutive sampling cycles.

[0014] Compared with the prior art, the beneficial effects of this application are as follows: 1. Significantly improved thickness uniformity: By using independent feedback in zones and pattern density feedforward, the thickness uniformity of thick gold layers can be improved from the traditional global control precision, meeting the requirements of high-end packaging bonding processes.

[0015] 2. Strong process stability: The global concentration balance mechanism ensures no concentration drift during long-term operation in mass production, and the upper and lower limits of flow rate constraints avoid side effects such as coating porosity and self-decomposition caused by excessive reducing agent.

[0016] 3. High compensation accuracy: It simultaneously corrects hardware aging disturbances and sub-region cross-diffusion interference, reducing the deviation between the compensation amount and actual needs.

[0017] 4. Full-stage coverage: Feedforward compensation in the nucleation stage eliminates initial nucleation unevenness and effectively improves the uniformity of the final coating.

[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a reduction electroplating method for optimizing the uniformity of a thick gold layer on a packaging substrate, as described in an embodiment of the present invention. Figure 2 This is a structural diagram of a reduction electroplating system for optimizing the uniformity of a thick gold layer on a packaging substrate, as described in an embodiment of the present invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] Reduction plating refers to a process in which gold ions in a plating solution are reduced by a chemical reducing agent and deposited onto the conductive circuit surface of a packaged substrate under conditions without an external current. It is essentially equivalent to chemical gold plating (electrolytic plating) and embodies the deposition mechanism driven by a reducing agent.

[0023] Plating solution system: Chemical gold plating solution with sodium gold sulfite The main salt, with sodium borohydride Sodium hypophosphite As a reducing agent, it is supplemented with a stabilizer (such as disodium ethylenediaminetetraacetate), a complexing agent (sodium sulfite), and a pH buffer (a mixture of sodium dihydrogen phosphate and disodium hydrogen phosphate, with the pH controlled at 7.0 ± 0.2). The sodium borohydride system is used as the main example in this paper, and its basic concentration is defined as the reducing agent concentration required to achieve an average plating rate of 0.1 μm / min under the conditions of a plating bath temperature of 65℃ and pH = 7.0. A typical value is... =0.5 g / L. The baseline concentration of the sodium hypophosphite system is determined separately, but the principle is the same.

[0024] The physical definition of the near-wall boundary layer of a plating solution refers to the liquid region adjacent to the surface of the substrate circuitry, and its thickness is taken as the velocity boundary layer thickness of the substrate surface. Under forced circulation conditions of the plating solution (with an impeller pump driving the solution to flow from bottom to top at a velocity of approximately 0.1-0.3 m / s), the velocity boundary layer thickness is... The boundary layer is in the range of 0.5 mm to 5 mm. Here, the boundary layer is the near-wall velocity gradient region in a hydrodynamic sense, which is a different scale concept from the Nernst diffusion layer (typically ≤100 μm, dominated by mass transfer and diffusion). The reducing agent replenishment is injected into this near-wall region, where a concentration gradient is formed through slow convection and molecular diffusion, and then reaches the reaction interface via the Nernst diffusion layer.

[0025] Partitioning and Sampling Period Constraints: The feature size and sampling period of the controlled sub-regions must meet the following requirements: within a single sampling period, the proportion of local concentration deviation caused by the lateral diffusion of the reducing agent between sub-regions is less than 5%, ensuring that lateral cross-interference between sub-regions is negligible. Root mean square distance formula for lateral diffusion: ,in, For the effective diffusion coefficient, The sampling period is [value]. When [condition]... hour, When the feature size of the sub-region is ≥2mm, the diffusion distance ratio is ≤5.5%, which can be approximated as cross-interference <5% in engineering, satisfying the premise of independent control of the partition. This embodiment uses a 5mm×5mm mesh, which is much higher than this constraint threshold.

[0026] In this embodiment, the total reducing agent replenishment amount of the multi-channel liquid supply system matches the total global consumption of the process to maintain a stable concentration in the plating bath. The zone compensation only distributes the total replenishment amount to each sub-region according to the compensation model, and does not add extra flow rate on top of the global amount. The total global consumption is calculated from the target plating rate, total reaction area, and Faraday's law, and fine-tuning is achieved through online detection of the bath concentration.

[0027] Example 1: This invention proposes a reduction electroplating method to optimize the uniformity of thick gold layers on packaging substrates, such as... Figure 1 As shown, it includes: Step 1: Obtain the surface circuit pattern data of the substrate to be plated, divide the substrate surface into several control sub-regions, and calculate the pattern density factor of each sub-region based on the circuit area ratio. Acquire surface circuit pattern data of the substrate. Implementation method: Use a linear CCD scanning camera (5μm / pixel resolution) to perform area scan imaging on the substrate, acquire grayscale images and transmit them to an industrial control computer; run the Canny edge detection algorithm to extract the line contour vector coordinates, and output the line layout data in DXF or GDSII format.

[0028] The substrate surface is divided into several control sub-regions. The simplest division method is to divide it into equal-area rectangular grids. For example, a 100mm × 100mm substrate is divided into 5mm × 5mm grids, resulting in 400 control sub-regions. A more refined adaptive division method is given in Example 4.

[0029] For each sub-region i, calculate the graphic density factor based on the proportion of the line area. The simplest implementation is the proportion of the two-dimensional top surface projection: ,in, It is the sum of the projected areas of the top surfaces of all lines within sub-region i. This represents the total area of ​​the sub-region. This two-dimensional density is a simplified version of the graphic density factor; the precise three-dimensional calculation method is given in Example 2.

[0030] Step 2: Immerse the substrate to be plated into the electroless gold plating solution and perform a phased reduction electroplating, which includes at least a nucleation stage, a rapid deposition stage and a densification stage performed sequentially. Each stage corresponds to a different set of basic process parameters. The nucleation stage performs differentiated feedforward supply of reducing agent based on the pattern density factor, and the rapid deposition stage and the densification stage perform thickness feedback control. The substrate to be plated is immersed in a chemical gold plating solution, and three stages are performed sequentially along the time axis. Feedforward compensation is introduced in the nucleation stage. The parameters of each stage are shown in Table 1. Table 1 Parameter Table for Each Stage

[0031] Nucleation stage feedforward logic: Based on the graph density factor obtained in step 1 The reducing agent supply for each sub-region during the nucleation stage is allocated according to the following formula: ,in, It is the feedforward gain coefficient of the pattern density in the nucleation stage, dimensionless, with a typical value of 0.6~0.8, used to adjust the compensation amplitude of the nucleating reducing agent in different pattern density regions; The basic replenishment flow rate of the single-channel reducing agent during the nucleation stage is measured in ml / min, with a typical value of 7 ml / min (corresponding to 70% of the basic concentration), matching the overall consumption rate during the nucleation stage.

[0032] The uneven initial nucleation rate caused by differences in pattern density is offset by differentiated supply. There is no thickness feedback during the nucleation stage; initial uniformity is ensured solely by feedforward.

[0033] The physical basis of the nucleation stage: Although the Ni / Pd substrate already possesses catalytic activity, the initial deposition rate of electroless gold plating directly affects the size distribution of gold grains and the bonding quality with the substrate. A high initial rate leads to rapid, uneven nucleation of large grains, accompanied by localized hydrogen evolution and the formation of porosity. Lowering the rate results in finer, more uniform grains, forming a smooth and dense initial capping layer.

[0034] The timing of switching between each stage: the switching between the nucleation stage and the rapid deposition stage is controlled by a preset fixed duration and does not depend on the coating thickness feedback; the switching from the rapid deposition stage to the densification stage is automatically determined by the controller based on the cumulative thickness feedback.

[0035] Step 3: Starting from the rapid deposition stage, the gold plating thickness of each sub-region is collected in real time at a fixed sampling period. The average value of multiple sampling points in each sub-region is used as the gold plating thickness. An immersion-type multi-channel XRF thickness measurement system is adopted. The hardware consists of a Si-PIN semiconductor detector array with 16-64 channels, each channel equipped with a beryllium window (thickness 25μm) and a tungsten collimator (aperture 1mm); the X-ray source is a miniature Rh target or Ag target X-ray tube (tube voltage 40kV, tube current 20μA), fixed on the side of the detector, and irradiating the measurement point at an incident angle of 45°.

[0036] Immersion operation: The detector array is fixed on a two-dimensional moving platform, 5-8 mm from the substrate surface, and immersed in the plating solution. A beryllium window prevents the plating solution from contacting the detector chip.

[0037] Measurement principle: Primary X-rays excite inner-shell electrons in gold atoms, generating AuLα characteristic fluorescence (energy 9.71 keV), which enters the detector through a beryllium window. The relationship between fluorescence count rate (CPS) and gold layer thickness is determined through a two-point linear calibration: before each process, measurements are taken in the plating bath using standard gold foils of known thicknesses (1 μm and 10 μm), and a calibration curve is fitted. .

[0038] Sub-region thickness sampling rules: Nine sampling points (3×3) are set within each control sub-region, evenly distributed across the entire sub-region; the integration time for a single point acquisition is ≤0.2s, and the total acquisition time for a single region is ≤1.8s. The arithmetic mean of the nine sampling points is used as the representative thickness of the sub-region to eliminate the influence of minor fluctuations within the sub-region on feedback control. For adaptively divided irregular sub-regions, sampling points are arranged using area weighting to ensure full sampling coverage.

[0039] Scanning method: The 2D moving platform traverses all sub-regions along a zigzag path at a speed of ≥50mm / s. One complete scan takes ≤5s, corresponding to a fixed sampling period of 5s.

[0040] Measurement accuracy: Within the thickness range of 1-10μm, the standard deviation is ≤0.12μm, and the measurement system meets the process control requirements relative to the target tolerance of ±0.25μm.

[0041] Maintenance instructions: If a trace amount of gold is deposited on the surface of the beryllium window, it will introduce a background shift. The solution is to wipe the beryllium window with dilute aqua regia every 10 process cycles and refit the calibration coefficients in the next calibration cycle.

[0042] Step 4: Based on the thickness of the gold plating layer, calculate the instantaneous thickness deviation of each sub-region; remember For the target gold layer thickness, The gold layer thickness measured in the i-th sub-region at time t ( (where k is the sampling period and k is the sampling sequence number).

[0043] Instantaneous thickness deviation: , This indicates that the current coating is too thin and a reducing agent needs to be added to accelerate deposition; This indicates that the current coating is too thick, and the reducing agent needs to be reduced to slow down deposition. This definition allows the deviation sign to directly correspond to the compensation direction, facilitating the implementation of control logic.

[0044] Step 5: Based on the graphic density factor and instantaneous thickness deviation, use the compensation model to generate the adjustment value of reducing agent replenishment for each sub-region; The compensation model will adjust the deviation. and graphic density factor Mapped to reducing agent replenishment flow rate adjustment value The formula is as follows: ,in, This is the basic replenishment flow rate for a single channel during the rapid deposition phase. This value is obtained by dividing the total global consumption by the number of sub-regions, matching the overall consumption rate of the process to ensure stable bulk concentration. This is the error gain coefficient. Set based on: target tolerance ±0.25. At this time, the maximum flow rate adjustment range is ±12.5%, which ensures both response speed and avoids overshoot. This value can be adjusted by the critical proportionality method to adapt to different plating solution systems.

[0045] The density sensitivity coefficient is dimensionless. It was calibrated offline: test substrates with different density gradients were prepared. The optimal compensation amplitude is measured under a fixed thickness deviation, and the regression slope is obtained.

[0046] Let be the graph density factor of subregion i, which is dimensionless.

[0047] As a reference density factor, all sub-regions are taken. The arithmetic mean of , dimensionless.

[0048] To supplement the lower limit of traffic, take 20% to avoid completely cutting off the supply, which would cause local concentrations to be too low and disrupt the diffusion field.

[0049] To supplement the traffic limit, take The concentration of the reducing agent is 200% of the base concentration to ensure that the peak concentration of the local reducing agent does not exceed 150% of the base concentration, thus avoiding increased self-decomposition of the reducing agent and excessive hydrogen evolution, which can lead to porosity in the coating.

[0050] The clip function is a limiting function that constrains the output between upper and lower limits, ensuring that local process parameters are always within the acceptable window.

[0051] This is a proportional error feedback term; if the thickness is too large, adjust it to be low; if the thickness is too small, adjust it to be high. As a feedforward term for graphics density, areas with high density consume more resources, thus requiring additional supply. The product of two terms retains the cross-coupling effect and has higher accuracy than first-order linear superposition.

[0052] The controller calculates each sub-region sequentially in each sampling period. The total computation time is ≤0.2ms, which meets the real-time requirements.

[0053] Step 6: Using a multi-channel liquid supply system, the adjustment value is first cross-corrected based on the cross-interference coupling coefficient of adjacent sub-regions, and then a reducing agent is quantitatively added to the near-wall boundary layer of the plating solution in each sub-region according to the corrected adjustment value. Hardware components: multi-channel micro-injection pump (such as a 48-channel push-pull injection pump), equipped with independent liquid supply lines (0.5mm inner diameter PTFE tubing) and micro nozzles corresponding to each sub-area (0.2mm outlet inner diameter, 316L stainless steel).

[0054] Nozzle arrangement: The nozzle is located directly above the corresponding sub-region, 1-2 mm from the substrate surface (in the near-wall area), and the liquid is discharged horizontally at a small tilt angle of 0°-5°, so that the replenishing liquid spreads slowly along the surface of the sub-region.

[0055] Replenishment solution formulation: A high-concentration reducing agent solution, with a concentration 3-5 times the base concentration (i.e., 1.5-2.5 g / L), dissolved in the same buffer system as the plating bath. A small volume of replenishment solution carries sufficient reducing agent without significantly diluting the local plating solution.

[0056] Injection rate: Outlet flow velocity ≤ 0.05 m / s, outlet Reynolds number about 14, which is far below the critical value for laminar-turbulent transition. The replenishing liquid flows out slowly in a laminar flow, without causing significant disturbance to the laminar flow state near the wall.

[0057] Execution mechanism: The controller sends pulse signals to the syringe pump, with the number of pulses corresponding to the replenished volume. In actual deployment, the volume of each injection step must be calibrated according to the syringe pump model and syringe specifications. The calibration method is to weigh the mass of pure water injected in multiple steps using an analytical balance, calculate the actual volume of each step, and record it in the controller parameter table.

[0058] The complete disturbance compensation and cross-interference correction scheme is presented in Example 6.

[0059] Step 7: Repeat steps 3 to 6 until the gold plating thickness of all sub-regions is within the allowable tolerance zone of the preset target thickness for m0 consecutive sampling periods, where m0 is the preset number of consecutive qualified judgment periods, which is a positive integer.

[0060] Loop termination condition: The thickness of all sub-regions satisfies the condition for m0 consecutive sampling periods. , (i.e. 5) (Target thickness ±5% tolerance zone). Setting three consecutive confirmations is used to filter out noise from single measurements and avoid spurious convergence.

[0061] Example 2: This invention proposes a reduction electroplating method to optimize the uniformity of thick gold layers on packaging substrates, which calculates the pattern density factor for each sub-region, including: Obtain the line area ratio of each control sub-region and calculate the initial plane density factor of each control sub-region; Meanwhile, based on the copper thickness data of the circuit of the substrate to be plated, a depth topography constraint map of each sub-region is constructed. The depth topography constraint map is used to characterize the range of reducing agent dissipation influence corresponding to the three-dimensional topography of the circuit and to delineate the effective area of ​​the sidewall effect. Within the effective area defined by the depth topography constraint map, the effective reaction area of ​​the sidewall is compensated for the corresponding initial planar density factor to generate a constraint correction density factor. Based on the local gradient change of the constraint correction density factor, each sub-region is adaptively divided into multiple mass transfer micro-domains, and the consumption weight within each mass transfer micro-domain is assigned according to the specific surface area of ​​the exposed lines in each micro-domain. The constraint correction density factor of each mass transfer micro-domain within the same sub-region is normalized and weighted by the corresponding consumption weight to obtain the graphic density factor of the corresponding sub-region.

[0062] In this embodiment, the initial planar density factor is: ,in, The sum of the projected areas of the top surfaces of all lines within sub-region i is obtained through polygon area integration in the DXF / GDSII file. It is a two-dimensional reference density that completely disregards the effects of sidewalls and linewidth.

[0063] Using copper thickness distribution data (spatial resolution 2 μm, measurement accuracy ±0.1 μm) obtained by white light interferometer measurement during the substrate pre-process (after copper electroplating), a digital elevation model of the circuit body is first constructed. Then, with the liquid phase plating solution region as the solution domain, the steady-state diffusion equation of the reducing agent is solved to obtain the scalar field of dissipation effect. Finally, the equipotential surface is extracted as a depth topography constraint map to delineate the effective range of the sidewall dissipation effect. The complete construction process is described in Example 3.

[0064] Within the effective area of ​​the sidewall effect defined by the depth topography constraint map, calculate the equivalent depositional area of ​​sub-region i. : ,in, The projected area of ​​the top surface of the line within the sub-region, in units of ; The total area of ​​the line sidewalls exposed to the plating solution within the sub-region is obtained by integrating the DEM data along all edge pixels. ,unit ; The sidewall deposition efficiency coefficient is dimensionless. , representing the average catalytic activity ratio of the sidewall surface to the top surface.

[0065] Calibration method: Take 10 isolated microstrip lines (30μm linewidth, 5mm length) with known copper thickness (15μm) and place them on the test substrate. Deposit to the target thickness of 5μm under standard process conditions. Cut each microstrip line cross-section using FIB (Film Injection Block), and measure the sidewall gold layer thickness and top surface gold layer thickness under SEM. Take the arithmetic mean of the thickness ratios of the 10 cross-sections to obtain the final thickness. Calibration results show that within the line width range of 10-100 μm... It is approximately a constant and has no significant correlation with linewidth.

[0066] It should be noted that if the sidewalls of the circuit are covered by photoresist or passivation layer, then the corresponding Do not enter Summation is achieved by reading the opening mask data in the design file during engineering implementation.

[0067] by Alternative The density factor is recalculated to obtain the constrained modified density factor: .

[0068] In this embodiment, the constraint correction density factor is used. This represents the overall average density of the sub-region, but local densities vary at different locations within the sub-region, with higher-density local hotspots exhibiting stronger energy consumption. To refine this distinction, mass transfer micro-domains are segmented: the local distribution of the constraint-corrected density factor within the sub-region is taken. (Using a grid with a sampling interval of 20 μm), calculate the local gradient on each pixel. In the neighborhood of gradient extrema, the SLIC superpixel segmentation algorithm (a pixel-based clustering-based region segmentation algorithm used to divide regions with varying density gradients into microdomains with uniform internal mass transfer properties) is employed. The superpixel compactness parameter m=20, and each microdomain contains approximately 50-100 pixels. Each segmented mass transfer microdomain... internal If the change is gradual (local CV < 3%), it can be approximated as having uniform mass transfer characteristics.

[0069] For each mass transfer micro-domain Calculate the specific surface area of ​​its internal circuitry. : ,in, For micro-domains The top surface area of ​​the internal wiring, ; For micro-domains Internal wiring sidewall area ; For micro-domains The corresponding near-wall plating solution volume, , For the horizontal projected area of ​​the micro-domain , The thickness is the region near the wall.

[0070] The larger the specific surface area, the larger the reaction area per unit volume of plating solution, and the faster the reducing agent is consumed, thus increasing the consumption weight. Pick Normalized value: .

[0071] The constraint-corrected density factors of each mass transfer micro-domain within the same sub-region are normalized and weighted summed with corresponding consumption weights to obtain the final graphical density factor of the sub-region: ,in, It is a micro-domain The constrained correction density factor local mean.

[0072] Example 3: This invention provides a reduction electroplating method for optimizing the uniformity of thick gold layers on a packaging substrate. Based on the copper thickness data of the circuit lines on the packaging substrate to be plated, a depth topography constraint map of each sub-region is constructed, including: Obtain the copper thickness distribution of the line within the corresponding sub-region, and construct a digital elevation model of the line body based on the plane coordinates of the sub-region and the copper thickness value of each point as the height value. On the digital elevation model, a gradient operator is used to extract the edge gradient vector field at the edge of the line, wherein the edge gradient vector field includes the maximum slope direction and the corresponding gradient magnitude of each edge point; Based on the edge gradient vector field, the gradient amplitude is used as the modulation coefficient and combined with the preset sidewall dissipation coefficient to construct the sidewall consumption flux boundary condition, where the consumption flux is the outflow flux of the liquid phase solution domain; the liquid phase plating solution region within the sub-region is used as the solution domain, the outer boundary of the sub-region is used as the first boundary of the plating solution body concentration, and the line sidewall is used as the second boundary of the consumption flux. The two-dimensional steady-state diffusion equation is solved to obtain the reducing agent concentration distribution field covering the entire liquid phase domain of the sub-region. Map the reducing agent concentration distribution field as a scalar field of dissipative effects; The equipotential surface of the dissipation influence scalar field is extracted, and the equipotential surface corresponding to the preset dissipation influence threshold is selected as the dissipation boundary surface. The dissipation boundary surface serves as the depth topography constraint map of the corresponding sub-region to characterize the effective boundary range of the reaction dissipation influence caused by the three-dimensional topography of the line.

[0073] In this embodiment, on the 500×500 pixel grid of sub-region i, the elevation value of pixel p is the copper thickness of the line at the corresponding location. The elevation of the area outside the line is 0, which constitutes the digital elevation model (DEM) of the line body.

[0074] The gradient in the x-direction of the DEM was calculated using the 3×3 Sobel operator. gradient in the y-direction Edge point determination criteria: (Elevation variations exceeding 1μm between adjacent pixels are considered valid edges). Record the gradient vectors of all valid edge points. With spatial coordinates.

[0075] In this embodiment, the sidewall consumption flux and diffusion equations are solved by: constructing sidewall consumption flux boundary conditions for all effective edge points in sub-region i. Calculate the normalized gradient magnitude: The denominator is the maximum gradient magnitude of all valid edge points within sub-region i. The normalized gradient magnitude reflects the relative copper thickness at the edge points. , This represents the maximum copper thickness within the sub-region.

[0076] The reducing agent consumption flux at the edge point is defined as the product of the normalized gradient magnitude and the sidewall dissipation coefficient. ,in, , representing the equivalent consumption rate of reducing agent per unit area of ​​sidewall under standard process conditions.

[0077] The liquid phase plating region within sub-region i is the planar region excluding the solid circuit body. The diffusion equation holds only in the liquid phase; there is no material diffusion inside the solid circuit.

[0078] Boundary conditions: Outer boundary: The first boundary is the Dirichlet fixed concentration boundary. That is, the outer boundary of the sub-region is equal to the concentration of the plating solution, and the concentration is kept constant by the macroscopic circulation of the plating solution. The reducing agent diffuses from the outer boundary to the inside to replenish the sidewall consumption. Inner boundary (line sidewall): The second boundary is the Neumann flux boundary. That is, the diffusion flux at the sidewall is equal to the consumption flux.

[0079] Solve the two-dimensional steady-state diffusion equation (Laplace equation): diffusion coefficient , is the effective diffusion coefficient of gold ions in the plating solution at 65℃ and pH=7.0, determined by a rotating disk electrode experiment.

[0080] The two-dimensional model is applicable when the circuit is elongated and the concentration gradient along its length is negligible; the two-dimensional planar model can accurately characterize the mass transfer properties. For circuits of arbitrary shape, the two-dimensional model is an engineering approximation with an error ≤10%, meeting the accuracy requirements for process compensation. If higher accuracy is required, it can be extended to a three-dimensional model, with the same calculation logic.

[0081] The finite element method was used for numerical solution. The mesh was divided into unstructured triangular meshes, and the mesh near the edge of the line was refined to a side length of 2μm. The steady-state concentration field C(x,y) of the reducing agent in the entire liquid phase was obtained. The high concentration at the outer boundary diffused inward, and the sidewalls continuously consumed the reducing agent, forming a stable concentration gradient distribution.

[0082] Map the concentration field to a scalar field of dissipative effects. The mapping rule is as follows: the intensity of dissipation effect is positively correlated with the degree of local consumption; the closer to the edge of the line, the more reducing agent is consumed, the lower the concentration, and the stronger the dissipation effect. Mapping formula: ,in, This represents the bulk concentration of the liquid phase (the concentration at the outer boundary of the sub-region). This represents the average concentration at all line edge points. This mapping ensures that at the line edges... The corresponding dissipation effect is strongest in the body region far from the edge. The corresponding dissipation effect is negligible.

[0083] Effects of dissipation on scalar fields Perform equipotential surface extraction. Select a preset threshold. As an effective demarcation value for dissipation effects: The region where dissipation effects are significant needs to be included in the sidewall effect compensation scope. In the region where dissipation effects are weak and sidewall effects are negligible.

[0084] Equipotential surfaces This is the dissipation boundary surface, which is also the depth topographic constraint map of the corresponding sub-region. This surface is a concentration contour line in the liquid phase, characterizing the effective boundary range of the reducing agent dissipation effect caused by the three-dimensional topography of the circuit.

[0085] Threshold selection instructions: 0.5 is taken as a compromise value in engineering. If the threshold is too high, the effective range will be too small, which will underestimate the sidewall consumption; if the threshold is too low, the effective range will be too large, which will introduce too many weakly influential areas. In actual production, this threshold can be iteratively optimized through calibration experiments, with the goal of achieving optimal uniformity.

[0086] Example 4: This invention provides a reduction electroplating method for optimizing the uniformity of a thick gold layer on a packaging substrate. The method involves obtaining surface circuit pattern data of the packaging substrate to be plated, and dividing the substrate surface into several control sub-regions, including: Line entity recognition is performed on the surface line graphic data, and geometric feature parameters of each line entity are extracted, wherein the geometric feature parameters include at least line width, line spacing and line centerline curvature; Using the geometric feature distribution clustering of line entities as the primary classification criterion, line entities with geometric feature similarity higher than a preset threshold are grouped into the same primary partition, generating several geometrically homogeneous regions. Within each geometrically homogeneous region, a spatial adjacency graph of the line entities is constructed. The spatial adjacency graph is constructed with each line entity as a node and the minimum distance between two line entities as an edge connection relationship based on whether it is less than a preset diffusion coupling threshold. Under the constraints of the spatial adjacency graph, a region partitioning cost function is constructed with the dual optimization objectives of local density equilibrium of line area and equilibrium of equivalent deposition rate distribution of line within the sub-region. The spatial adjacency graph is iteratively segmented using a graph cut algorithm. Each segmentation selects the edge set that maximizes the decrease in the region partitioning cost function for cutting, until the coefficient of variation of the local density of the line area and the coefficient of variation of the equivalent deposition rate distribution of the line in the spatial region corresponding to each segmented subgraph are reduced to below the preset equilibrium threshold, and the total number of control sub-regions obtained by segmentation does not exceed the preset maximum number of partitions. Each spatial region obtained from the final segmentation is taken as the control sub-region.

[0087] In this embodiment, the netlist is read from the design file (Gerber or ODB++ format), and all line segments or filled polygons of the same net number are merged into a single line entity. Geometric feature parameters are extracted: a vectorized central axis is extracted for each line entity, and the line width, line spacing, and centerline curvature are calculated point-by-point along the central axis. Functional attribute tags (power line, ground line, high-speed signal line, ordinary line) can be attached. These tags are only used for subsequent partitioning results and engineering alignment with functional modules; they do not participate in core partitioning determination. The chemical plating deposition rate is only related to the geometric morphology and local plating solution parameters, and has no causal relationship with the electrical function of the line.

[0088] Using the geometric feature distribution of line entities as the primary classification criterion, the DBSCAN density clustering algorithm is employed to group line entities with line width distribution and local density similarity exceeding a preset threshold into the same primary partition, generating several geometrically homogeneous regions. The overall morphological characteristics of the lines within each homogeneous region are similar, providing a foundation for subsequent fine-grained partitioning.

[0089] Construct an undirected graph within each geometrically homogeneous region. : Each node represents a line entity; If the minimum distance between the two line entities Diffusion Coupling Threshold If there is an edge connecting the two nodes, then there exists an edge connecting them.

[0090] Diffusion Coupling Threshold This threshold is determined by the thickness of the Nernst diffusion layer. Determine: Mass transfer coefficient (Measured by a rotating disk electrode at 65℃ and 200 rpm), calculated as follows The thickness of the diffusion layer is taken as twice the thickness and rounded up to 150 μm to ensure that the mass transfer coupling relationship of the adjacent lines is completely captured.

[0091] The core objective of region partitioning is to ensure that the coating consumption characteristics within each sub-region are as uniform as possible, i.e., uniform local density and uniform distribution of equivalent deposition rate. The detailed construction method of the cost function is explained in Example 5.

[0092] The Normalized Cut algorithm (a graph-theoretic optimal partitioning algorithm used to achieve balanced subregion partitioning under spatial adjacency graph constraints) is employed for iterative partitioning. The initial input is a complete spatial adjacency graph of a geometrically homogeneous region. Iterative process: Calculate the change in cost function for all current binary splitting methods. ; Choose to Cut the edge set with the largest descent. Evaluate the coefficient of variation of the weighted equivalent density within the corresponding region of each segmented submap. Coefficient of variation with equivalent deposition rate .

[0093] Termination condition: The condition is met simultaneously within each sub-region. and And the total number of partitions (Matching the maximum number of channels in a multi-channel liquid supply system).

[0094] If the number of partitions reaches If the CV requirement is still not met, a merging strategy is executed: select the sub-region with the largest current internal CV, merge it with the neighboring sub-region with the closest density factor, re-evaluate the CV, and continue until all sub-regions meet the conditions or reach the optimal approximate result.

[0095] Equal-area meshing is the ultimate simplification of this adaptive meshing: when the substrate circuitry is uniformly distributed and there is no obvious density clustering, the cut result naturally approaches equal-area rectangular meshing. The two methods represent different levels of precision in the same partitioning step; meshing can be used for simple graphics, while adaptive cut is used for complex graphics.

[0096] Example 5: This invention provides a reduction electroplating method for optimizing the uniformity of thick gold layers on packaging substrates, comprising constructing a region segmentation cost function, including: The line entities within the geometrically homogeneous region are classified into at least two line width levels according to their line width dimensions, and a corresponding area conversion weight is assigned to each line width level. The area conversion weight corresponding to narrow lines is greater than that corresponding to wide lines. For any candidate sub-region, the equivalent deposition area is obtained by multiplying the projected area of ​​each line entity by the area conversion weight of the corresponding line width level of the line entity, and the weighted equivalent density of the candidate sub-region is calculated based on the equivalent deposition area. The maximum value of the coefficient of variation of the weighted equivalent density within each sub-region is taken as the first cost term, and the maximum value of the coefficient of variation of the linewidth equivalent deposition rate within each sub-region is taken as the second cost term. The region partitioning cost function is constructed by the weighted sum of the first and second cost terms.

[0097] Electroless plating exhibits a significant edge effect: the smaller the linewidth, the higher the edge proportion, the hemispherical distribution of the diffusion field, the lower the mass transfer resistance, the higher the average deposition rate per unit area, and the faster the reducing agent is consumed; conversely, the larger the linewidth, the thicker the diffusion layer in the central region, the greater the mass transfer resistance, and the lower the deposition rate per unit area. This principle is common knowledge in the field of PCB surface treatment.

[0098] Based on this, the line entities within the geometrically homogeneous region are divided into two levels according to their line width, and different area conversion weights are assigned to them, as shown in Table 2: Table 2 Weight Table

[0099] The area conversion weight is calibrated by electrochemical limiting current density testing: the limiting current density at each linewidth is measured on a dedicated test substrate (containing isolated microstrip lines of different widths). With wide lines Using the normalized benchmark, the regression yields .

[0100] For any candidate sub-region S, calculate the equivalent depositional area: ,in, Let be the projected area of ​​the top surface of line entity j. The area conversion weight is used for the corresponding line width level.

[0101] Weighted equivalent density: ,in, The geometric area of ​​the candidate sub-region is represented by . The weighted equivalent density combines the area ratio with the linewidth edge effect, more closely reflecting the actual reducing agent consumption intensity.

[0102] The cost function consists of two weighted terms, both of which follow the worst-case subregion priority optimization principle to ensure that uniformity is achieved within all subregions: ,in, , The weights were optimized using a grid search to achieve the best partition uniformity on 50 test substrates with different patterns.

[0103] First cost item The maximum value of the weighted equivalent density variation coefficient within all current sub-regions, i.e. The coefficient of variation (CV) is the ratio of the standard deviation to the mean of the local density within a sub-region, measuring the density uniformity within a single sub-region. The maximum CV across all sub-regions is taken as the cost term, following the worst-case scenario optimization principle: if even one sub-region exhibits density non-uniformity, the partitioning fails to meet the target.

[0104] Second cost item : The maximum value of the coefficient of variation of the equivalent deposition rate within all current sub-regions, i.e. The equivalent deposition rate is calculated using the weights corresponding to the linewidth levels. This metric measures the uniformity of the deposition rate within a single sub-region. The maximum CV across all sub-regions is taken as the cost term, consistent with the logic of the first cost term, to jointly ensure uniform coating characteristics within each sub-region.

[0105] Example 6: This invention provides a reduction electroplating method for optimizing the uniformity of thick gold layers on packaging substrates, which uses a multi-channel liquid supply system to quantitatively replenish the reducing agent, including: Acquire the operation status monitoring data of the multi-channel liquid supply system during the current process cycle. The operation status monitoring data includes at least the opening and closing response delay of each channel nozzle, the amplitude of pipeline residual pressure fluctuation, and the near-wall temperature drift of the sub-region. Normalize the deviation between the current value of each disturbance in the operation status monitoring data and the historical benchmark value of the corresponding disturbance to construct a multi-source disturbance deviation vector. Based on the preset perturbation coupling coefficient matrix, the multi-source perturbation deviation vector is mapped to the hardware perturbation correction factor of each sub-region. At the same time, based on the preset adjacent sub-region cross-interference coupling coefficient matrix, the concentration contribution caused by the diffusion of the supplementary amount in adjacent sub-regions is calculated, and the cross-interference correction factor is generated. The adjustment value is multiplicatively corrected by the product of the hardware disturbance correction factor and the cross-interference correction factor to obtain the actual replenishment amount, and quantitative replenishment is performed according to the actual replenishment amount, wherein the actual replenishment amount is constrained between preset upper and lower limits of flow.

[0106] The following four types of real-time monitoring data are collected by sensor arrays distributed across the multi-channel liquid supply system, as shown in Table 3: Table 3 Monitoring Table

[0107] The normalized bias for each type of disturbance is calculated by subtracting the historical baseline value from the current value and then dividing by the maximum allowable range for that type: , , ,in: , is the average opening delay of the nozzle under standard driving pressure during factory testing; , is the steady-state value of pressure fluctuation when the new pipeline is accepted at the working temperature of the plating solution; The baseline value is 0℃ (no drift); , , =3℃.

[0108] each It is a dimensionless value that falls in the interval [-1, 1], and j1 = 1, 2, 3.

[0109] The perturbation coupling coefficient matrix C2 has a size of N×3 (N sub-regions, 3 types of perturbations), and its elements are... Characterize the coupling strength of the j2th disturbance to the supplementary flow in the i2th sub-region. Calibration method: Design an orthogonal experiment. Using nozzle opening / closing delay, pipeline residual pressure fluctuation, and near-wall temperature as three factors, with four levels for each factor (covering the normal operating range), experiments were conducted group by group on an offline test bench. Under the baseline conditions of a base plating solution temperature of 65℃ and a base flow rate of 10ml / min, the actual steady-state flow deviation of each sub-channel outlet was measured in each experimental group. With the normalized perturbation as the independent variable and the relative flow deviation as the dependent variable, the perturbation coupling coefficient matrix C2 was obtained through multiple linear regression fitting. Validation criterion: the coefficient of determination of the regression fit. Only after the calibration results are obtained can the product be put into production and use.

[0110] Due to macroscopic flow and molecular diffusion of the bath solution, the reducing agent added to a single sub-region will partially diffuse to adjacent sub-regions, causing cross-interference in the compensation amount. This solution uses a cross-interference coupling coefficient matrix for pre-correction: Coupling coefficient calibration: In the offline test tank, the plating solution temperature and main flow rate were kept consistent with the mass production process, and only the standard flow rate of reducing agent replenishment solution was injected into the i2th sub-region; at the center of the near-wall boundary layer of the adjacent sub-region, the steady-state concentration increase value 10s after injection was detected by a micro electrochemical sensor; the ratio of the concentration increase value of the adjacent sub-region to the concentration increase value of the injected sub-region center was used as the concentration contribution ratio. The calibration only considers the coupling effect of first-level adjacent sub-regions. The contribution of second-level and above coupling is less than 1% and can be ignored. It forms the cross-interference coupling coefficient matrix KJ. The matrix elements represent the contribution of the reducing agent supplemented in the row sub-region to the concentration of the column sub-region after it diffuses to the column sub-region.

[0111] Cross-correction calculation: Let the original adjusted value vector be Q, and the vector after cross-interference pre-correction be... Satisfying Q=KJ· Where KJ is the cross-interference coupling coefficient matrix, the pre-corrected supplementary vector is obtained by matrix inversion: This is to counteract the concentration superposition effect caused by cross-diffusion.

[0112] In practical applications, only the coupling of first-level adjacent sub-regions is considered, and the coupling of second-level and above can be ignored. The matrix is ​​a sparse matrix, and the computational cost is small.

[0113] Cross-interference correction factor: For the i2th sub-region, the cross-interference correction factor obtained after pre-correction by the cross-coupling matrix is ​​denoted as... , which represents the multiplicative correction ratio of cross-diffusion compensation.

[0114] Hardware disturbance correction factor: ,in, The final actual replenishment amount is a multiplicative correction of the original adjustment value by multiplying the cross-interference correction factor and the hardware disturbance correction factor, i.e.: In the formula, To compensate for the original adjustment values ​​output by the model, the clip function constrains the final output between the upper and lower limits of the flow rate, ensuring compliance of local process parameters. This is a hardware disturbance correction item.

[0115] For the irregular and varying-sized control sub-regions obtained through adaptive partitioning, the following matching method is adopted: The nozzle array maintains a fixed grid arrangement, and a single sub-region can be covered by one or more nozzles; If a sub-region contains N nozzles, then the total replenishment amount for that sub-region is evenly distributed among the corresponding N nozzles. The nozzles located at the boundaries of sub-regions are allocated flow based on the proportion of their coverage area in the adjacent sub-regions; The sum of the total flow rates of all nozzles equals the total global replenishment, ensuring a balanced concentration in the body.

[0116] Example 7: This invention proposes a reduction electroplating method to optimize the uniformity of thick gold layers on packaging substrates, constructing a multi-source disturbance deviation vector, including: The deviations between the current values ​​of each disturbance in the operation status monitoring data and the corresponding historical benchmark values ​​are mapped to the flow attenuation rate of the corresponding channel, the increase in pipeline flow resistance, and the cumulative near-wall temperature drift of each sub-region in the current process cycle through a preset empirical conversion relationship, and are used as the loss input parameters for the current cycle. The uniformity deviation of the gold plating thickness in the corresponding sub-region of multiple substrates that have completed plating under historical process cycles is recorded, as well as the flow rate attenuation rate, pipeline flow resistance increment, and near-wall temperature drift accumulation in the multiple completed plating cycles. Using the uniformity deviation record as the output reference, and taking the flow rate attenuation rate and pipeline flow resistance increment as the first input dimension, and the flow rate attenuation rate and near-wall temperature drift accumulation as the second input dimension, a first-order loss feature mapping table and a second-order coupling loss feature mapping table are established for each sub-region. The first-order loss feature mapping table records the first-order system deviation value of the supplementary amount corresponding to a single input dimension combination, and the second-order coupling loss feature mapping table records the coupling compensation term that exceeds the linear superposition of the corresponding first-order system deviation value under the cross combination of two input dimensions. Based on the loss input parameters of the current period, determine the current first-order loss input combination and second-order coupling loss input combination of each sub-region, and obtain the first-order system deviation value and second-order coupling compensation term from the first-order loss feature mapping table and the second-order coupling loss feature mapping table, respectively. The first-order system deviation value is added to the second-order coupling compensation term to obtain the composite system deviation value, and the deviation of each disturbance quantity is biased and corrected using the composite system deviation value to obtain the biased disturbance quantity deviation. The bias values ​​of each disturbance quantity after bias correction are combined to form the multi-source disturbance bias vector.

[0117] In this embodiment, historical data collection involves extracting data from 50-100 batches of historical production data that have already undergone plating. Uniformity deviation record: the range or standard deviation of the gold layer thickness after plating on each batch of substrates; Flow attenuation rate (Initial calibration flow - current calibration flow) / initial calibration flow × 100%, calibrated monthly; Increment of pipeline flow resistance : The difference between the nozzle inlet pressure and the factory acceptance benchmark value at a fixed pump speed, in units of Collected once a week; Near-wall temperature drift accumulation Within a single process cycle The integral value over time is stored and summarized for each batch.

[0118] The real-time monitoring quantity in Example 6 is a dynamic instantaneous value, while the lookup parameter in Example 7 is a periodic cumulative / steady-state value. The two are bridged by an empirical conversion relationship. All conversion relationships need to be calibrated for specific hardware systems through regression testing using more than 30 batches of historical data. They are only applicable to the normal wear and tear range of the hardware. If the wear and tear exceeds the range, recalibration or the addition of a direct detection unit is required, as shown in Table 4. Table 4 Lookup

[0119] High-precision alternative: If higher compensation accuracy is required, a steady-state flow sensor and resistance detection unit can be directly added to the pipeline to replace the empirical conversion formula, directly obtain the flow attenuation rate and pipeline resistance increment, and eliminate empirical fitting error.

[0120] First-order loss characteristic mapping table: with ( ) is the input dimension, where, Select 5 levels (0%-20%, step size 5%). Select 5th gear (0-5kPa) min / ml, step size 1kPa (min / ml), forming a 5×5 grid. For each grid level, select production batches that have historically met the uniformity standard (in-plane uniformity CV≤3%) and whose input parameters fall within that level. Calculate the relative deviation between the actual reducing agent replenishment amount and the theoretical consumption amount for that batch. Take the average of the deviations of all corresponding batches as the first-order systematic deviation value of the replenishment amount for that level and store it in the corresponding position of the mapping table.

[0121] Second-order coupling loss characteristic mapping table: with ( ) is the input dimension. Same as above, 5 gears Select 5 gears (0-500℃) s, step size 100℃ (s), forming a 5×5 grid. For each grid level, calculate the actual total deviation value under that input combination, subtract the sum of the deviation values ​​of the corresponding flow attenuation alone and the temperature drift alone, and the remaining nonlinear coupling deviation is the second-order coupling compensation term, which is stored in the corresponding position of the mapping table to correct the nonlinear superposition error under the combined effect of multiple factors.

[0122] The basis for distinguishing between first-order and second-order: Flow rate attenuation and pipeline resistance increment: Both are related to the mechanical characteristics of fluid transport, and the superposition of deviations is mainly additive, and is classified as first order. Flow rate attenuation and temperature drift accumulation: the former affects the accuracy of liquid supply, and the latter affects the reaction rate. The cross-effect is a coupling term and cannot be simply added together, so it is classified as second-order.

[0123] Obtain the first-order system deviation value from the two tables. and second-order coupling compensation term The sums are used to obtain the composite system deviation value. The unit of measurement is ml / min.

[0124] Composite system bias is used to correct the bias vector of multi-source disturbances. In engineering, a uniform distribution method is used to allocate the total bias to the three types of disturbances: ,in, The normalized baseline is 50% of the base flow rate, determined by the 95th percentile of the absolute value of the supply deviation of the historical 50 batches. The bias allocation weight represents the proportion of the composite system deviation distributed across the j1-th type of disturbance. (Default value) That is, evenly distributed.

[0125] Distribution rule explanation: Uniform distribution simplifies the engineering scheme. If the sensitivity differences of various disturbances are known, weights can be allocated according to the sensitivity ratio, resulting in higher correction accuracy. Uniform distribution does not change the magnitude of the final total correction amount, but only affects the numerical distribution of the intermediate vector.

[0126] Revised The combined vectors are used to calculate the hardware disturbance correction factor in Example 6.

[0127] Example 8: This invention provides a reduction electroplating system for optimizing the uniformity of thick gold layers on packaging substrates, such as... Figure 2 As shown, it includes: The factor acquisition module is used to acquire the surface circuit pattern data of the substrate to be plated and packaged, divide the substrate surface into several control sub-regions, and calculate the pattern density factor of each sub-region based on the proportion of the circuit area. The gold plating module is used to immerse the substrate to be plated into a chemical gold plating solution and perform a phased reduction electroplating, which includes at least a nucleation stage, a rapid deposition stage and a densification stage performed sequentially. Each stage corresponds to a different set of basic process parameters. In the nucleation stage, the reducing agent is fed forward in a differentiated manner based on the pattern density factor. In the rapid deposition stage and the densification stage, thickness feedback control is performed. The thickness acquisition module is used to collect the gold plating thickness of each sub-region in real time at a fixed sampling period starting from the rapid deposition stage. The average value of multiple sampling points in each sub-region is used as the gold plating thickness. The deviation calculation module is used to calculate the instantaneous thickness deviation of each sub-region based on the thickness of the gold plating layer; The adjustment determination module is used to generate adjustment values ​​for the amount of reducing agent replenishment for each sub-region based on the graphic density factor and instantaneous thickness deviation using a compensation model; The quantitative replenishment module is used to first cross-correct the adjustment value based on the cross-interference coupling coefficient of adjacent sub-regions through a multi-channel liquid supply system, and then quantitatively replenish the reducing agent to the near-wall boundary layer of the plating solution in each sub-region according to the corrected adjustment value. The cyclic adjustment module is used to repeatedly execute the thickness acquisition module, deviation calculation module, adjustment determination module, and quantitative supplementation module until the gold plating thickness of all sub-regions is within the allowable tolerance zone of the preset target thickness within m0 consecutive sampling cycles.

[0128] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A reduction electroplating method for optimizing the uniformity of a thick gold layer on a packaging substrate, characterized in that, include: Step 1: Obtain the surface circuit pattern data of the substrate to be plated, divide the substrate surface into several control sub-regions, and calculate the pattern density factor of each sub-region based on the circuit area ratio. Step 2: Immerse the substrate to be plated into the electroless gold plating solution and perform a phased reduction electroplating, which includes at least a nucleation stage, a rapid deposition stage and a densification stage performed sequentially. Each stage corresponds to a different set of basic process parameters. The nucleation stage performs differentiated feedforward supply of reducing agent based on the pattern density factor, and the rapid deposition stage and the densification stage perform thickness feedback control. Step 3: Starting from the rapid deposition stage, the gold plating thickness of each sub-region is collected in real time at a fixed sampling period. The average value of multiple sampling points in each sub-region is used as the gold plating thickness. Step 4: Based on the thickness of the gold plating layer, calculate the instantaneous thickness deviation of each sub-region; Step 5: Based on the graphic density factor and instantaneous thickness deviation, use the compensation model to generate the adjustment value of reducing agent replenishment for each sub-region; Step 6: Using a multi-channel liquid supply system, the adjustment value is first cross-corrected based on the cross-interference coupling coefficient of adjacent sub-regions, and then a reducing agent is quantitatively added to the near-wall boundary layer of the plating solution in each sub-region according to the corrected adjustment value. Step 7: Repeat steps 3 to 6 until the gold plating thickness of all sub-regions is within the allowable tolerance zone of the preset target thickness for m0 consecutive sampling periods.

2. The reduction electroplating method according to claim 1, characterized in that, Calculate the graphic density factor for each sub-region, including: Obtain the line area ratio of each control sub-region and calculate the initial plane density factor of each control sub-region; Meanwhile, based on the copper thickness data of the circuit of the substrate to be plated, a depth topography constraint map of each sub-region is constructed. The depth topography constraint map is used to characterize the range of reducing agent dissipation influence corresponding to the three-dimensional topography of the circuit and to delineate the effective area of ​​the sidewall effect. Within the effective area defined by the depth topography constraint map, the effective reaction area of ​​the sidewall is compensated for the corresponding initial planar density factor to generate a constraint correction density factor. Based on the local gradient change of the constraint correction density factor, each sub-region is adaptively divided into multiple mass transfer micro-domains, and the consumption weight within each mass transfer micro-domain is assigned according to the specific surface area of ​​the exposed lines in each micro-domain. The constraint correction density factor of each mass transfer micro-domain within the same sub-region is normalized and weighted by the corresponding consumption weight to obtain the graphic density factor of the corresponding sub-region.

3. The reduction electroplating method according to claim 2, characterized in that, Based on the copper thickness data of the circuits on the substrate to be plated, a depth topography constraint map of each sub-region is constructed, including: Obtain the copper thickness distribution of the line within the corresponding sub-region, and construct a digital elevation model of the line body based on the plane coordinates of the sub-region and the copper thickness value of each point as the height value. On the digital elevation model, a gradient operator is used to extract the edge gradient vector field at the edge of the line, wherein the edge gradient vector field includes the maximum slope direction and the corresponding gradient magnitude of each edge point; Based on the edge gradient vector field, the gradient amplitude is used as the modulation coefficient and combined with the preset sidewall dissipation coefficient to construct the sidewall consumption flux boundary condition, where the consumption flux is the outflow flux of the liquid phase solution domain; the liquid phase plating solution region within the sub-region is used as the solution domain, the outer boundary of the sub-region is used as the first boundary of the plating solution body concentration, and the line sidewall is used as the second boundary of the consumption flux. The two-dimensional steady-state diffusion equation is solved to obtain the reducing agent concentration distribution field covering the entire liquid phase domain of the sub-region. Map the reducing agent concentration distribution field as a scalar field of dissipative effects; The equipotential surface of the dissipation influence scalar field is extracted, and the equipotential surface corresponding to the preset dissipation influence threshold is selected as the dissipation boundary surface. The dissipation boundary surface serves as the depth topography constraint map of the corresponding sub-region to characterize the effective boundary range of the reaction dissipation influence caused by the three-dimensional topography of the line.

4. The reduction electroplating method according to claim 1, characterized in that, Obtain the surface circuit pattern data of the substrate to be plated, and divide the substrate surface into several control sub-regions, including: Line entity recognition is performed on the surface line graphic data, and geometric feature parameters of each line entity are extracted, wherein the geometric feature parameters include at least line width, line spacing and line centerline curvature; Using the geometric feature distribution clustering of line entities as the primary classification criterion, line entities with geometric feature similarity higher than a preset threshold are grouped into the same primary partition, generating several geometrically homogeneous regions. Within each geometrically homogeneous region, a spatial adjacency graph of the line entities is constructed, wherein the spatial adjacency graph is constructed with each line entity as a node and the minimum distance between two line entities as an edge connection relationship based on whether it is less than a preset diffusion coupling threshold. Under the constraints of the spatial adjacency graph, a region partitioning cost function is constructed with the dual optimization objectives of local density equilibrium of line area and equilibrium of equivalent deposition rate distribution of line within the sub-region. The spatial adjacency graph is iteratively segmented using a graph cut algorithm. Each segmentation selects the edge set that maximizes the decrease in the region partitioning cost function for cutting, until the coefficient of variation of the local density of the line area and the coefficient of variation of the equivalent deposition rate distribution of the line in the spatial region corresponding to each segmented subgraph are reduced to below the preset equilibrium threshold, and the total number of control sub-regions obtained by segmentation does not exceed the preset maximum number of partitions. Each spatial region obtained from the final segmentation is taken as the control sub-region.

5. The reduction electroplating method according to claim 4, characterized in that, Constructing the region partitioning cost function includes: The line entities within the geometrically homogeneous region are classified into at least two line width levels according to their line width dimensions, and a corresponding area conversion weight is assigned to each line width level. The area conversion weight corresponding to narrow lines is greater than that corresponding to wide lines. For any candidate sub-region, the equivalent deposition area is obtained by multiplying the projected area of ​​each line entity by the area conversion weight of the corresponding line width level of the line entity, and the weighted equivalent density of the candidate sub-region is calculated based on the equivalent deposition area. The maximum value of the coefficient of variation of the weighted equivalent density within each sub-region is taken as the first cost term, and the maximum value of the coefficient of variation of the linewidth equivalent deposition rate within each sub-region is taken as the second cost term. The region partitioning cost function is constructed by the weighted sum of the first and second cost terms.

6. The reduction electroplating method according to claim 1, characterized in that, Quantitative replenishment of reducing agent is performed through a multi-channel liquid supply system, including: Acquire the operation status monitoring data of the multi-channel liquid supply system during the current process cycle. The operation status monitoring data includes at least the opening and closing response delay of each channel nozzle, the amplitude of pipeline residual pressure fluctuation, and the near-wall temperature drift of the sub-region. Normalize the deviation between the current value of each disturbance in the operation status monitoring data and the historical benchmark value of the corresponding disturbance to construct a multi-source disturbance deviation vector. Based on the preset perturbation coupling coefficient matrix, the multi-source perturbation deviation vector is mapped to the hardware perturbation correction factor of each sub-region. At the same time, based on the preset adjacent sub-region cross-interference coupling coefficient matrix, the concentration contribution caused by the diffusion of the supplementary amount in adjacent sub-regions is calculated, and the cross-interference correction factor is generated. The adjustment value is multiplicatively corrected by the product of the hardware disturbance correction factor and the cross-interference correction factor to obtain the actual replenishment amount, and quantitative replenishment is performed according to the actual replenishment amount, wherein the actual replenishment amount is constrained between preset upper and lower limits of flow.

7. The reduction electroplating method according to claim 6, characterized in that, Constructing a multi-source perturbation bias vector includes: The deviations between the current values ​​of each disturbance in the operation status monitoring data and the corresponding historical benchmark values ​​are mapped to the flow attenuation rate of the corresponding channel, the increase in pipeline flow resistance, and the cumulative near-wall temperature drift of each sub-region in the current process cycle through a preset empirical conversion relationship, and are used as the loss input parameters for the current cycle. The uniformity deviation of the gold plating thickness in the corresponding sub-region of multiple substrates that have completed plating under historical process cycles is recorded, as well as the flow rate attenuation rate, pipeline flow resistance increment, and near-wall temperature drift accumulation in the multiple completed plating cycles. Using the uniformity deviation record as the output reference, and taking the flow rate attenuation rate and pipeline flow resistance increment as the first input dimension, and the flow rate attenuation rate and near-wall temperature drift accumulation as the second input dimension, a first-order loss feature mapping table and a second-order coupling loss feature mapping table are established for each sub-region. The first-order loss feature mapping table records the first-order system deviation value of the supplementary amount corresponding to a single input dimension combination, and the second-order coupling loss feature mapping table records the coupling compensation term that exceeds the linear superposition of the corresponding first-order system deviation value under the cross combination of two input dimensions. Based on the loss input parameters of the current period, determine the current first-order loss input combination and second-order coupling loss input combination of each sub-region, and obtain the first-order system deviation value and second-order coupling compensation term from the first-order loss feature mapping table and the second-order coupling loss feature mapping table, respectively. The first-order system deviation value is added to the second-order coupling compensation term to obtain the composite system deviation value, and the deviation of each disturbance quantity is biased and corrected using the composite system deviation value to obtain the biased disturbance quantity deviation. The bias values ​​of each disturbance quantity after bias correction are combined to form the multi-source disturbance bias vector.

8. A reduction electroplating system for optimizing the uniformity of a thick gold layer on a packaging substrate, characterized in that, include: The factor acquisition module is used to acquire the surface circuit pattern data of the substrate to be plated, divide the substrate surface into several control sub-regions, and calculate the pattern density factor of each sub-region based on the proportion of the circuit area. The gold plating module is used to immerse the substrate to be plated into a chemical gold plating solution and perform a phased reduction electroplating, which includes at least a nucleation stage, a rapid deposition stage and a densification stage performed sequentially. Each stage corresponds to a different set of basic process parameters. In the nucleation stage, the reducing agent is fed forward in a differentiated manner based on the pattern density factor. In the rapid deposition stage and the densification stage, thickness feedback control is performed. The thickness acquisition module is used to collect the gold plating thickness of each sub-region in real time at a fixed sampling period starting from the rapid deposition stage. The average value of multiple sampling points in each sub-region is used as the gold plating thickness. The deviation calculation module is used to calculate the instantaneous thickness deviation of each sub-region based on the thickness of the gold plating layer; The adjustment determination module is used to generate adjustment values ​​for the amount of reducing agent replenishment for each sub-region based on the graphic density factor and instantaneous thickness deviation using a compensation model; The quantitative replenishment module is used to first cross-correct the adjustment value based on the cross-interference coupling coefficient of adjacent sub-regions through a multi-channel liquid supply system, and then quantitatively replenish the reducing agent to the near-wall boundary layer of the plating solution in each sub-region according to the corrected adjustment value. The cyclic adjustment module is used to repeatedly execute the thickness acquisition module, deviation calculation module, adjustment determination module, and quantitative supplementation module until the gold plating thickness of all sub-regions is within the allowable tolerance zone of the preset target thickness within m0 consecutive sampling cycles.

Images