Copper foil resistivity data analysis system for fusion surface processing processes

Abstract

This invention belongs to the field of industrial data processing and electrical performance analysis technology, specifically relating to a copper foil resistivity data analysis system integrating surface treatment processes. The system includes a data input and processing unit, a length coordinate mapping unit, a substrate contribution estimation unit, a stage-gated decomposition unit, and a residual write-back unit. It receives copper foil roll number, specification information, resistivity measurement point data, and surface treatment process stage records. Based on linear velocity, stage dwell time, and process time, it generates a process fingerprint sequence corresponding to the measurement points. Combining this with stable batches of the same specification, adjacent sections of the same roll, and substrate process records, it obtains the substrate resistivity contribution value and decomposes the resistivity offset contribution value of each surface treatment stage. Then, based on the measured resistivity, it forms abnormal residuals and writes back the process fingerprint weights. This invention can distinguish between substrate differences and surface treatment disturbances under the same length coordinate, forming traceable analysis results of abnormal sections, dominant stages, and residual sources.

Description

Technical Field

[0001] This invention belongs to the field of industrial data processing and electrical performance analysis technology, specifically relating to a copper foil resistivity data analysis system that integrates surface treatment processes. Background Technology

[0002] After the copper foil substrate is formed, it typically undergoes surface treatment processes such as roughening, curing, anti-oxidation, and passivation to achieve a surface condition that meets the requirements of subsequent applications. Existing production data systems generally record the surface treatment process based on roll number, specifications, production time, processing section name, and set parameters. Process parameters are saved in chronological order according to equipment time, while resistivity test data is saved according to test batch, measurement point location, or sampling order. These two types of data are stored in separate process and test databases. In routine analysis, technicians usually first retrieve the surface treatment records based on the roll number or batch number, then retrieve the corresponding resistivity results, and judge the electrical performance status of the copper foil using batch average, maximum, minimum values, number of out-of-limit points, and trend curves. This method can achieve basic traceability, but the copper foil moves continuously on the surface treatment line, and the entry time, exit time, and residence time of different length sections of the same roll are not exactly the same in each processing stage. Furthermore, the resistivity measurement points formed at the testing end are often recorded later than those at the processing end. Surface treatment records are based on time, while resistivity test results are based on length measurement points, resulting in a natural difference in their coordinate systems. Simply associating with roll number, batch number, or testing time can only establish a coarse-grained correspondence. It is difficult to reflect the actual process trajectory of a certain resistivity measurement point in multiple surface treatment stages, and it is also difficult to determine whether the difference between adjacent measurement points is caused by changes in the length and position of the roll material or by changes in the state of the processing stage.

[0003] Existing, relatively close computer data analysis solutions involve statistically correlating copper foil resistivity test values ​​with surface treatment process settings to generate batch-level or time-window-level process quality analysis results. These solutions typically convert records such as linear velocity, processing time, processing solution state, current density, and processing temperature into time series, and then perform regression, correlation calculations, anomaly rule matching, or trend comparisons with the resistivity mean, dispersion, or sampled measurement points of the same batch. When resistivity shifts, the system provides an anomaly alert based on process setting deviations or equipment record fluctuations within the same time window. This analysis path assumes a direct correspondence between process records and test results in the time dimension, and also assumes that resistivity changes are mainly caused by parameter changes within the current processing window. However, there is a length transfer, stage dwell, and detection lag between the copper foil entering a certain surface treatment stage and the formation of test results. Furthermore, there is sequential coupling between roughening, curing, anti-oxidation, and passivation processes. Surface state changes caused by one stage may be amplified, weakened, or masked by subsequent stages. Ordinary statistical correlation can only provide the correlation between parameters and results, but it cannot distinguish different process experiences under the same resistivity shift, nor can it separate matrix differences, stage processing contributions, and detection residuals into calculable independent components. When the number of outlier samples is small or there are multiple stages of linked changes within the same process window, the statistical results are also prone to attribution drift.

[0004] The main technical problem arising from this is that existing copper foil resistivity data analysis systems struggle to establish a coordinate correspondence between surface treatment process stage records and resistivity measurement points along the length direction. This makes it impossible to separate the sources of resistivity anomalies between differences in the substrate copper foil and disturbances from the surface treatment stage. Specifically, substrate thickness, previous substrate process conditions, batch stability of the same specification, and continuity of the same roll length all affect the resistivity benchmark. Processing stages such as roughening, curing, anti-oxidation, and passivation have a cumulative effect on resistivity shifts. Without length coordinate mapping based on linear velocity, stage dwell time, and process timing, as well as mechanisms for substrate contribution estimation, stage gating decomposition, and residual write-back, the system can only obtain the final resistivity value, batch anomaly label, or rough trend judgment. It cannot determine the dominant processing stage corresponding to a certain abnormal length range, cannot distinguish between substrate contribution items and surface treatment-induced contribution items, and cannot use the residual results in reverse to correct similar process fingerprints. The same abnormal value may correspond to different combinations of processing stages, and fluctuations in the same processing stage may also manifest as different resistivity shifts due to different substrate conditions. Traditional result discrimination is difficult to form a closed calculation relationship between process stages, substrate contributions, and abnormal residuals, and therefore cannot support resistivity decomposition analysis for surface treatment processes. As a result, the attribution of anomalies still relies on human experience and post-event comparison. Summary of the Invention

[0005] The purpose of this invention is to provide a copper foil resistivity data analysis system that integrates surface treatment processes, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A copper foil resistivity data analysis system integrating surface treatment processes includes a data access and processing unit, a length coordinate mapping unit, a substrate contribution estimation unit, a stage-gated decomposition unit, and a residual write-back unit. The data access and processing unit receives copper foil roll number, specification information, resistivity measurement point data along the length direction, and surface treatment process stage records. The length coordinate mapping unit converts the surface treatment process stage records into a process fingerprint sequence corresponding to the resistivity measurement points along the length direction based on linear velocity, stage dwell time, and process time. The substrate contribution estimation unit generates substrate resistivity contribution values ​​based on stable batches of the same specification, adjacent sections of the same roll, and substrate process records. The stage-gated decomposition unit outputs resistivity offset contribution values ​​corresponding to each surface treatment stage based on the process fingerprint sequence. The residual write-back unit generates abnormal residuals based on measured resistivity, substrate resistivity contribution values, and resistivity offset contribution values, and writes back the process fingerprint weights.

[0007] Preferably, the length coordinate mapping unit includes a stage trajectory resampling processing unit and a measurement point assignment processing unit; the stage trajectory resampling processing unit divides the roughening, curing, anti-oxidation and passivation process stage records into continuous length segments according to the copper foil running direction, and configures the stage category, entry time, exit time, dwell time and process setting change trajectory for each continuous length segment; the measurement point assignment processing unit uses the length coordinate of the resistivity measurement point as an index to retrieve the continuous length segment corresponding to the measurement point in each surface treatment stage, and splices the multi-stage segments into the process fingerprint sequence according to the actual passing order.

[0008] Preferably, the substrate contribution estimation unit includes a same-specification benchmark screening processing unit, an in-roll continuous reference processing unit, and a substrate fusion processing unit; the same-specification benchmark screening processing unit extracts batch segments with consistent specifications and stable surface treatment states from historical data to form a specification benchmark sample set; the in-roll continuous reference processing unit extracts reference segments in the same roll of copper foil that have not experienced abrupt changes in the process stage according to the length adjacency relationship; the substrate fusion processing unit generates a substrate resistivity contribution value corresponding to each resistivity measurement point based on the specification benchmark sample set, reference segments, and substrate process records.

[0009] Preferably, the stage gating decomposition unit includes a stage encoding processing unit, a gating allocation processing unit, and a contribution constraint processing unit; the stage encoding processing unit segments and encodes the process category, dwell interval, and set change trajectory of each surface treatment stage in the process fingerprint sequence; the gating allocation processing unit configures learnable gating weights for different stage codes according to the stage sequence and outputs the resistivity offset contribution value corresponding to each stage; the contribution constraint processing unit adds length direction continuity constraints, stage sequence constraints, and batch consistency constraints of the same specification during the gating weight update process.

[0010] Preferably, the stage trajectory resampling processing unit is equipped with a time-series breakpoint correction processing flow. The time-series breakpoint correction processing flow identifies length segment breakpoints based on the linear velocity changes, stage entry time, and stage exit time of adjacent process records, generates candidate compensation intervals for cross-stage overlapping segments and missing segments, and determines the corrected continuous length segments based on the continuity of dwell time of adjacent stages and the increasing relationship of copper foil length. The measurement point attribution processing unit uses the corrected continuous length segments as the basis for generating the process fingerprint sequence.

[0011] Preferably, the substrate fusion processing unit includes a reference difference elimination processing flow. The reference difference elimination processing flow maps the resistivity distribution in the same specification reference sample set and the resistivity distribution in the same roll reference section to a unified length coordinate scale, extracts the substrate bias items corresponding to the thickness specification, length position and previous substrate process records, and performs joint weighting on the substrate bias items according to the same specification batch stability and in-roll continuity to generate the substrate resistivity contribution value after excluding the disturbance of the surface treatment stage.

[0012] Preferably, the contribution constraint processing unit includes a stage contribution separable processing flow. The stage contribution separable processing flow establishes a stage contribution matrix based on the multi-stage encoding corresponding to the same resistivity measurement point, and uses the stage contribution difference between adjacent length measurement points, the cross-batch contribution difference within the same stage, and the correlation between contributions of different stages as constraint terms. When updating the gating weight, it restricts length mutation, batch drift, and stage aliasing respectively. The gating allocation processing unit outputs the resistivity offset contribution value of each surface treatment stage according to the restricted gating weight.

[0013] Preferably, the stage coding processing unit includes an abnormal sample enhancement processing flow, which extracts comparison segments with similar process fingerprint sequences but different resistivity offset directions from historical batches, pairs the stage codes of the comparison segments with the corresponding substrate resistivity contribution values, and forms a control sample group in the stage gating decomposition training; the gating allocation processing unit distinguishes the resistivity offset contribution values ​​caused by the surface treatment stage and the resistivity offset contribution values ​​caused by substrate differences based on the control sample group.

[0014] Preferably, the residual write-back unit includes an abnormal segment aggregation processing flow. The abnormal segment aggregation processing flow differs the measured resistivity with the matrix resistivity contribution value and the resistivity offset contribution value of each stage to obtain a length measurement point residual sequence. Adjacent length measurement points are aggregated into abnormal length intervals according to the continuity of residual sign, similarity of residual amplitude, and similarity of process fingerprint. The abnormal length interval is associated with the dominant stage code in the stage contribution matrix and generates residual samples for updating the process fingerprint weights.

[0015] Preferably, the residual write-back unit includes a weight iterative update process, which performs a joint indexing of the abnormal length interval, dominant stage code, and process fingerprint sequence in the residual sample, extracts the stage residual pattern that appears repeatedly in multiple batches, and writes the stage residual pattern into the process fingerprint weight update set; the stage gating decomposition unit calls the process fingerprint weight update set during subsequent batch processing to pre-correct the gating weights corresponding to similar process fingerprint sequences.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention establishes a correspondence between surface treatment process records and resistivity measurement points along the length of copper foil in the same coordinate system through collaborative processing of data access, length coordinate mapping, substrate contribution estimation, stage-gated decomposition, and residual write-back. After linear velocity, stage dwell time, and process time are converted into a process fingerprint sequence oriented towards length segments, each resistivity measurement point can be associated with its actual coarsening, curing, anti-oxidation, and passivation stages, avoiding misalignment caused by rough matching based on batch time or detection time. Substrate contribution estimation uses stable batches of the same specification, adjacent sections of the same roll, and substrate process records to generate substrate resistivity contribution values. This allows resistivity analysis to first isolate the baseline differences caused by specifications, continuous states within the roll, and previous substrate processes, and then calculate the resistivity offset contribution values ​​corresponding to each surface treatment stage through stage-gated decomposition. After the measured resistivity, matrix resistivity contribution value, and resistivity offset contribution value at each stage are jointly used in the residual calculation, the abnormal results are transformed from simple out-of-limit markings into decomposed results including length intervals, dominant stages, and residual sources. Resistivity anomalies can form a clear attribution relationship between matrix differences and surface treatment disturbances. Length coordinate mapping provides the basis for the correspondence between measurement points and stage processes, matrix contribution estimation provides the offset stripping benchmark, stage-gated decomposition provides stage-level contribution output, and residual write-back provides the correction basis for similar process fingerprints. Each processing link is mutually restrictive, so that process data, resistivity data, and historical residual data no longer exist in isolation, but jointly form a closed data chain from processing to electrical results to weight updates. For resistivity fluctuations in adjacent length segments in the same roll, the system can distinguish between intra-roll benchmark drift and surface treatment-induced offset based on continuous length segments and stage contribution values. For similar resistivity anomalies in different batches, the system can maintain consistency in attribution caliber by using stable batches of the same specification and historical residual records, and reduce stage misjudgments caused by accidental fluctuations in a single batch, so that the anomaly source decomposition results are consistent with the actual path of surface treatment.

[0017] 2. This invention forms a continuous data constraint chain through stage trajectory resampling, temporal breakpoint correction, benchmark difference elimination, separable stage contribution processing, anomalous sample enhancement, and weight iterative updates. Temporal breakpoint correction corrects overlapping and missing segments across stages based on linear velocity changes, stage entry times, and stage exit times, ensuring the process fingerprint sequence maintains an increasing length relationship and stage dwell continuity, reducing measurement point assignment bias caused by recording breakpoints. Benchmark difference elimination maps benchmark samples of the same specification and reference segments within the same volume to a unified length coordinate scale, and generates a matrix bias term based on batch stability and intra-volume continuity, reducing matrix factor contamination in surface treatment contribution calculation. Separable stage contribution processing uses the contribution difference between adjacent length measurement points, the contribution difference across batches within the same stage, and the correlation of contributions from different stages to limit length abrupt changes, batch drift, and stage aliasing. Anomalous sample enhancement corrects stage encoding through comparison segments with similar process fingerprints but different resistivity offset directions. Residual write-back uses a joint index of the anomaly length range, dominant stage encoding, and process fingerprint sequence to write recurring stage residual patterns from multiple batches into the process fingerprint weight update set. The gating weights for subsequent similar process trajectories can be pre-corrected based on historical residuals. This ensures consistent decomposition results for similar anomalies across different batches, allows the stage contribution matrix to gradually converge with actual resistivity residuals, and enables process fingerprint weights to absorb reusable residual information from subsequent detection results, rather than relying solely on initial training data. This constraint chain also maintains data consistency between measurement point attribution, matrix estimation, stage decomposition, and residual updates, preventing deviations in a single processing step from being amplified in subsequent calculations. It allows resistivity anomaly analysis results to be jointly corrected across three dimensions: copper foil length, processing stage, and historical batches, and maintains stable and reliable contribution boundaries between different stages. Residual samples can also continuously supplement the process fingerprint weights required for stage gating decomposition, providing a traceable data source for the analysis results of similar process fingerprints in subsequent batches. Attached Figure Description

[0018] Figure 1 This is a flowchart of the overall closed-loop process of the copper foil resistivity data analysis system integrating surface treatment technology of the present invention. Figure 2 This is a flowchart of the length coordinate mapping and matrix contribution estimation of the present invention; Figure 3 This is a flowchart of the stage-gated decomposition and residual write-back process of the present invention. Detailed Implementation

[0019] refer to Figure 1In one embodiment, the copper foil resistivity data analysis system integrating surface treatment processes comprises a data access and processing unit, a length coordinate mapping unit, a matrix contribution estimation unit, a stage-gated decomposition unit, and a residual write-back unit. The data access and processing unit receives copper foil roll number, specification information, resistivity measurement point data along the length direction, and surface treatment process stage records. These records include the process time, linear velocity, stage dwell time, stage category, and process setting change trajectory for each stage (roughing, curing, anti-oxidation, passivation). The length coordinate mapping unit converts the time-based surface treatment records into a process fingerprint sequence based on copper foil length, enabling the same resistivity measurement point to correspond to its actual experience in each surface treatment stage. The matrix contribution estimation unit utilizes stable batches of the same specification, adjacent sections of the same roll, and matrix process records. The system generates a matrix resistivity contribution value. The stage-gated decomposition unit outputs the resistivity offset contribution value corresponding to each surface treatment stage based on the process fingerprint sequence. The residual write-back unit generates abnormal residuals based on the measured resistivity, matrix resistivity contribution value, and resistivity offset contribution value, and writes back the process fingerprint weights. During system operation, the volume number is used as the data association entry, the length coordinate is used as the process alignment entry, the matrix contribution and stage contribution are used as the decomposition calculation entry, and the residual sample is used as the subsequent weight correction entry, forming a closed data chain from process data to resistivity result data and then to fingerprint weight update. The advantage of this embodiment is that it can transform the copper foil resistivity anomaly from a single result judgment into a decomposition analysis between matrix differences, surface treatment stage disturbances, and abnormal residuals, avoiding confusion of the source of anomalies caused by comparing the batch average or detection time with the process record.

[0020] In one embodiment, the data access processing unit performs unified identification, field cleaning, and time-series merging processing on the input data. The roll number is used to define consecutive production objects of the same copper foil, and the specification information is used to define thickness, width, and product category. The length-direction resistivity measurement point data includes the measurement point number, measurement point length coordinates, detection time, and measured resistivity. Surface treatment process stage records are stored according to the processing segment name, stage entry time, stage exit time, linear velocity record, stage dwell time, and set change trajectory. After receiving the data, the data access processing unit generates a composite key for each record consisting of the roll number, stage category, and recording time, and performs time-series merging on duplicate records under the same composite key. Furthermore, records lacking a length field but possessing linear velocity and time fields are marked as records to be mapped. Records lacking specification information but traceable by volume number are supplemented with specification fields from the main record of the same volume. Detection data exceeding the processing time range of the same volume are isolated and stored. The processed data is output to the length coordinate mapping unit and the matrix contribution estimation unit. The length coordinate mapping unit obtains process records with complete stage time and linear velocity sequences, and the matrix contribution estimation unit obtains detection records with specification consistency and intra-volume continuity. The advantage of this embodiment is that it provides a unified data reference basis for subsequent length mapping and contribution estimation, reducing calculation deviations caused by volume number mismatch, missing time, and duplicate records.

[0021] refer to Figure 2 In one embodiment, the length coordinate mapping unit generates length coordinates based on linear velocity, stage dwell time, and process time, and the calculation method uses the following formula:

[0022] in, This indicates the surface treatment stage number, with values ​​from 1 to 4, corresponding to roughening, curing, anti-oxidation, and passivation, respectively. This indicates the sequence number of the process record within the same stage. Indicates the first Phase 1 The converted length coordinates of each process record. Indicates the first Phase 1 The linear velocity corresponding to each sampling interval Indicates the time interval between adjacent process records. Indicates resistivity measurement point In the Process fingerprint elements formed in each stage Indicates the measuring point Entering the At this stage, Indicates the measuring point Leaving the At this stage, Indicates the measuring point In the The process setting change trajectory experienced in each stage; when the length coordinate of a certain stage is 120 meters, the linear velocity is 8 meters per minute, and the time interval is 0.5 minutes, the converted length coordinate is 124 meters. When the entry time of measuring point 8 in the coarsening stage is 10.0 minutes, the exit time is 12.5 minutes, and the dwell time is 2.5 minutes, the corresponding process fingerprint element record is stage number 1, entry time 10.0, exit time 12.5, dwell time 2.5, and the process setting change trajectory within this time period. The advantage of this embodiment is that it converts the surface treatment record originally stored in time order into a length coordinate record that can be directly indexed by resistivity measuring points, so that the process and the detection results have the same calculation scale.

[0023] Table 1. Configuration of Process Fingerprint Sequence Fields ; The fields shown in Table 1 are used to illustrate that the process fingerprint sequence is not a set of ordinary process parameters, but a record of corresponding relationships formed around the copper foil roll number, length measurement point, and surface treatment stage. The entry time, exit time, dwell time, and set change trajectory are simultaneously written into the stage fingerprint of the same measurement point, so that the subsequent matrix contribution estimation and stage gating decomposition can be calculated around the same measurement point. The advantage of this embodiment is that it enables the surface treatment stage data to have a traceable, segmentable, and splicable structured expression.

[0024] In one embodiment, the length coordinate mapping unit includes a stage trajectory resampling processing unit and a measurement point assignment processing unit. The stage trajectory resampling processing unit divides the roughening, curing, anti-oxidation, and passivation process stages into continuous length segments according to the copper foil running direction. The segmentation uses the stage category as the segment boundary, the stage entry and exit times as the segment time range, the linear velocity integral result as the segment length range, and the process setting change trajectory as the segment attribute. The measurement point assignment processing unit uses the length coordinates of the resistivity measurement point as an index to retrieve the corresponding continuous length segment of the measurement point in each surface treatment stage, and splices the multi-stage segments according to the actual passing order of roughening, curing, anti-oxidation, and passivation. When a measurement point corresponds to a length segment of 120 to 128 meters in the roughening stage... When the length segment corresponding to the solidification stage is 120.2 meters to 128.2 meters, the length segment corresponding to the anti-oxidation stage is 120.4 meters to 128.4 meters, and the length segment corresponding to the passivation stage is 120.5 meters to 128.5 meters, the measurement point assignment processing unit searches for the corresponding segment of each stage segment according to the measurement point length coordinates, and splices the search results into a process fingerprint sequence containing fingerprint elements of 4 stages. If there are multiple set change sub-segments in a certain stage, the measurement point assignment processing unit extracts the sub-segment trajectory according to the actual residence interval of the measurement point in that stage and merges it into the corresponding stage fingerprint element. The advantage of this embodiment is that a single resistivity measurement point no longer corresponds to only one batch record, but corresponds to a complete multi-stage processing trajectory, providing stage-level input for subsequent decomposition analysis.

[0025] In a preferred embodiment, the stage trajectory resampling processing unit includes a timing breakpoint correction process. After generating continuous length segments, this process detects the coordinate and temporal relationships between adjacent segments. When the endpoint of the length calculated from the linear velocity of an adjacent process record is greater than the starting point of the next record length, the overlapping portion is marked as a cross-stage overlapping segment. When the endpoint of the length of an adjacent process record is less than the starting point of the next record length, the missing portion is marked as a missing segment. For cross-stage overlapping segments, the timing breakpoint correction process determines the segment to be retained based on the continuity of the entry time, exit time, and dwell time of adjacent stages. For missing segments, the timing breakpoint correction process determines the segment to be retained based on the continuity of the entry time, exit time, and dwell time of adjacent stages. Candidate compensation intervals are generated based on the changing trends of adjacent linear velocities and the increasing relationship of copper foil length. The candidate compensation intervals are then compared with the set change trajectories of the preceding and following stages for continuity. Only when the compensation intervals meet the requirements of no inversion of stage time, increasing length coordinates, and no negative dwell time are they written into the corrected continuous length segment. The measurement point attribution processing unit uses the corrected continuous length segment as the basis for the measurement point index to avoid resistivity measurement points being incorrectly assigned to adjacent stages or omitted in the stage gaps. The advantage of this embodiment is that by correcting the timing breakpoints, the consistency of the process fingerprint sequence in the length direction and stage order is maintained, so that the contribution decomposition of subsequent stages will not cause attribution drift due to misalignment of the basic mapping.

[0026] In one embodiment, the substrate contribution estimation unit includes a same-specification benchmark screening processing unit, an in-roll continuous reference processing unit, and a substrate fusion processing unit. The same-specification benchmark screening processing unit extracts batch segments with consistent specifications and stable surface treatment states from historical data. Consistent specifications refer to the same thickness, width, and product category as the current copper foil. Stable surface treatment states mean that the set change trajectory of the surface treatment stage in the batch segment does not show abrupt changes and the resistivity distribution does not show continuous abnormal residuals. The in-roll continuous reference processing unit extracts reference segments in the same roll of copper foil that do not show abrupt changes in process stages according to length adjacency, and uses the resistivity distribution of the reference segments as the source of in-roll benchmark changes. The substrate fusion processing unit... The processing unit reads the matrix state data from the preceding matrix process record and maps the same specification benchmark sample set, the same roll reference section, and the matrix process record to the length coordinates of the current measurement point, generating the matrix resistivity contribution value corresponding to each resistivity measurement point. In this embodiment, the matrix contribution estimation does not directly replace the benchmark with the final resistivity result, but provides cross-batch benchmarks with stable batches of the same specification, provides intra-roll continuous benchmarks with adjacent sections of the same roll, and provides the matrix formation process benchmark with the preceding matrix process record. The advantage of this embodiment is that it can remove matrix differences before analyzing the contribution of the surface treatment stage, so that the resistivity shift under the same surface treatment trajectory will not be directly mixed into the stage contribution by matrix fluctuations.

[0027] In one embodiment, the substrate fusion processing unit generates the substrate resistivity contribution value using the following formula:

[0028] in, Indicates resistivity measurement point The corresponding matrix resistivity contribution value, This indicates that stable batches of the same specification are mapped to measurement points. Reference resistivity after length coordinates This indicates that adjacent reference sections in the same volume are mapped to measurement points. Reference resistivity after the length coordinate. This represents the estimated matrix resistivity obtained from the preceding matrix process records. Indicates the stability weight for the same specification. Indicates the intra-volume continuity weight. Indicates the weight of the integrity of the matrix process record; when 1.680, 1.700, For 1.690, 0.5 0.3 When it is 0.2, The calculated result is 1.688, and this result is used as the measurement point. The advantage of this embodiment, which does not consider the matrix contribution input during the surface treatment stage, is that by jointly weighting the matrix contribution values ​​from three sources—same specification, same roll, and previous matrix—it simultaneously reflects historical stability, intra-roll continuity, and the matrix formation process record.

[0029] In a preferred embodiment, the matrix fusion processing unit includes a reference difference elimination process. This process maps the resistivity distribution in a set of reference samples of the same specification to the resistivity distribution in a reference section of the same roll to a unified length coordinate scale. During mapping, adjacent windows are extracted centered on the length coordinate of the current measurement point, and historical data with different length sampling densities are interpolated and normalized to ensure that historical reference sections, reference sections of the same roll, and the current section have the same measurement point index. Subsequently, the thickness specification, length position, and matrix offset items corresponding to the previous matrix process record are extracted. The thickness specification offset item is used to reflect different specifications. The influence of the matrix on the resistivity reference is reflected by the length position bias term, which is used to reflect the continuous drift of the beginning and end sections of the same roll, and the previous matrix process bias term, which is used to reflect the influence of the matrix formation state on the resistivity reference. The reference difference elimination process jointly weights the above matrix bias terms according to the batch stability and roll continuity of the same specification, and outputs the weighted result separately from the measured input of the resistivity measurement point as the matrix contribution part. The advantage of this embodiment is that the matrix difference enters the subsequent decomposition process as a calculable bias term, avoiding the stage gated decomposition from misassigning the differences from the matrix source to the roughening, curing, anti-oxidation or passivation stages.

[0030] refer to Figure 3 In one embodiment, the stage gating decomposition unit includes a stage encoding processing unit, a gating allocation processing unit, and a contribution constraint processing unit. The stage encoding processing unit segments and encodes the process category, residence interval, and set change trajectory of each surface treatment stage in the process fingerprint sequence. During encoding, the stage sequence, stage residence time, set change direction within the stage, and change segment length are retained. The gating allocation processing unit configures learnable gating weights for different stage codes according to the stage sequence and outputs the resistivity offset contribution value corresponding to each stage according to each stage code. The contribution constraint processing unit adds length direction continuity constraints, stage sequence constraints, and same-specification batch consistency constraints during the gating weight update process. The length direction continuity constraints are used to limit discontinuous contribution jumps between adjacent measurement points under the condition of no stage abrupt change. The stage sequence constraints are used to limit the contribution of subsequent stages from being incorrectly pre-allocated to the previous stage. The same-specification batch consistency constraints are used to limit the contribution decomposition results of similar process fingerprints in the same specification stable batch from drifting without basis. The advantage of this embodiment is that the contribution components of the coarsening, curing, anti-oxidation, and passivation stages are distinguished and constrained in the same model, forming a stage-level contribution output that can explain the source of resistivity offset.

[0031] In one embodiment, the stage-gated decomposition unit calculates the predicted resistivity and stage-gated weights using the following formula:

[0032] in, Indicates resistivity measurement point The predicted resistivity, Indicates resistivity measurement point The contribution value of matrix resistivity, Indicates the measuring point In the The candidate resistivity offset contribution value formed under stage coding. Indicates the measuring point In the Stage gating weights Indicates the measuring point In the Phase-based gating scoring This represents the stage index in the normalization calculation. Indicates exponentiation; when It is 1.688, with 4 stages. The values ​​are 0.006, -0.002, 0.004, and 0.001 in four stages, respectively. The normalized values ​​obtained when the values ​​are 0.2, 1.0, 0.3, and -0.5 respectively are... The values ​​are approximately 0.207, 0.461, 0.229, and 0.103, respectively, and the weighted sum of the stage contributions is approximately 0.0013. The value is approximately 1.6893. The advantage of this embodiment is that it uses the matrix contribution value as the calculation base and uses stage gating weights to distinguish the contribution ratio of each surface treatment stage to the resistivity shift, so that the model output can correspond to the specific processing stage.

[0033] In a preferred embodiment, the contribution constraint processing unit includes a stage contribution separable processing flow. This flow establishes a stage contribution matrix based on the multi-stage encoding corresponding to the same resistivity measurement point. The rows of the matrix correspond to resistivity measurement points along the length direction, and the columns correspond to coarsening, curing, anti-oxidation, and passivation stages. Matrix elements correspond to the resistivity offset contribution value of a measurement point at a certain stage. The contribution constraint processing unit calculates the stage contribution difference between adjacent length measurement points, the cross-batch contribution difference within the same stage, and the correlation between contributions from different stages. During gating weight updates, it restricts length abrupt changes, batch drift, and stage aliasing, respectively. The limitation is based on the similarity of process fingerprints between adjacent measurement points. When adjacent measurement points experience the same stage trajectory, their stage contribution difference is constrained to the allowable range of the current roll residual distribution. The batch drift limitation takes the stable batch of the same specification as a reference. When the same stage fingerprint appears in different batches, their stage contribution direction remains consistent. The stage aliasing limitation reduces the unfounded correlation between different stage contribution columns, so that the setting change of a certain stage will not be repeatedly interpreted by multiple stages. The advantage of this embodiment is that it makes the stage gating decomposition result have a separable matrix structure, reducing the contribution confusion caused by the simultaneous change of multiple surface treatment stages.

[0034] In a preferred embodiment, the stage coding processing unit includes an anomaly sample enhancement processing flow. This flow extracts comparison segments from historical batches that have similar process fingerprint sequences but different resistivity offset directions. The stage codes of these comparison segments are paired with the corresponding matrix resistivity contribution values. The criteria for determining similar process fingerprint sequences include consistent stage categories, consistent stage order, overlapping residence intervals, and consistent set change trajectory directions. Different resistivity offset directions refer to opposite residual signs or residual amplitude distributions falling into different directional intervals after matrix contribution stripping. After pairing, a control sample group is formed, and gating allocation is performed. During training, the processing unit simultaneously reads the stage code and matrix contribution value of the control sample group. When two groups of samples have similar surface treatment trajectories but large differences in matrix contribution, the gating allocation processing unit reduces the same-direction offset weight directly generated by the stage code. When two groups of samples have similar matrix contributions but a certain surface treatment stage code differs, the gating allocation processing unit retains the correlation between the stage code and the residual direction. The advantage of this embodiment is that by comparing samples, the model is forced to distinguish between resistivity offset caused by surface treatment stage and resistivity offset caused by matrix difference, avoiding distortion of stage contribution judgment caused by a small number of historical abnormal samples.

[0035] In one embodiment, the residual write-back unit includes an abnormal section aggregation processing flow. This flow differs the measured resistivity with the substrate resistivity contribution value and the resistivity offset contribution value at each stage to obtain a length measurement point residual sequence. Adjacent length measurement points are then aggregated into abnormal length intervals based on the continuity of residual sign, similarity of residual amplitude, and similarity of process fingerprints. The calculation method uses the following formula:

[0036] in, Indicates resistivity measurement point Abnormal residuals Indicates resistivity measurement point The measured resistivity, Indicates resistivity measurement point The predicted resistivity, Indicates the first An abnormal length interval, This indicates the operation of taking values ​​in the direction of the residual. This represents the residual difference boundary generated by the dispersion of the current convolution residual amplitude. This indicates the distance between adjacent measurement point process fingerprint sequences. This indicates the fingerprint similarity boundary generated by the current roll-to-roll fingerprint distribution; when the measurement point When the measured resistivity is 1.695 and the predicted resistivity is 1.6893, If the residuals of adjacent measuring points are 0.0057, have the same sign, and the difference between the residuals is no more than 0.0003, then the residuals are 0.0054. And the process fingerprint distance is no more than 0.08. Then the measuring point The advantage of this embodiment is that the isolated measurement point residuals are converted into abnormal intervals with length continuity and process similarity, which makes it easier to establish a correspondence with the dominant stage code in the stage contribution matrix.

[0037] Table 2 Residual Sample Record Field Configuration ; The fields shown in Table 2 are used to explain that the residual sample is not a single anomaly record, but a write-back data structure composed of anomaly length range, residual direction, dominant stage code, process fingerprint sequence, and matrix contribution value. The dominant stage code is used to limit the stage gating weight that needs to be corrected, the matrix contribution value is used to prevent matrix source differences from being rewritten into the surface treatment stage weight, and the write-back batch identifier is used to retrieve recurring stage residual patterns in batches of the same or similar specifications. The advantage of this embodiment is that it gives the residual write-back clear data boundaries and stage boundaries, avoiding the residual sample being used as a general anomaly label.

[0038] In one embodiment, the residual write-back unit further includes a weight iterative update process. This process jointly indexes the abnormal length intervals, dominant stage codes, and process fingerprint sequences in the residual samples, extracts recurring stage residual patterns across multiple batches, and writes these patterns into the process fingerprint weight update set. The stage gating decomposition unit calls the process fingerprint weight update set during subsequent batch processing to pre-correct the gating weights corresponding to similar process fingerprint sequences. The calculation method uses the following formula:

[0039] in, Indicates the first The updated process fingerprint correction coefficient Indicates the first Process fingerprint correction coefficient before phase update Indicates the first Historical correction retention factor for each stage Indicates the first The set of residual patterns that repeat across multiple batches in a given stage. Indicates the number of residual patterns in the set. Indicates the abnormal length range The residual representative value, This indicates an operation that limits the representative value of the residual to the range of the historical residuals of the stage. and They represent the first The lower and upper bounds of the stage residual truncation; when For 0.010, When the value is 0.6, and the truncated residual representative values ​​of the three outlier intervals in the repeated residual pattern set are 0.004, 0.005, and 0.003 respectively, The value is 0.0076. When similar process fingerprint sequences appear in subsequent batches, the stage gating decomposition unit writes the correction coefficient into the preset correction item before the corresponding stage gating score calculation. The advantage of this embodiment is that the stage offsets that repeatedly appear in the historical residuals can enter the subsequent calculation process, while single occasional residuals will not directly change the process fingerprint weight.

[0040] In a preferred embodiment, the process fingerprint weight update set is stored hierarchically according to specification information, stage category, fingerprint similarity range, and residual direction. When subsequent batches enter the stage gating decomposition unit, the gating allocation processing unit performs a similarity search between the process fingerprint sequence of the current measurement point and the historical process fingerprint sequence in the weight update set. During the search, the stage category, stage order, degree of overlap of the dwell interval, and set change trajectory direction are compared. If there are multiple candidate historical patterns, the candidate pattern reading order is determined according to the closeness of the matrix contribution value, the consistency of specifications, and the number of times the residual pattern is repeated. The correction coefficient after reading only applies to the dominant stage code corresponding to the candidate pattern and does not directly rewrite the gating weights of other stages. If the matrix contribution value of the current measurement point differs greatly from the matrix contribution value in the historical residual sample, the weight of the historical correction coefficient in the gating score is reduced so that the matrix difference is not mistakenly regarded as the same surface treatment stage offset. The advantage of this embodiment is that the weight write-back has stage limitation and matrix limitation, ensuring that the historical residual pattern is still matched with the current process fingerprint and the current matrix contribution when subsequent batches call the historical residual pattern.

[0041] In a preferred embodiment, when the system performs continuous processing on multiple resistivity measurement points of the same roll of copper foil, the cleaning data output by the data access processing unit is written into the roll-level cache, the process fingerprint sequence generated by the length coordinate mapping unit is written into the measurement point-level index, the matrix resistivity contribution value generated by the matrix contribution estimation unit is written into the measurement point benchmark table, the stage contribution matrix generated by the stage gating decomposition unit is written into the stage decomposition table, and the abnormal length interval and residual sample generated by the residual write-back unit are written into the residual sample table. The roll-level cache is used to maintain the consistency of the data processing range of the same roll, the measurement point-level index is used to ensure that each measurement point can be traced back to its multi-stage processing trajectory, the measurement point benchmark table is used to support matrix contribution verification, the stage decomposition table is used to support dominant stage code query, and the residual sample table is used to support multi-batch repetitive pattern retrieval. The data tables are associated with each other through roll number, measurement point length coordinate, and stage number. Data closure can be completed without introducing additional hardware structure. The advantage of this embodiment is that the output results of each processing unit have a traceable data carrier, which facilitates cross-verification of the same abnormal length interval among process trajectory, matrix contribution, stage contribution, and residual sample.

[0042] In one embodiment, after completing the processing of a roll of copper foil, the system generates resistivity decomposition analysis results. The analysis results include the measured resistivity of each resistivity measurement point, the matrix resistivity contribution value, the roughening stage offset contribution value, the curing stage offset contribution value, the anti-oxidation stage offset contribution value, the passivation stage offset contribution value, the predicted resistivity, the abnormal residual, the abnormal length interval number, and the dominant stage code. The generation process does not directly use the out-of-limit label as a conclusion. Instead, it first determines whether there is an offset that exceeds the matrix interpretation range based on the difference between the matrix contribution value and the measured resistivity. Then, it determines which surface treatment stage the offset is mainly concentrated in based on the stage contribution matrix. Finally, it determines whether the offset forms a continuous length interval based on the residual aggregation result. If the dominant stage code is consistent and the process fingerprint is similar within an abnormal length interval, the residual write-back unit writes the interval into the residual sample table. If the dominant stage code is inconsistent within the abnormal length interval, the interval is split into multiple sub-intervals and written back separately. The advantage of this embodiment is that it preserves the hierarchical structure of the output results from measurement point to interval, from matrix to stage, and from stage to residual, avoiding merging resistivity changes from different sources into the same abnormal category.

[0043] In a preferred embodiment, when analyzing a new batch of copper foil, the system reuses historical stable batches, historical residual samples, and historical process fingerprint weight update sets. After the data access processing unit completes the merging of the new batch data, the length coordinate mapping unit generates a new batch process fingerprint sequence. The matrix contribution estimation unit extracts historical benchmarks from stable batches of the same specification and generates matrix contribution values ​​by combining them with reference segments of the same roll in the new batch. Before calculating the stage gating weights, the stage gating decomposition unit searches the historical weight update set. If the process fingerprint of a certain length segment of the new batch matches the historical repeated residual pattern, the corresponding stage correction coefficient is added to the stage gating score. If the difference between the matrix contribution distribution of the new batch and the historical residual samples exceeds the current specification stable range, the effect of the historical correction coefficient on the current measurement point is canceled. After the analysis of the new batch is completed, the residual write-back unit continues to write the abnormal length intervals that form continuous evidence into the residual sample table. The advantage of this embodiment is that the system can maintain the consistency of stage attribution for similar surface treatment trajectories during the continuous accumulation of batches, while avoiding the mechanical reuse of residual patterns under different matrix conditions.

[0044] In one embodiment, the system employs a combination of interval interpolation and confidence marking to address situations where adjacent measurement point data is missing or there are local detection gaps. When a resistivity measurement point is missing in a certain length segment, the length coordinate mapping unit still generates the process fingerprint sequence for that segment based on the surface treatment process record. The matrix contribution estimation unit generates the matrix contribution estimate for that segment based on adjacent reference segments in the same volume and stable batches of the same specification. The stage gating decomposition unit marks the stage contribution matrix elements corresponding to the missing measurement points as pending verification. The residual write-back unit only calculates the residuals for measurement points with measured resistivity and forms abnormal length intervals. If the missing segment is located between two abnormal sub-intervals with the same residual direction and similar process fingerprints, the missing segment is recorded as an interval in the abnormal length interval, but its residual representative value is not written into the weight update set. The residual sample is updated based on the measured resistivity after subsequent detection data is added. The advantage of this embodiment is that it maintains the continuous expression of process fingerprints and matrix contributions when the detection data is incomplete, while avoiding data without measured residual support from directly affecting the stage gating weights.

[0045] In a preferred embodiment, the system employs a segmented retention processing method for situations where short-term abrupt changes occur in the surface treatment stage records. When the stage encoding processing unit identifies a directional abrupt change in the set change trajectory of a certain stage within a short length range, it does not merge the records before and after the abrupt change into the same average value. Instead, it splits the stage fingerprint element into multiple sub-elements according to the length coordinates corresponding to the abrupt change time. The gating allocation processing unit calculates the candidate resistivity offset contribution value for each sub-element and then merges them according to the proportion of the residence time of the measurement point in each sub-element. The contribution constraint processing unit establishes continuous constraints on both sides of the abrupt change position when restricting the continuity of the length direction, so as to avoid the real process abrupt change being smoothed out by the continuity constraint. When the residual write-back unit aggregates abnormal sections, it uses the abrupt change position as the optional interval boundary and determines whether to split the abnormal length interval based on the continuity of the residual sign. The advantage of this embodiment is that it enables short-term process abrupt changes to be used as independent inputs for stage contribution decomposition, while keeping the continuity constraints of adjacent stable sections unaffected by abrupt change segments.

[0046] In one embodiment, the system adopts a dynamic update method for screening stable batches of the same specification. Before a historical batch enters the stable sample set, the same specification benchmark screening processing unit reads the stage contribution matrix, abnormal residual sequence, and residual sample label of the batch. If the surface treatment stage contribution distribution of a certain batch is stable and the abnormal residual does not form a continuous abnormal length interval, the corresponding length segment of the batch is written into the candidate set of stable batches of the same specification. If a certain batch only forms an abnormal length interval in a local segment, the abnormal interval is removed and the remaining segments that meet the process fingerprint stability conditions are retained. The intra-volume continuous reference processing unit also adopts the same interval removal mechanism for the reference segment of the current volume. When calculating the matrix resistivity contribution value, the matrix fusion processing unit reads the updated stable sample set and reference segment set. The advantage of this embodiment is that the historical benchmark used for matrix contribution estimation can be continuously corrected with the detection data, avoiding segments that have been identified as surface treatment abnormalities from entering the matrix benchmark calculation.

[0047] In a preferred embodiment, a bidirectional verification process is established between the stage contribution matrix and the residual sample table. After generating the stage contribution matrix, the stage contribution separable processing flow takes the stage column with the largest contribution value in each abnormal length interval as the candidate dominant stage. The residual write-back unit then combines the process fingerprint similarity and residual direction of the interval to determine whether the candidate dominant stage has continuous evidence. If the candidate dominant stage of adjacent measurement points in the same interval switches frequently, but the process fingerprint sequence does not have a corresponding stage mutation, the contribution constraint processing unit adds stage aliasing restrictions again and updates the gating weights. If the switching position of the candidate dominant stage is consistent with the mutation position of the stage fingerprint sub-element, the switching result is retained and the abnormal length interval is split into multiple sub-intervals. The sub-intervals are associated with different dominant stage codes and written into the residual sample table. The advantage of this embodiment is that it enables the stage contribution matrix and the residual aggregation result to verify each other, reducing the misjudgment of the dominant stage caused by the contribution fluctuation of a single measurement point.

[0048] In one embodiment, the complete processing chain of the system operates in the order of data flow. The data access processing unit receives and merges the copper foil roll number, specification information, resistivity measurement point data, and surface treatment stage records. The length coordinate mapping unit generates continuous length segments and completes the measurement point assignment based on the linear velocity, stage dwell time, and process time. The substrate contribution estimation unit generates the substrate resistivity contribution value based on the stable batch of the same specification, adjacent sections of the same roll, and substrate process records. The stage gated decomposition unit generates the stage contribution matrix based on the process fingerprint sequence, substrate contribution value, and stage constraints. The residual write-back unit generates the residual sequence and aggregates abnormal length intervals based on the difference between the measured resistivity and the predicted resistivity. The weight iterative update processing flow writes the stage residual patterns that recur in multiple batches into the process fingerprint weight update set. During subsequent batch analysis, the stage gated decomposition unit reads the correction coefficients corresponding to similar fingerprints. The advantage of this embodiment is that the output of each processing unit becomes the input of the subsequent unit, and each residual result can return to the process fingerprint weight update set, forming a continuous decomposition and analysis mechanism for the copper foil surface treatment process and resistivity data.

Claims

1. A copper foil resistivity data analysis system integrating surface treatment processes, characterized in that, It includes a data access and processing unit, a length coordinate mapping unit, a matrix contribution estimation unit, a stage-gated decomposition unit, and a residual write-back unit; The data access and processing unit is used to receive copper foil roll number, specification information, resistivity measurement point data in the length direction, and surface treatment process stage records. The length coordinate mapping unit converts the surface treatment process stage record into a process fingerprint sequence corresponding to the resistivity measurement point along the length direction based on the linear velocity, stage dwell time, and process time. The matrix contribution estimation unit generates the matrix resistivity contribution value based on the same specification stable batch, adjacent sections of the same roll, and matrix process records. The stage-gated decomposition unit outputs the resistivity offset contribution value corresponding to each surface treatment stage based on the process fingerprint sequence. The residual write-back unit generates abnormal residuals and writes back the process fingerprint weights based on the measured resistivity, the contribution value of the matrix resistivity, and the contribution value of the resistivity offset.

2. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 1, characterized in that, The length coordinate mapping unit includes a stage trajectory resampling processing unit and a measurement point attribution processing unit; The stage trajectory resampling processing unit divides the roughening, curing, anti-oxidation and passivation process stage records into continuous length segments according to the copper foil running direction, and configures the stage category, entry time, exit time, dwell time and process setting change trajectory for each continuous length segment. The measuring point attribution processing unit uses the length coordinates of the resistivity measuring point as an index to retrieve the corresponding continuous length segments of the measuring point in each surface treatment stage, and splices the multi-stage segments into the process fingerprint sequence according to the actual process sequence.

3. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 2, characterized in that, The matrix contribution estimation unit includes a same-specification benchmark screening processing unit, an intra-volume continuous reference processing unit, and a matrix fusion processing unit; The same specification benchmark screening and processing unit extracts batch segments with consistent specifications and stable surface treatment status from historical data to form a specification benchmark sample set. The continuous reference processing unit within the roll extracts reference sections in the same roll of copper foil that do not exhibit abrupt changes in the process stage, based on their length adjacency. The substrate fusion processing unit generates a substrate resistivity contribution value corresponding to each resistivity measurement point based on the specification benchmark sample set, reference section, and substrate process record.

4. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 3, characterized in that, The stage gating decomposition unit includes a stage coding processing unit, a gating allocation processing unit, and a contribution constraint processing unit; The stage coding processing unit segments and encodes the process category, dwell interval, and set change trajectory of each surface treatment stage in the process fingerprint sequence. The gating allocation processing unit encodes and configures learnable gating weights for different stages in the order of stages, and outputs the resistivity offset contribution value corresponding to each stage. The contribution constraint processing unit adds length direction continuity constraints, stage sequence constraints, and batch consistency constraints of the same specification during the gating weight update process.

5. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 4, characterized in that, The stage trajectory resampling processing unit is equipped with a time-series breakpoint correction processing flow. The time-series breakpoint correction processing flow identifies length segment breakpoints based on the linear velocity changes, stage entry time, and stage exit time recorded by adjacent processes. It generates candidate compensation intervals for cross-stage overlapping segments and missing segments, and uses the continuity of dwell time between adjacent stages and the increasing relationship of copper foil length as screening conditions to determine the corrected continuous length segments. The measurement point attribution processing unit uses the corrected continuous length segment as the basis for the measurement point index to generate the process fingerprint sequence.

6. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 5, characterized in that, The substrate fusion processing unit includes a benchmark difference elimination process. This process maps the resistivity distribution in the same specification benchmark sample set and the resistivity distribution in the same roll reference section to a unified length coordinate scale. It extracts the substrate bias items corresponding to the thickness specification, length position, and previous substrate process records, and performs joint weighting on the substrate bias items according to the same specification batch stability and in-roll continuity to generate the substrate resistivity contribution value after excluding surface treatment stage disturbances.

7. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 6, characterized in that, The contribution constraint processing unit includes a stage contribution separable processing flow. The stage contribution separable processing flow establishes a stage contribution matrix based on the multi-stage encoding corresponding to the same resistivity measurement point. The stage contribution difference between adjacent length measurement points, the cross-batch contribution difference in the same stage, and the correlation of contributions in different stages are used as constraint terms. When updating the gating weight, length mutation, batch drift, and stage aliasing are restricted respectively. The gating allocation processing unit outputs the resistivity offset contribution value of each surface treatment stage according to the restricted gating weight.

8. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 7, characterized in that, The stage coding processing unit includes an abnormal sample enhancement processing flow. The abnormal sample enhancement processing flow extracts comparison segments with similar process fingerprint sequences but different resistivity offset directions from historical batches, pairs the stage codes of the comparison segments with the corresponding matrix resistivity contribution values, and forms a control sample group in the stage gated decomposition training. The gating and allocation processing unit distinguishes between the resistivity shift contribution value caused by the surface treatment stage and the resistivity shift contribution value caused by matrix differences based on the control sample group.

9. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 1, characterized in that, The residual write-back unit includes an abnormal section aggregation process. The abnormal section aggregation process differentially analyzes the measured resistivity with the contribution value of the matrix resistivity and the contribution value of the resistivity offset at each stage to obtain a length measurement point residual sequence. Adjacent length measurement points are aggregated into abnormal length intervals according to the continuity of residual sign, similarity of residual amplitude and similarity of process fingerprint. The abnormal length interval is associated with the dominant stage code in the stage contribution matrix and a residual sample is generated for updating the process fingerprint weights.

10. The copper foil resistivity data analysis system based on the integrated surface treatment process according to claim 9, characterized in that, The residual write-back unit includes a weight iterative update process. The weight iterative update process performs a joint indexing of the abnormal length range, dominant stage code, and process fingerprint sequence in the residual sample, extracts the stage residual pattern that appears repeatedly in multiple batches, and writes the stage residual pattern into the process fingerprint weight update set. The stage-gating decomposition unit calls the process fingerprint weight update set during subsequent batch processing to pre-correct the gating weights corresponding to similar process fingerprint sequences.

Images