Optimization methods for mass production process parameters of SiC MOSFETs adapted to 6-inch production lines

Abstract

The application discloses a SiC MOSFET mass production process parameter optimization method suitable for a 6-inch production line, and particularly relates to the field of SiC MOSFET mass production process optimization, and comprises the following steps: taking wafer numbers as chain heads in edge computing nodes, establishing wafer chain records of cavity numbers, loading sequences, formula parameters, failure coordinates and abnormal types in a 6-inch mass production batch, and generating batch chain tables according to batch numbers at the end of the process and the loading sequence; and through chain formation of the cavity numbers, the loading sequences, the formula parameters and the electric measurement failure coordinates according to the wafer numbers, and based on device fragments, formula fragments and failure clusters, performing element replacement group operation and non-backtracking attribution shunting, so as to solve the problem that an electrical abnormality caused by device execution or loading position is mistakenly corrected as a process formula parameter abnormality.

Description

Technical Field

[0001] This invention relates to the field of SiC MOSFET mass production process optimization technology, and more specifically, to a method for optimizing SiC MOSFET mass production process parameters to suit 6-inch production lines. Background Technology

[0002] In the mass production of SiC MOSFETs, the production line typically uses MES to store batch, recipe and flow records, and uses FDC to collect temperature, pressure, gas flow, running time and alarm data of oxidation furnace, annealing furnace, injection machine and metallization equipment. Combined with wafer-level electrical test results, batch corrections are made to oxidation time, annealing temperature, injection dosage and metal annealing conditions. When the process shifts from small-size wafers to 6-inch production lines, the same batch of wafers needs to be processed continuously in multi-cavity equipment according to the loading sequence. Online measurement, electrical test pattern generation, and parameter distribution for the next batch must be completed within a limited cycle time. Edge computing nodes can only make judgments based on the cavity number, loading sequence, process execution time, equipment process curve, and electrical test failure coordinates already generated for the current batch. At this point, if a certain type of leakage current increase, breakdown voltage dispersion, or threshold voltage deviation occurs concentrated in the same cavity, loading sequence, or several batches after equipment maintenance, simply relying on batch yield, average electrical properties, or single-process SPC results to revert to the formula parameters will misattribute cavity thermal field deviations, loading position differences, or equipment recovery period drift to oxidation, annealing, or injection parameter deviations. This leads to the next batch formula being modified, yet the anomalies still reappear with the cavity or loading sequence. Even if a Markov model is used to describe batch state changes, if the equipment state, loading state, and electrical failure state are not linked according to the same wafer number and process sequence, it is impossible to distinguish between equipment compensation, loading adjustment, and formula correction. Therefore, the technical problem to be solved by this application is: how to determine the state transmission relationship between equipment cavity, wafer loading and electrical failure based on edge computing within the mass production cycle of 6-inch SiC MOSFET, so as to avoid miscorrecting electrical abnormalities caused by equipment execution or loading position as abnormal process formula parameters. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for optimizing SiCMOSFET mass production process parameters adapted to 6-inch production lines. This method links cavity numbers, loading order, recipe parameters, and electrical failure coordinates according to wafer numbers, and performs substitution group operations and non-backtracking attribution shunting based on equipment segments, recipe segments, and failure clusters to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing SiC MOSFET mass production process parameters to suit 6-inch production lines, comprising: S1. In the edge computing node, with the wafer number as the chain head, establish a wafer chain record for the cavity number, loading order, formula parameters, failure coordinates and anomaly type within the 6-inch mass production batch, and assign batch tag numbers according to the process end time and loading order to generate a batch chain list. S2. Based on the batch linked list, wafer records with the same device key and adjacent anomalies across recipes are grouped into device segments, wafer records with the same recipe and similar anomalies across cavities and loading are grouped into recipe segments, and adjacent failure coordinates in the same set are grouped into failure clusters to generate a segment table. S3. Perform Kikuchi substitution group operation on the fragment table to form a ternary group of equipment fragments, recipe fragments and failure clusters under the same anomaly type. Generate equipment transfer amount based on recipe replacement breaks after maintaining cavity number and loading order, and generate recipe transfer amount based on cavity number and loading order replacement breaks after maintaining recipe parameters, and generate group transfer table. S4. Perform non-backward confidence propagation operation on the cluster transfer table, generate an inbound edge table with cluster number, cluster edge direction and previous cluster number, set the reverse back edge to zero, and perform cross suppression on the equipment transfer amount and formula transfer amount within the same cluster number to generate equipment attribution value and formula attribution value. S5. The attribution difference is generated by subtracting the formula attribution value from the equipment attribution value under the same group number. When the attribution difference is positive, an equipment correction record is generated based on the equipment segment and a formula blocking mark is written. When the attribution difference is negative, a formula correction record is generated based on the formula segment. When the attribution difference is zero, a formula blocking mark is written.

[0005] In a preferred embodiment, S1 includes: S1-1. Using the wafer number as the chain head, assemble the cavity number, loading order, formula parameters and process end time into a process row, and generate chain position number according to the order of process end times within the same wafer number to obtain the wafer process chain. S1-2. Using the wafer number as an index, associate the failure coordinates and anomaly types with the process line at the end of the chain number in the wafer process chain, and mark the process lines without associated failure coordinates as empty failure lines to obtain the wafer failure chain. S1-3. Rearrange the wafer failure chain according to the process end time. If the process end times are the same, rearrange them according to the loading order and generate batch tag numbers based on the rearranged row numbers to obtain the batch linked list.

[0006] In a preferred embodiment, S2 includes: S2-1. Remove empty failure rows from the batch list, combine the cavity number, loading order and anomaly type into a device key, combine the recipe parameters and anomaly type into a recipe key, combine the failure coordinates and anomaly type into a coordinate key, and retain the batch tag number to generate an anomaly row table. S2-2. For abnormal rows with the same device key, read them in ascending order of batch number. Merge abnormal rows with adjacent batch numbers and changing recipe keys into the same device segment. Cut out abnormal rows with the same recipe key and disconnected batch numbers from the device segment to generate a device segment row.

[0007] In a preferred embodiment, S2 further includes: S2-3. For abnormal rows with the same recipe key, merge abnormal rows that cross cavity numbers, cross loading order and have the same abnormal type into the same recipe segment, and exclude abnormal rows that fall only into a single device key from the recipe segment to generate a recipe segment line. S2-4. Expand the failure coordinates according to the coordinate key, merge the failure coordinates that are adjacent and whose wafer number sets fall into the equipment segment line or recipe segment line into failure clusters, and generate a segment table according to the intersection of wafer number sets.

[0008] In a preferred embodiment, S3 includes: S3-1. Using the same anomaly type in the fragment table as the group number, read the equipment fragment, recipe fragment and failure cluster that have an intersection in the wafer number set. Combine the cavity number, loading order and wafer number set in the equipment fragment, the recipe parameters and wafer number set in the recipe fragment, and the failure coordinate and wafer number set in the failure cluster into a ternary group, and take the intersection of the wafer number sets of the three to generate the wafer set within the group. S3-2. Perform Kikuchi substitution operation on the ternary group, arrange the wafer set in the group according to the batch number to form the group position string, keep the cavity number and loading order in the group position string unchanged and replace the position string of the formula parameter to generate the equipment substitution string, keep the formula parameter in the group position string unchanged and replace the position string of the cavity number and loading order to generate the formula substitution string.

[0009] In a preferred embodiment, S3 further includes: S3-3. Read the failure coordinates bit by bit along the equipment replacement string. If the failure coordinates after replacing the formula parameters still fall into the failure cluster, mark this position as the equipment reserved position. If the failure coordinates after replacing the formula parameters leave the failure cluster, mark this position as the equipment cut-off position and truncate the equipment reserved positions after this position to generate the equipment cut-off string. S3-4. Read the failure coordinates bit by bit along the formula substitution string. If the failure coordinates after replacing the cavity number and loading order still fall into the failure cluster, mark this position as a formula retention position. If the failure coordinates after replacing the cavity number and loading order leave the failure cluster, mark this position as a formula break and truncate the formula retention positions after this position to generate a formula truncation string.

[0010] In a preferred embodiment, S3 further includes: S3-5. Perform the expected propagation breakpoint re-feedback operation on the equipment truncated string and the recipe truncated string. Re-feed the equipment reservation in the equipment truncated string to the equipment segment according to the batch tag number, and re-feed the recipe reservation in the recipe truncated string to the recipe segment according to the batch tag number. If the same batch tag number has both equipment reservation and recipe reservation, a conflict bit is generated; otherwise, a one-way reservation is generated. S3-6. Perform non-backtracking deduction operation on conflict bits and one-way reservation bits, synchronously delete conflict bits from equipment reservation bits and recipe reservation bits, generate equipment net strings from the reserved equipment reservation bits, and generate recipe net strings from the reserved recipe reservation bits. S3-7. Generate the device transfer amount from the bit length of the device net string and the formula transfer amount from the bit length of the formula net string. Then, merge the group number, the fault type, the wafer set within the group, the device transfer amount, and the formula transfer amount to generate the group transfer table.

[0011] In a preferred embodiment, S4 includes: S4-1. Read the equipment transfer amount and formula transfer amount of the same ternary group using the group number in the group transfer table. Record the edge from the ternary group to the equipment segment as the equipment group edge and the edge from the ternary group to the formula segment as the formula group edge. Generate the inbound edge table with the group number, the group edge direction and the previous group number. S4-2. Perform non-backtracking shadow edge operation on the inbound edge table. Record the propagation items where the previous cluster number is swapped with the current cluster number and the cluster edge direction is opposite as back edges. Reset the propagation weight of the back edges to zero. Retain the original propagation weights of the remaining propagation items and generate a de-back weight table.

[0012] In a preferred embodiment, S4 further includes: S4-3. Perform cross-suppression operation based on the weight table. When there are both equipment group edge and formulation group edge in the same group number, generate equipment net weight by subtracting the positive value of formulation transfer amount from equipment transfer amount, and generate formulation net weight by subtracting the positive value of equipment transfer amount from formulation transfer amount, and generate net weight table. S4-4. Based on the net weight table, the net weight of equipment under the same equipment segment is recursively accumulated by group number to generate the equipment attribution value, and the net weight of formula under the same formula segment is recursively accumulated by group number to generate the formula attribution value.

[0013] In a preferred embodiment, S5 includes: S5-1. The attribution difference is generated by subtracting the formula attribution value from the equipment attribution value under the same group number. When the attribution difference is positive, the equipment position is generated; when the attribution difference is negative, the formula position is generated; when the attribution difference is zero, the parallel position is generated. A decision table is generated from the group number, equipment segment, formula segment, and attribution difference. S5-2. For the equipment position in the decision table, generate equipment correction records based on the cavity number, loading order and abnormality type in the equipment segment, and generate formula blocking markers for formula segments and formula parameters under the same group number. S5-3. For the recipe position in the decision table, generate a recipe correction record based on the recipe parameters and anomaly type in the recipe segment; for the parallel positions in the decision table, write both the equipment segment and the recipe segment under the same group number into the recipe blocking flag.

[0014] The technical effects and advantages of this invention are as follows: 1. By chaining the cavity number, loading order and failure coordinates according to the wafer number through edge computing, and then performing attribution and diversion processing, the situation where equipment execution deviation is mistakenly judged as abnormal formula parameters can be relatively reduced. 2. By using equipment keys, recipe keys, and coordinate keys to separate equipment segments, recipe segments, and failure clusters, the source of anomalies is changed from batch average judgment to segment relationship judgment, and the parameter correction objects are more in line with the actual production line. 3. By constructing equipment substitution strings and recipe substitution strings through Kikuchi substitution group operations, and generating transfer amounts by retaining or breaking failure clusters, the transmission paths of equipment loading factors and recipe factors can be relatively distinguished. 4. By re-injecting the expected propagation breakpoint and deleting conflicting bits without retrospective deduction, the repeated enhancement of the same failure coordinate on the equipment side and the formulation side is suppressed, making the attribution results more suitable for subsequent correction and diversion. 5. By generating device bits, recipe bits, and parallel bits through attribution difference, and writing recipe blocking flags under device bits or parallel bits, the probability of erroneously modifying oxidation, annealing, and injection parameters can be relatively reduced. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1 The present invention provides a method for optimizing SiC MOSFET mass production process parameters for 6-inch production lines, comprising: S1. In the edge computing node, with the wafer number as the chain head, establish a wafer chain record for the cavity number, loading order, formula parameters, failure coordinates and anomaly type within the 6-inch mass production batch, and assign batch tag numbers according to the process end time and loading order to generate a batch chain list. In this embodiment, the edge computing node first converts the scattered process execution data and wafer electrical test data generated within the 6-inch SiC MOSFET mass production batch into a chain record under the same wafer number, so that subsequent equipment segments, recipe segments, and failure clusters can all read fields from the same batch chain list; the wafer number is used to connect the process lines of the same wafer in different processes; the process end time is used to form the chain position number inside the wafer; the loading order is used to process the batch sorting under the same end time; the failure coordinates and anomaly type are connected to the chain tail as the end observation results after the wafer completes the process, avoiding the mixing of electrical test results with intermediate process lines; this implementation process includes the following steps: In S1-1, the edge computing node uses the wafer number as the chain head, reads the cavity number from the equipment process record, the loading order from the batch loading record, the recipe parameters from the process recipe record, and the process end time from the equipment end record. The recipe parameters consist of a process identifier, parameter name, and parameter value. When the same wafer record contains multiple parameters, they are formed into a recipe parameter string according to the process execution order. The edge computing node assembles the cavity number, loading order, recipe parameters, and process end time under the same wafer number and the same process end time into a process line. When there are multiple process lines under the same wafer number, they are arranged from first to last according to the process end time, and the arrangement position generates a chain position number, which increments from the first process line. The resulting wafer process chain includes a chain head, a chain position number, and a process line. The chain head is used to distinguish wafers, the chain position number indicates the sequential position of process lines within the same wafer, and the process line is used to retain the cavity number, loading order, recipe parameters, and process end time. In S1-2, the edge computing node uses the wafer number as an index to read the failure coordinates and anomaly types from the wafer electrical test record. The failure coordinates are represented by chip row and column coordinates or wafer plane coordinates, and the anomaly types are taken from the electrical test judgment results, including threshold voltage anomalies, on-resistance anomalies, leakage anomalies, breakdown anomalies, or gate-drain anomalies. Since the wafer electrical test occurs after the current wafer completes the set process flow, the edge computing node associates the failure coordinates and anomaly types under the same wafer number with the process row at the end of the chain number in the wafer process chain, making the electrical test results the end observation item of the current wafer chain. If no failure coordinates are obtained for the same wafer number, the process row at the end of the chain number is marked as an empty failure row. This results in a wafer failure chain, which retains the process execution process and receives the electrical test results at the end of the chain, providing input for subsequent removal of empty failure rows and generation of anomaly row table. In S1-3, the edge computing node reads the wafer failure chain under each wafer number, takes the last process line of the chain position number in the wafer failure chain as the batch sorting object, and rearranges them from first to last according to the process end time; if two or more process lines have the same process end time, they are rearranged from first to last according to the loading order; after the rearrangement is completed, the edge computing node generates the rearranged line number as the batch position number, which represents the global position of the tail record of each wafer chain in the current mass production batch; the chain position number still represents the position of the process line inside a single wafer, and the batch position number represents the processing position at the batch level, and the two are retained respectively; thus, a batch linked list is obtained, and each row in the batch linked list includes at least the wafer number, chain position number, batch position number, cavity number, loading order, recipe parameters, process end time, failure coordinate, anomaly type, and empty failure line marker; Through this implementation, the fields scattered in the equipment process record, batch loading record, process recipe record and wafer electrical test record are organized into a batch linked list with the wafer number as the head, the chain position number indicating the internal process sequence of the wafer, and the batch position number indicating the global sequence of the batch. This allows subsequent steps to directly read abnormal records within the same batch according to the equipment key, recipe key and coordinate key, reducing the situation of misconnection between electrical test results and process execution records. In practical applications: A batch of 6-inch SiC MOSFETs contains multiple wafers. Wafer W01 completes the oxidation process in cavity A, loading order 2, with the process ending at 10:15. The formula parameter string is recorded as a combination of values ​​for "oxidation temperature / oxidation time / nitriding annealing time". After completion, an edge leakage anomaly occurs during electrical testing, with the failure coordinates being from row 34, column 8 to row 34, column 10. The edge computing node first assembles the oxidation process data into the process row of W01 and generates a chain number. Then, it associates the edge leakage anomaly and failure coordinates with the end of the process chain of wafer W01. Subsequently, it generates batch numbers for W01 and other wafers in the same batch according to the process ending time and loading order, forming a batch linked list for subsequent fragment merging and cluster transfer operations.

[0018] S2. Based on the batch linked list, wafer records with the same device key and adjacent anomalies across recipes are grouped into device segments, wafer records with the same recipe and similar anomalies across cavities and loading are grouped into recipe segments, and adjacent failure coordinates in the same set are grouped into failure clusters to generate a segment table. In this embodiment, the batch list has arranged the process chain tail records of each wafer according to the batch tag number. S2 is used to extract records with electrical test anomalies from the batch list and split the anomaly records into equipment segments, recipe segments, and failure clusters. Equipment segments are used to represent the situation where anomalies still occur adjacent to each other along the batch tag number after the recipe parameters change, under the same cavity number and the same loading order. Recipe segments are used to represent the situation where anomalies are still retained after crossing cavity numbers and loading orders under the same recipe parameters. Failure clusters are used to represent the adjacent clustering results of similar anomalies in the wafer space coordinates. The implementation process includes the following steps: In S2-1, the edge computing node reads the batch linked list, first deleting records marked with empty failure rows, and retaining only wafer records containing failure coordinates and anomaly types to form an anomaly record set. For each row in the anomaly record set, the cavity number, loading order, and anomaly type are combined into a device key in a fixed field order. The device key is used to represent the same type of anomaly under the same equipment loading conditions. The recipe parameters and anomaly type are combined into a recipe key, which is used to represent the same type of anomaly under the same recipe conditions. The failure coordinates and anomaly type are combined into a coordinate key, which is used to represent the position of the same anomaly type in wafer space. When generating the device key, recipe key, and coordinate key, the edge computing node retains the batch reference number, wafer number, cavity number, loading order, recipe parameters, failure coordinates, and anomaly type to obtain an anomaly row table. In S2-2, the edge computing node groups the abnormal row table according to the device key, and reads the abnormal rows in each device key group in ascending order of batch tag number. When the current abnormal row has the same device key as the previous abnormal row, the batch tag number is adjacent, and the recipe key has changed, it indicates that the same type of abnormality under the same cavity number and the same loading order is still retained after the recipe parameter changes, and the edge computing node merges the current abnormal row into the previous device segment. When the current abnormal row has the same device key as the previous abnormal row but the batch tag number is not adjacent, the edge computing node ends the previous device segment and generates the next device segment from the current abnormal row. When the current abnormal row has the same device key and recipe key as the previous abnormal row but the batch tag number is disconnected, the current abnormal row is not used as the basis for the continuation of the previous device segment. Each device segment row contains at least the device segment number, device key, wafer number set, batch tag number range, cavity number, loading order, and abnormality type. In S2-3, the edge computing node groups the abnormal row table according to the recipe key, and reads the cavity number, loading order, and abnormal type within each recipe key group. If an abnormal row under the same recipe key covers more than two cavity numbers and more than two loading orders, and the abnormal type is the same, it means that the same type of abnormality under the same recipe parameter is not limited by a single cavity number or a single loading order, and the edge computing node merges these abnormal rows into the same recipe segment. If an abnormal row under the same recipe key falls only into a single device key, the edge computing node excludes these abnormal rows from the recipe segment and retains them on the device segment side for subsequent calculations. Each recipe segment row contains at least the recipe segment number, recipe key, wafer number set, recipe parameter, abnormal type, cavity number set, and loading order set. In S2-4, the edge computing node reads the failure coordinates from the anomaly row table based on the coordinate key and performs coordinate expansion under the same anomaly type. If the difference between the horizontal coordinates of two failure coordinates is one and their vertical coordinates are the same, or the difference between their vertical coordinates is one and their horizontal coordinates are the same, then the two failure coordinates are determined to be adjacent. The edge computing node uses the adjacent failure coordinates as the expansion direction and merges the failure coordinates with adjacent coordinates and wafer number sets entering the equipment segment row or recipe segment row into the same failure cluster. After generating the failure cluster, the edge computing node reads the wafer number sets of the equipment segment row, recipe segment row, and failure cluster row respectively, and establishes the connection relationship among the three based on the intersection of the wafer number sets to generate a segment table. The segment table contains at least the equipment segment number, recipe segment number, failure cluster number, anomaly type, equipment key, recipe key, wafer number set, failure coordinate set, and batch number segment. In this implementation, the abnormal records in the batch list are divided into three types of objects: equipment fragments, recipe fragments, and failure clusters. Equipment fragments retain evidence that "abnormalities still occur adjacently after the equipment loading conditions remain unchanged and the recipe parameters change." Recipe fragments retain evidence that "abnormalities still occur after the recipe parameters remain unchanged and the cavity number and loading order change." Failure clusters retain the clustering positions of similar abnormalities in the wafer space. After the three types of objects are joined together by the wafer number set, the subsequent S3 can directly construct ternary clusters around the equipment fragments, recipe fragments, and failure clusters, avoiding the mistaken inclusion of abnormalities in a single equipment key into the recipe correction path. In practical applications: In a batch of 6-inch SiC MOSFET wafers, wafers in cavity A and loading order 2 still exhibit leakage anomalies in adjacent batch reference numbers after the recipe parameter is changed from P1 to P2. Edge computing nodes classify these anomaly rows into device segments. Leakage anomalies under the same recipe parameter P3 appear simultaneously in wafers in cavity A, cavity B, and loading order 1 and loading order 3. Edge computing nodes classify these anomaly rows into recipe segments. If the failure coordinates of these wafers are concentrated in row 34, column 8 to row 34, column 10, with adjacent coordinates and the same anomaly type, a failure cluster is generated, and a segment table is generated by the intersection of wafer number sets.

[0019] S3. Perform Kikuchi substitution group operation on the fragment table to form a ternary group of equipment fragments, recipe fragments and failure clusters under the same anomaly type. Generate equipment transfer amount based on recipe replacement breaks after maintaining cavity number and loading order, and generate recipe transfer amount based on cavity number and loading order replacement breaks after maintaining recipe parameters, and generate group transfer table. In this embodiment, S3 takes the fragment table as input and places the equipment fragment, recipe fragment, and failure cluster under the same anomaly type for cluster structure calculation. This is used to distinguish whether the anomaly is retained along the equipment loading factors or along the recipe parameters. The calculation process does not regenerate virtual electrical test results. Instead, it searches for existing wafer records in the fragment table and batch chain list. By combining the fields after substitution, it checks whether the failure coordinates still belong to the original failure cluster. Then, it uses breakpoint truncation, breakpoint reinjection, and non-backtracking deduction to form the equipment transfer amount and recipe transfer amount. This implementation process includes the following steps: In S3-1, the edge computing node uses the anomaly type in the fragment table as the basis for the cluster number, and reads equipment fragments, recipe fragments, and failure clusters with the same anomaly type. Among them, the equipment fragment provides cavity number, loading order, and wafer number set; the recipe fragment provides recipe parameters and wafer number set; and the failure cluster provides failure coordinates and wafer number set. The edge computing node first determines whether there is an intersection between the wafer number set of the equipment fragment, the wafer number set of the recipe fragment, and the wafer number set of the failure cluster. If the intersection is not empty, the equipment fragment, recipe fragment, and failure cluster are merged into a ternary cluster, and the intersection of the wafer number sets of the three is taken as the wafer set within the cluster. If the intersection is empty, no ternary cluster is generated. The cluster number is formed by combining the anomaly type, equipment fragment number, recipe fragment number, and failure cluster number, so that multiple ternary clusters under the same anomaly type can participate in subsequent calculations respectively. In S3-2, the edge computing node performs Kikuchi substitution cluster operations on the ternary cluster. First, it arranges the wafer sets within the cluster according to the batch reference number in ascending order to obtain the cluster reference string. Each bit in the cluster reference string retains the wafer number, cavity number, loading order, recipe parameters, failure coordinates, and batch reference number. When generating the equipment substitution string, the edge computing node keeps the cavity number and loading order in the cluster reference string unchanged and reads the recipe parameters from other recipe fragments under the same anomaly type as the replacement source, replacing the original recipe parameters bit by bit. When generating the recipe substitution string, the edge computing node keeps the recipe parameters in the cluster reference string unchanged and reads the cavity number and loading order from other equipment fragments under the same anomaly type as the replacement source, replacing the original cavity number and original loading order bit by bit. The field combinations formed after substitution are all returned to the fragment table and batch chain table to query existing wafer records, avoiding the generation of failure coordinates out of thin air. In S3-3, the edge computing node reads the combined fields after substitution bit by bit along the device substitution string according to the batch tag number, and queries the fragment table for existing wafer records with unchanged cavity number and loading order and replaced formula parameters. If the failure coordinates of the found existing wafer record still belong to the failure coordinate set of the current failure cluster, the current position is marked as device reserved. If no existing wafer record is found, or the found failure coordinates do not belong to the current failure cluster, the current position is marked as device disengaged. After the first occurrence of device disengagement, the edge computing node removes the device reserved positions after the current position from the device substitution string and no longer counts them in the device-side retention results. The device truncated string is generated from the device reserved positions retained before the first device disengagement. This rule is used to express that after the formula parameters are replaced, the failure cluster is still retained along the same cavity number and the same loading order until the disengagement. In S3-4, the edge computing node reads the combined fields after substitution bit by bit along the recipe substitution string according to the batch number, and queries the fragment table for existing wafer records after the recipe parameters remain unchanged and the cavity number and loading order have been replaced. If the failure coordinates of the existing wafer record found still belong to the failure coordinate set of the current failure cluster, the current position is marked as a recipe retention position. If no existing wafer record is found, or the failure coordinates found do not belong to the current failure cluster, the current position is marked as a recipe break position. After the recipe break position appears for the first time, the edge computing node truncates the recipe retention positions after the current position from the recipe substitution string and no longer counts them in the recipe-side retention results. The recipe truncation string is generated from the recipe retention positions retained before the first recipe break position. This rule is used to express that after the cavity number and loading order are replaced, the failure cluster is still retained along the same recipe parameters up to before the break position. In S3-5, the edge computing node performs the expected propagation breakpoint refeeding operation on the device truncated string and the recipe truncated string. The device hold-in in the device truncated string is refeeded to the original device segment according to the batch tag number, forming the retention evidence of the device segment at the corresponding batch tag number. The recipe hold-in in the recipe truncated string is refeeded to the original recipe segment according to the batch tag number, forming the retention evidence of the recipe segment at the corresponding batch tag number. If the same batch tag number receives both device hold-in and recipe hold-in, the edge computing node marks the batch tag number as a conflict bit. If the same batch tag number only receives device hold-in or only receives recipe hold-in, it is marked as a one-way hold-in. This operation refeeds the breakpoint result after substitution to the segment object, so that the evidence conflict between the device segment and the recipe segment under the same batch tag number can be subsequently deducted and processed. In S3-6, edge computing nodes perform non-backtracking deduction operations on conflicting bits and unidirectional hold bits. For conflicting bits, edge computing nodes synchronously delete failure coordinates that simultaneously enter equipment hold bits and formula hold bits under the same batch tag number from both sides to prevent the same failure coordinate from increasing the equipment transfer amount and formula transfer amount at the same time. For unidirectional hold bits, edge computing nodes retain the hold bits in their respective directions and do not delete them. After deleting conflicting bits, the remaining equipment hold bits are arranged by batch tag number to generate a net equipment string, and the remaining formula hold bits are arranged by batch tag number to generate a net formula string. This rule prevents evidence of repeated backflow within ternary groups from amplifying the attribution results in the reverse direction. In S3-7, the edge computing node calculates the bit length of the device net string and uses the number of device reserved bits in the device net string as the device transfer amount; it also calculates the bit length of the recipe net string and uses the number of recipe reserved bits in the recipe net string as the recipe transfer amount. Subsequently, the edge computing node merges the cluster number, anomaly type, wafer set within the cluster, device net string, recipe net string, device transfer amount, and recipe transfer amount into a cluster transfer row and arranges them by cluster number to generate a cluster transfer table. The cluster transfer table is read by subsequent non-backtracking belief propagation operations, where the device transfer amount represents the length of the failed cluster retained in the recipe replacement path after maintaining the cavity number and loading order, and the recipe transfer amount represents the length of the failed cluster retained in the device loading replacement path after maintaining the recipe parameters. In this implementation, equipment fragments, formula fragments, and failure clusters are no longer attributed solely based on their frequency of occurrence. Instead, they are grouped into a cluster transfer table through substitution, dislocation, refilling, and deduction. If a failure cluster remains along the same cavity number and loading order after replacing formula parameters, the equipment transfer amount increases. If a failure cluster remains along the same formula parameters after replacing cavity number and loading order, the formula transfer amount increases. If the same batch tag number supports both paths simultaneously, conflicting bits are deleted to avoid redundant enhancement. In practical applications: Under the same leakage anomaly type, the equipment segment shows that cavity A and loading order 2 still exhibit adjacent batch anomalies after changes between recipe parameters P1 and P2. The recipe segment shows that recipe parameter P3 still exhibits leakage anomalies across multiple cavities and loading orders, with failure clusters concentrated in the wafer edge coordinate region. The edge computing nodes first form ternary clusters, then keep cavity A and loading order 2 to replace the recipe parameters. If the existing wafer record after replacement still falls into the original edge failure cluster, an equipment reservation is formed. At the same time, keep the cavity number and loading order of P3 to replace. If the existing wafer record after replacement still falls into the original edge failure cluster, a recipe reservation is formed. If the reservations on both sides meet in the same batch number, they are deleted as conflict bits. The remaining reservations form the equipment net string and the recipe net string, and the equipment transfer amount and recipe transfer amount are obtained from the bit length.

[0020] S4. Perform non-backward confidence propagation operation on the cluster transfer table, generate an inbound edge table with cluster number, cluster edge direction and previous cluster number, set the reverse back edge to zero, and perform cross suppression on the equipment transfer amount and formula transfer amount within the same cluster number to generate equipment attribution value and formula attribution value. In this embodiment, S4 takes the cluster transfer table as input and converts the equipment transfer amount and formulation transfer amount formed in S3 into attribution propagation results; the ternary cluster does not directly output the correction action, but first forms a propagation path through the equipment cluster edge and the formulation cluster edge, then cuts off the back edge propagation using non-backtracking shadow edge operation, and then performs cross-suppression on the equipment side and the formulation side within the same cluster number, so that the equipment attribution value and formulation attribution value under the same anomaly type are recursively generated from the net weight after backtracking; this implementation process includes the following steps: In S4-1, the edge computing node uses the cluster number in the cluster transfer table as an index to read the equipment transfer volume, formula transfer volume, equipment fragment, and formula fragment under the same ternary cluster. The propagation edge from the ternary cluster to the equipment fragment is recorded as the equipment cluster edge, and the propagation weight of the equipment cluster edge is taken as the equipment transfer volume. The propagation edge from the ternary cluster to the formula fragment is recorded as the formula cluster edge, and the propagation weight of the formula cluster edge is taken as the formula transfer volume. The previous cluster number is obtained from the same equipment fragment or the same formula fragment. The edge computing node arranges the cluster numbers in ascending order according to the starting point of the batch number segment, and takes the item before the current cluster number as the previous cluster number. If there is no previous cluster number, the previous cluster number is set to an empty flag. The edge computing node merges the cluster number, the previous cluster number, the cluster edge direction, the propagation weight, the equipment fragment, and the formula fragment to generate an inbound edge table. In S4-2, the edge computing node performs non-backtracking shadow edge operations on the incoming edge table. For each propagation item, the edge computing node reads the current round cluster number, the previous cluster number, and the cluster edge direction. If the previous round propagation item enters a certain cluster number via the device cluster edge, and the current round propagation item returns from the same cluster number to the previous cluster number via the formula cluster edge, or if the previous round propagation item enters a certain cluster number via the formula cluster edge, and the current round propagation item returns from the same cluster number to the previous cluster number via the device cluster edge, then the current round propagation item is recorded as a back edge. The propagation weight of the back edge is reset to zero, and the back edge record is still retained in the table for path tracing. Propagation items that have not formed a back edge retain their original propagation weight. After the back edge is reset to zero, the edge computing node generates a back-to-back weight table, which includes at least the cluster number, the previous cluster number, the cluster edge direction, the original propagation weight, and the propagation weight after back-to-back. In S4-3, edge computing nodes perform cross-suppression operations based on the de-propagation weight table. For cases where both equipment and formulation edges exist within the same cluster number, the edge computing node reads the de-propagation weight of the equipment edge as the equipment transfer amount and the de-propagation weight of the formulation edge as the formulation transfer amount. If the result of subtracting the formulation transfer amount from the equipment transfer amount is greater than zero, the result is used as the equipment net weight; otherwise, the equipment net weight is zero. Similarly, if the result of subtracting the equipment transfer amount from the formulation transfer amount is greater than zero, the result is used as the formulation net weight; otherwise, the formulation net weight is zero. For cases where only equipment edges exist within the same cluster number, the equipment net weight is the de-propagation weight of the equipment edge, and the formulation net weight is zero. For cases where only formulation edges exist, the formulation net weight is the de-propagation weight of the formulation edge, and the equipment net weight is zero. This generates a net weight table, which retains the cluster number, equipment segment, formulation segment, equipment net weight, and formulation net weight. In S4-4, the edge computing nodes generate attribution results based on the net weight table. For net weight records under the same equipment segment, the edge computing nodes read the equipment net weights from smallest to largest according to the batch number segment associated with the group number, and accumulate them item by item to obtain the equipment attribution value. For net weight records under the same formulation segment, the edge computing nodes read the formulation net weights in the same sorting manner, and accumulate them item by item to obtain the formulation attribution value. If the equipment net weight and formulation net weight of a certain group number are both zero, then the group number does not participate in the attribution value accumulation, but is retained in the net weight table as a record after being zeroed by the back edge or cross-suppression. The generated equipment attribution value and formulation attribution value continue to S5 to form the attribution difference and subsequent correction and diversion. In this implementation, S4 does not directly accumulate the equipment transfer amount and the formulation transfer amount. Instead, it first identifies the back edge in the propagation path, then resets the back edge weight to zero, and then suppresses the mutual amplification between the equipment side and the formulation side within the same group number. Finally, it recursively forms the equipment attribution value and the formulation attribution value according to the group number. This can avoid the same failure cluster from being repeatedly propagated between the equipment group edge and the formulation group edge and simultaneously enhancing the attribution on both sides. This allows the attribution difference read by S5 to come from the propagation result after de-back and cross-suppression. In practical applications: Group number G1 is associated with cavity A, the equipment segment with loading order 2, and the formula segment with formula parameter P3. S3 obtains an equipment transfer amount of 4 and a formula transfer amount of 1. Group number G2 is connected to the same equipment segment and was transferred from G1 through the equipment group edge in the previous round. If G2 returns to G1 through the formula group edge in this round, the propagation item is recorded as a return edge, and the propagation weight on the formula side is reset to zero. If both equipment group edges and formula group edges exist in G2, the equipment net weight or formula net weight is generated according to the difference between the equipment transfer amount and the formula transfer amount. The edge computing nodes then accumulate the equipment net weight under the same equipment segment according to the batch position number segment order of G1 and G2 to form the equipment attribution value, and accumulate the formula net weight under the same formula segment according to the same rule to form the formula attribution value.

[0021] S5. The attribution difference is generated by subtracting the formula attribution value from the equipment attribution value under the same group number. When the attribution difference is positive, an equipment correction record is generated based on the equipment segment and a formula blocking mark is written. When the attribution difference is negative, a formula correction record is generated based on the formula segment. When the attribution difference is zero, a formula blocking mark is written. In this implementation, S5 takes the equipment attribution value and formula attribution value obtained from S4 and splits the attribution results under the same group number; the edge computing node first reads the connected equipment segment and formula segment according to the group number, and then judges the anomaly by the attribution difference, which is more likely to be due to equipment loading factors, formula parameter factors, or the two cannot be separated; then, it generates equipment correction records, formula correction records, or formula blocking flags respectively, so that the formula parameters are not mistakenly modified when equipment factors are dominant or the attributions are parallel; this implementation process includes the following steps: In S5-1, the edge computing node uses the cluster number in the cluster transfer table as an index to read the equipment fragments and formula fragments associated with the same cluster number. Then, it reads the equipment attribution value under the same equipment fragment and the formula attribution value under the same formula fragment from the attribution results generated in S4. The edge computing node subtracts the formula attribution value from the equipment attribution value, and the difference is used as the attribution difference. When the attribution difference is positive, it indicates that the equipment fragment retains more evidence than the formula fragment after substitution, breakpoint refeedback, and non-backtracking propagation, and generates an equipment bit. When the attribution difference is negative, it indicates that the formula fragment retains more evidence than the equipment fragment, and generates a formula bit. When the attribution difference is zero, it indicates that the equipment fragment and the formula fragment do not form a unilateral dominant relationship under the current cluster number, and generates a parallel bit. The edge computing node merges the cluster number, equipment fragment, formula fragment, equipment attribution value, formula attribution value, attribution difference, and corresponding bit into a decision table. In S5-2, the edge computing node reads the device bit in the decision table and obtains the cavity number, loading order, and anomaly type based on the device segment, generating a device correction record. The device correction record includes at least the group number, device segment number, cavity number, loading order, anomaly type, device attribution value, and attribution difference, used to indicate that the current anomaly should be prioritized for device compensation or loading adjustment. For formula segments under the same group number, the edge computing node reads the formula parameters and formula segment number, generating a formula blocking flag. The formula blocking flag is bound to the group number, formula segment number, and formula parameters to prevent the formula parameters from entering the current round of formula correction records. No formula correction record is generated under the device bit, so that the same anomaly is not transferred to the formula parameter modification path when device attribution is dominant. In S5-3, the edge computing node reads the recipe bit in the decision table and obtains the recipe parameters and anomaly type based on the recipe fragment, generating a recipe correction record. The recipe correction record includes at least the group number, recipe fragment number, recipe parameters, anomaly type, recipe attribution value, and attribution difference, used to indicate that the current anomaly enters the recipe parameter correction path. For parallel bits in the decision table, the edge computing node does not generate equipment correction records and recipe correction records, but instead includes both equipment fragments and recipe fragments under the same group number in the recipe blocking flag. The records corresponding to the parallel bits will re-enter S2 to S5 after a new batch list is generated in the next mass production batch. The recipe blocking flag and the recipe correction record are mutually exclusive under the same group number, and the group number for which a recipe blocking flag has been generated will no longer generate a recipe correction record. In this implementation, equipment attribution values ​​and formula attribution values ​​are not directly used as manual judgment results. Instead, they are transformed into equipment bits, formula bits, and parallel bits through attribution differences, which trigger equipment correction records, formula correction records, or formula blocking flags, respectively. The equipment bit is used to prevent anomalies caused by cavity numbering or loading order from being rewritten as formula parameter problems. The formula bit is used to initiate parameter correction when formula factors are dominant. The parallel bit is used to pause formula modification and wait for the next batch of data to participate in attribution again. In practical applications: Under a certain group number G3, the equipment fragment associated with cavity A and loading order 2 obtains an equipment attribution value of 5, and the formula fragment associated with formula parameter P3 under the same group number obtains a formula attribution value of 2. The edge computing node calculates the attribution difference as 3 and generates the equipment bit, and then generates an equipment correction record containing cavity A and loading order 2, while generating a formula blocking flag for P3; if the equipment attribution value is 1 and the formula attribution value is 4 under another group number G4, the attribution difference is negative, and the edge computing node generates a formula correction record related to P3; if the attribution values ​​on both sides are equal under group number G5, only a formula blocking flag is generated, and no equipment correction record or formula correction record is output.

[0022] Working principle: First, edge computing nodes string together the cavity number, loading order, recipe parameters, electrical test failure coordinates, and anomaly type of the 6-inch SiC MOSFET mass production batch into a batch linked list. Then, equipment segments, recipe segments, and failure clusters are extracted from this list. Subsequently, using the state transfer concept of the Markov model, the three are grouped into a ternary cluster. Through substitution, breakpoint, backfeed, and non-backtracking propagation, it is determined whether the anomaly is retained along the cavity and loading order or along the recipe parameters. Finally, equipment attribution values ​​and recipe attribution values ​​are generated. If equipment factors are dominant, equipment correction records are output and recipe modifications are blocked. If recipe factors are dominant, recipe correction records are output, thereby avoiding mistaking equipment execution deviations for recipe parameter problems. In practical applications, if edge leakage anomalies are detected by electrical testing after the same batch of 6-inch SiC MOSFET wafers have undergone oxidation, annealing, or metallization, the edge computing nodes will first check whether these anomalies are concentrated in a certain cavity or loading order, or whether they still follow the same formula parameters in multiple cavities and loading orders. For example, if the same type of edge leakage repeatedly occurs in cavity A and loading order 2 after the formula parameters are changed, the system tends to determine that it is due to equipment or loading factors, and will first generate correction records related to cavity A or loading order, without directly modifying the oxidation or annealing formula. If the same formula parameters still show the same type of anomaly across different cavities and loading orders, the system will switch to the formula correction path to make the parameter adjustments for the next batch closer to the real cause.

[0023] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing SiC MOSFET mass production process parameters to suit 6-inch production lines, characterized in that, include: S1. In the edge computing node, with the wafer number as the chain head, establish a wafer chain record for the cavity number, loading order, formula parameters, failure coordinates and anomaly type within the 6-inch mass production batch, and assign batch tag numbers according to the process end time and loading order to generate a batch chain list. Using the wafer number as the chain head, the cavity number, loading order, formula parameters and process end time are assembled into a process row, and a chain position number is generated according to the order of the process end times within the same wafer number to obtain the wafer process chain; Using the wafer number as an index, the failure coordinates and anomaly types are associated with the process line at the end of the chain number in the wafer process chain, and process lines without associated failure coordinates are marked as empty failure lines, thus obtaining the wafer failure chain; The wafer failure chain is rearranged according to the process end time. If the process end times are the same, the chain is rearranged according to the loading order. A batch tag number is generated based on the rearranged row number to obtain a batch linked list. S2. Based on the batch linked list, wafer records with the same device key and adjacent anomalies across recipes are grouped into device segments, wafer records with the same recipe and similar anomalies across cavities and loading are grouped into recipe segments, and adjacent failure coordinates in the same set are grouped into failure clusters to generate a segment table. Remove empty failure rows from the batch list, combine the cavity number, loading order, and anomaly type into a device key, combine the recipe parameters and anomaly type into a recipe key, combine the failure coordinates and anomaly type into a coordinate key, and retain the batch tag number to generate an anomaly row table. For abnormal rows with the same device key, read them in ascending order of batch tag number. Abnormal rows with adjacent batch tag numbers and changing recipe keys are merged into the same device segment. Abnormal rows with the same recipe key and disconnected batch tag numbers are cut out of the device segment to generate a device segment row. S3. Perform Kikuchi substitution group operation on the fragment table to form a ternary group of equipment fragments, recipe fragments and failure clusters under the same anomaly type. Generate equipment transfer amount based on recipe replacement breaks after maintaining cavity number and loading order, and generate recipe transfer amount based on cavity number and loading order replacement breaks after maintaining recipe parameters, and generate group transfer table. S4. Perform non-backward confidence propagation operation on the cluster transfer table, generate an inbound edge table with cluster number, cluster edge direction and previous cluster number, set the reverse back edge to zero, and perform cross suppression on the equipment transfer amount and formula transfer amount within the same cluster number to generate equipment attribution value and formula attribution value. S5. The attribution difference is generated by subtracting the formula attribution value from the equipment attribution value under the same group number. When the attribution difference is positive, an equipment correction record is generated based on the equipment segment and a formula blocking mark is written. When the attribution difference is negative, a formula correction record is generated based on the formula segment. When the attribution difference is zero, a formula blocking mark is written.

2. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 1, characterized in that: S2 further includes: S2-3. For abnormal rows with the same recipe key, merge abnormal rows that cross cavity numbers, cross loading order and have the same abnormal type into the same recipe segment, and exclude abnormal rows that fall only into a single device key from the recipe segment to generate a recipe segment line. S2-4. Expand the failure coordinates according to the coordinate key, merge the failure coordinates that are adjacent and whose wafer number sets fall into the equipment segment line or recipe segment line into failure clusters, and generate a segment table according to the intersection of wafer number sets.

3. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 2, characterized in that: S3 includes: S3-1. Using the same anomaly type in the fragment table as the group number, read the equipment fragment, recipe fragment and failure cluster that have an intersection in the wafer number set. Combine the cavity number, loading order and wafer number set in the equipment fragment, the recipe parameters and wafer number set in the recipe fragment, and the failure coordinate and wafer number set in the failure cluster into a ternary group, and take the intersection of the wafer number sets of the three to generate the wafer set within the group. S3-2. Perform Kikuchi substitution operation on the ternary group, arrange the wafer set in the group according to the batch number to form the group position string, keep the cavity number and loading order in the group position string unchanged and replace the position string of the formula parameter to generate the equipment substitution string, keep the formula parameter in the group position string unchanged and replace the position string of the cavity number and loading order to generate the formula substitution string.

4. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 3, characterized in that: S3 further includes: S3-3. Read the failure coordinates bit by bit along the equipment replacement string. If the failure coordinates after replacing the formula parameters still fall into the failure cluster, mark this position as the equipment reserved position. If the failure coordinates after replacing the formula parameters leave the failure cluster, mark this position as the equipment cut-off position and truncate the equipment reserved positions after this position to generate the equipment cut-off string. S3-4. Read the failure coordinates bit by bit along the formula substitution string. If the failure coordinates after replacing the cavity number and loading order still fall into the failure cluster, mark this position as a formula retention position. If the failure coordinates after replacing the cavity number and loading order leave the failure cluster, mark this position as a formula break and truncate the formula retention positions after this position to generate a formula truncation string.

5. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 4, characterized in that: S3 further includes: S3-5. Perform the expected propagation breakpoint re-feedback operation on the equipment truncated string and the recipe truncated string. Re-feed the equipment reservation in the equipment truncated string to the equipment segment according to the batch tag number, and re-feed the recipe reservation in the recipe truncated string to the recipe segment according to the batch tag number. If the same batch tag number has both equipment reservation and recipe reservation, a conflict bit is generated; otherwise, a one-way reservation is generated. S3-6. Perform non-backtracking deduction operation on conflict bits and one-way reservation bits, synchronously delete conflict bits from equipment reservation bits and recipe reservation bits, generate equipment net strings from the reserved equipment reservation bits, and generate recipe net strings from the reserved recipe reservation bits. S3-7. Generate the device transfer amount from the bit length of the device net string and the formula transfer amount from the bit length of the formula net string. Then, merge the group number, the fault type, the wafer set within the group, the device transfer amount, and the formula transfer amount to generate the group transfer table.

6. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 5, characterized in that: S4 includes: S4-1. Read the equipment transfer amount and formula transfer amount of the same ternary group using the group number in the group transfer table. Record the edge from the ternary group to the equipment segment as the equipment group edge and the edge from the ternary group to the formula segment as the formula group edge. Generate the inbound edge table with the group number, the group edge direction and the previous group number. S4-2. Perform non-backtracking shadow edge operation on the inbound edge table. Record the propagation items where the previous cluster number is swapped with the current cluster number and the cluster edge direction is opposite as back edges. Reset the propagation weight of the back edges to zero. Retain the original propagation weights of the remaining propagation items and generate a de-back weight table.

7. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 6, characterized in that: S4 further includes: S4-3. Perform cross-suppression operation based on the weight table. When there are both equipment group edge and formulation group edge in the same group number, generate equipment net weight by subtracting the positive value of formulation transfer amount from equipment transfer amount, and generate formulation net weight by subtracting the positive value of equipment transfer amount from formulation transfer amount, and generate net weight table. S4-4. Based on the net weight table, the net weight of equipment under the same equipment segment is recursively accumulated by group number to generate the equipment attribution value, and the net weight of formula under the same formula segment is recursively accumulated by group number to generate the formula attribution value.

8. The method for optimizing SiC MOSFET mass production process parameters to suit a 6-inch production line according to claim 7, characterized in that: S5 includes: S5-1. The attribution difference is generated by subtracting the formula attribution value from the equipment attribution value under the same group number. When the attribution difference is positive, the equipment position is generated; when the attribution difference is negative, the formula position is generated; when the attribution difference is zero, the parallel position is generated. A decision table is generated from the group number, equipment segment, formula segment, and attribution difference. S5-2. For the equipment position in the decision table, generate equipment correction records based on the cavity number, loading order and abnormality type in the equipment segment, and generate formula blocking markers for formula segments and formula parameters under the same group number. S5-3. For the recipe position in the decision table, generate a recipe correction record based on the recipe parameters and anomaly type in the recipe segment; for the parallel positions in the decision table, write both the equipment segment and the recipe segment under the same group number into the recipe blocking flag.

Images