Search

CN-122022435-A - Full-flow intelligent management and control system and method for SiC device production

CN122022435ACN 122022435 ACN122022435 ACN 122022435ACN-122022435-A

Abstract

The invention discloses a full-flow intelligent control system and method for SiC device production, and belongs to the technical field of flow control. The method comprises the steps of integrating core data of each process of upstream and midstream to construct a full-flow time sequence data model, disassembling the full-flow time sequence data model, separating data sequences including trend item sequences, periodic item sequences and residual item sequences, collecting batch parameters and quantifying, constructing a batch initial scheduling frame based on a genetic algorithm by combining the sequences and the batch parameters to obtain a scheduling scheme, introducing an analytic hierarchy process to optimize the scheduling scheme to obtain an optimized scheduling scheme, aligning the optimized scheduling scheme with a graph structure time sequence data model in time sequence to obtain a scheduling table, and carrying out full-flow control on SiC device production based on the scheduling table. According to the invention, time sequence fine disassembly and parameter deep fusion are carried out, the scheduling basis is tamped, and the feasibility of a scheduling scheme is improved by adopting double-algorithm collaborative optimization and time sequence alignment.

Inventors

  • LI YIZHE
  • WU YUXUAN

Assignees

  • 南京宽能半导体有限公司

Dates

Publication Date
20260512
Application Date
20260224

Claims (10)

  1. 1. The full-flow intelligent control method for SiC device production is characterized by comprising the following steps of: Integrating core data of each procedure of upstream and midstream to construct a full-flow time sequence data model; Disassembling the full-flow time sequence data model, and separating the sequence of data, including a trend item sequence, a period item sequence and a residual item sequence; The method comprises the steps of combining sequence and batch parameters, constructing a batch initial scheduling frame based on a genetic algorithm to obtain a scheduling scheme, introducing an analytic hierarchy process to optimize the scheduling scheme to obtain an optimized scheduling scheme, and carrying out time sequence alignment on the optimized scheduling scheme and a time sequence data model of a graph structure to obtain a scheduling table; and performing full-flow control on SiC device production based on a schedule.
  2. 2. The full-flow intelligent control method for SiC device production according to claim 1, wherein the integrating the core data of each of the upstream and midstream processes constructs a full-flow time series data model, comprising: The core data of the upstream process includes substrate raw material purity Substrate crystalline integrity And substrate slice thickness tolerance Core data of the midstream process includes epitaxial layer thickness uniformity Alignment accuracy of lithography patterns And etch depth uniformity Will (i) be Association with , Association with , Association with , Association with , Association with , Association with , Association with ; Defining a node system, an edge system and an attribute system, constructing a graph structure, and integrating full-flow data based on the graph structure to obtain a full-flow time sequence data model.
  3. 3. The full-flow intelligent control method for SiC device production according to claim 2, wherein the defining a node system, an edge system and an attribute system, constructing a graph structure, performing full-flow data integration based on the graph structure, and obtaining a full-flow time sequence data model, includes: defining a node system by using process nodes and data nodes, wherein the upstream process nodes comprise raw material purification Crystal growth And slicing processing The midstream process node comprises epitaxial growth Photolithography process And etching process The upstream data node comprises 、 And The data node of midstream comprises 、 And ; Defining an edge system by adopting a time sequence edge, a home edge and a related edge, wherein the time sequence edge is connected with adjacent process nodes, the home edge is connected with a data node and a corresponding generated process node, and the related edge comprises a related edge in a process and a related edge corresponding to a cross process; In the attribute system, the process node attribute comprises a process ID, a standard execution duration and a pre-process ID, and the data node attribute comprises a data ID, an acquisition time stamp, a numerical range and an acquisition equipment ID; Constructing a graph structure, carrying out node instantiation and data mounting, aligning the acquisition time stamp of each data node with the execution time window of the corresponding process node, and sequencing the whole-flow data according to the production time sequence through the time sequence priority attribute of the time sequence edge, so that the graph structure simultaneously has data association logic and production time sequence logic, and a whole-flow time sequence data model is formed.
  4. 4. The full-flow intelligent control method for SiC device production according to claim 1, wherein the disassembling the full-flow time series data model, separating the sequence of data, including the trend term sequence, the period term sequence and the residual term sequence, includes: based on the full-flow time sequence data model, extracting a time sequence data set, wherein the time sequence data set comprises a procedure node time sequence data set and a data node time sequence data set, unifying time granularity and carrying out dimension alignment; The method comprises the steps of setting a fixed-length time window for time series data by adopting sliding window fitting logic, carrying out linear or nonlinear fitting on the data in the window, combining fitting results of all windows to form a continuous trend item sequence, removing the extracted trend items from a time series data set to obtain time series data after trend removal, determining the period length based on the inherent beat of a production process, carrying out period pattern matching on the time series data after trend removal, extracting repeated fluctuation components to form a period item sequence, subtracting the trend item sequence and the period item sequence by taking the original time series data as a reference, and taking the obtained difference as a residual error item sequence.
  5. 5. The full-flow intelligent control method for SiC device production according to claim 1, wherein the combining sequence and batch parameters, based on genetic algorithm, constructs a batch initial schedule framework, resulting in a schedule scheme, comprising: Extracting sequence characteristics, namely extracting a long-term change trend value of execution duration of each procedure and a numerical drift trend of each core data in a trend item sequence as procedure time-consuming reference correction basis, extracting periodic fluctuation amplitude of each procedure as a procedure time-consuming fluctuation constraint boundary based on inherent beats of a production process in a periodic item sequence, counting variance ranges of residuals of each procedure node and data node in a residual item sequence, screening a stable time sequence interval with the residuals smaller than a threshold value as a priority time window reference of batch scheduling, and establishing a corresponding relation between the sequence characteristics and batch parameters; The method comprises the steps of adopting a real number coding mode to enable chromosomes to correspond to a batch scheduling scheme one by one, defining constraint conditions of a genetic algorithm, fusing sequence characteristics and batch parameters to construct a multi-target weighting function by taking a maximized fitness value as a target, generating an initial population, carrying out selection operation, crossover operation and mutation operation, setting iteration termination conditions, and taking the chromosome with the highest fitness value in a final population as the scheduling scheme.
  6. 6. The full-flow intelligent control method for SiC device production according to claim 5, wherein the defining constraints of the genetic algorithm includes: constraint conditions of the genetic algorithm comprise hard constraint and soft constraint, wherein the hard constraint represents that the constraint must be met, and the soft constraint represents optimization target guidance; The hard constraint of the genetic algorithm comprises a time sequence constraint, a delivery constraint and a dependence constraint, wherein the time sequence constraint is that a batch process needs to follow the time sequence edge logic of a graph structure, a preceding process cannot start a subsequent process, the process duration cannot exceed the superposition range of a trend item correction reference and a periodic item fluctuation range; The soft constraint of the genetic algorithm comprises a priority constraint and a stability constraint, wherein the priority constraint is specifically a batch combination with higher process connection priority, a continuous time window is preferentially distributed, a batch with higher emergency grade weight is delivered, a time window with stable residual error is preferentially occupied, the stability constraint is specifically a cycle term law to be attached to the fluctuation range of the batch process duration, and the influence of the residual error term is controlled within an allowable range.
  7. 7. The full-flow intelligent control method for SiC device production according to claim 1, wherein the introducing a hierarchical analysis optimizing scheduling scheme to obtain an optimizing scheduling scheme includes: The method comprises the steps of taking an optimized SiC device production batch scheduling scheme as a core, constructing a target layer, a criterion layer, a sub-criterion layer and a scheme layer by combining constraint conditions of a genetic algorithm and sequence characteristics, strongly associating each layer with earlier data, ensuring that each layer corresponds to a full-flow time sequence data model and constraint conditions of the genetic algorithm, respectively constructing a criterion layer pair target layer judgment matrix, a sub-criterion layer pair criterion layer judgment matrix and a scheme layer sub-criterion layer judgment matrix by adopting a nine-level scale method based on actual production requirements and earlier data characteristics, carrying out weight calculation and consistency check, determining priority ordering of elements of each sub-criterion layer to elements of the corresponding criterion layer based on normalized weights, ordering of elements of each scheme layer to elements of the corresponding sub-criterion layer, combining weights of each layer, calculating comprehensive weights of candidate schemes of the scheme layer to the target layer, and taking the scheme with highest comprehensive weights as an optimized scheduling scheme.
  8. 8. The full-flow intelligent control method for SiC device production according to claim 1, wherein the performing time alignment of the optimized scheduling scheme with the graph structure time sequence data model to obtain the schedule includes: The method comprises the steps of according to a time sequence edge logic of a graph structure, checking the sequence and duration of a scheduling procedure, aligning a data node acquisition time stamp with a corresponding procedure time window based on a home edge, checking the time sequence engagement rationality of cross-procedure data based on an associated edge, checking time sequence conflict, correcting deviation according to hard constraint, controlling the influence of residual items in an allowable range, ensuring the scheme compliance after alignment, and summarizing the aligned time window, the process engagement relation, the delivery threshold and the stability index to generate a scheduling table.
  9. 9. The full-process intelligent control method for SiC device production according to claim 1, wherein the full-process control of SiC device production based on the schedule comprises: Dividing a schedule according to upstream and downstream procedures, synchronizing the schedule to each station, binding the time sequence edge of a graph structure with the constraint of an associated edge to ensure the alignment of an execution end to a control reference, collecting the execution data of each procedure node and the monitoring value of the data node in real time based on the attribution edge of the graph structure, comparing the execution data with the time sequence window of the schedule, synchronously tracking a trend item sequence, a period item sequence and a residual item sequence, checking the compliance of procedure engagement and cross-procedure data association, triggering early warning based on the confidence of the associated edge when the data deviate from a scheduling threshold value or time sequence logic conflict, and comparing actual data with the schedule after each batch of procedures are completed to update the time sequence model parameters of the graph structure.
  10. 10. A full-flow intelligent control system for SiC device production, using the full-flow intelligent control method for SiC device production according to any one of claims 1 to 9, characterized by comprising: the model building module comprises a data integration unit, a model building unit and a data processing unit, wherein the data integration unit integrates core data of each procedure of upstream and midstream; The model disassembly and parameter quantization module comprises a model disassembly unit, a parameter quantization unit, a parameter analysis unit and a data analysis unit, wherein the model disassembly unit disassembles a full-flow time sequence data model, separates data sequences including a trend item sequence, a period item sequence and a residual item sequence, and collects batch parameters and quantizes the batch parameters; The scheduling table generating module comprises a scheduling scheme constructing unit, a scheduling scheme optimizing unit, a scheduling table generating unit and a scheduling table generating unit, wherein the scheduling scheme constructing unit combines sequences and batch parameters, and constructs a batch initial scheduling frame based on a genetic algorithm to obtain a scheduling scheme; the full-flow control module comprises a full-flow control unit for carrying out full-flow control on SiC device production based on a schedule.

Description

Full-flow intelligent management and control system and method for SiC device production Technical Field The invention relates to the technical field of flow control, in particular to a full-flow intelligent control system and method for SiC device production. Background SiC, as a third generation wide bandgap semiconductor material, has excellent characteristics of large bandgap width, high thermal conductivity, high breakdown field strength, high electron saturation drift speed, and the like, and has a continuously rising application demand in the field of high-end equipment such as new energy automobiles, photovoltaic inverters, rail transit, smart grids, and the like. The production flow of the SiC device has the remarkable characteristics of long process chain, high process complexity and strong parameter relevance, and the core covers two major links of upstream substrate preparation and midstream device manufacture, and the process parameters and production time sequence of each process directly determine the performance and yield of the SiC device. In the prior art, data acquisition is carried out on single links of upstream substrate preparation or midstream device manufacture in SiC device production, upstream and downstream data are mutually split, and a unified integration frame is not provided. In the prior art, the analysis of production data stays on the surface, the trend, the period and the fine disassembly of residual error items are not carried out on the time series data, and the fluctuation rule of the working procedure duration, the data drift characteristic and the random interference range cannot be accurately captured. In the prior art, a single algorithm is adopted to construct a scheduling scheme, a multi-algorithm cooperative optimization mechanism is lacked, the single algorithm is difficult to consider multi-target requirements such as delivery time, process priority, time sequence stability and the like, and the scheduling scheme is easy to be disjointed from the actual production constraint. Disclosure of Invention The invention aims to provide a full-flow intelligent control system and method for SiC device production, which are used for solving the problems in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: In a first aspect, the present application provides a full-flow intelligent control method for SiC device production, comprising the steps of: Integrating core data of each procedure of upstream and midstream to construct a full-flow time sequence data model; Disassembling the full-flow time sequence data model, and separating the sequence of data, including a trend item sequence, a period item sequence and a residual item sequence; The method comprises the steps of combining sequence and batch parameters, constructing a batch initial scheduling frame based on a genetic algorithm to obtain a scheduling scheme, introducing an analytic hierarchy process to optimize the scheduling scheme to obtain an optimized scheduling scheme, and carrying out time sequence alignment on the optimized scheduling scheme and a time sequence data model of a graph structure to obtain a scheduling table; and performing full-flow control on SiC device production based on a schedule. With reference to the first aspect, in a first implementation manner of the first aspect of the present application, the integrating core data of each of the upstream and midstream processes constructs a full-flow time-series data model, including: The core data of the upstream process includes substrate raw material purity Substrate crystalline integrityAnd substrate slice thickness toleranceCore data of the midstream process includes epitaxial layer thickness uniformityAlignment accuracy of lithography patternsAnd etch depth uniformityWill (i) beAssociation with,Association with,Association with,Association with,Association with,Association with,Association with; Defining a node system, an edge system and an attribute system, constructing a graph structure, and integrating full-flow data based on the graph structure to obtain a full-flow time sequence data model. With reference to the first aspect, in a second implementation manner of the first aspect of the present application, the defining a node system, an edge system, and an attribute system, constructing a graph structure, and performing full-flow data integration based on the graph structure to obtain a full-flow time sequence data model, where the method includes: defining a node system by using process nodes and data nodes, wherein the upstream process nodes comprise raw material purification Crystal growthAnd slicing processingThe midstream process node comprises epitaxial growthPhotolithography processAnd etching processThe upstream data node comprises、AndThe data node of midstream comprises、And; Defining an edge system by adopting a time sequence edge, a home edge and a related edge,