A multi-scale thermal parameter incremental updating method and system for layout ECO

Abstract

The application discloses a multi-scale thermal parameter incremental updating method and system for layout ECO, and solves the technical problem of serious waste of computing resources caused by the existing multi-scale thermal parameter incremental updating method. The method comprises the following steps: obtaining an old layout file, a new layout file, an old multi-scale thermal parameter database and PDK process data, and outputting layout change segment data through layout difference detection; sequentially outputting a dirty window candidate set through multi-scale window mapping, outputting an affected window set through thermal influence propagation analysis, outputting normal and high-risk dirty windows through dirty window hierarchical judgment, and then outputting compatible format new thermal parameter data through differential thermal parameter updating, cross-scale thermal parameter updating and compatible packaging.

Description

Technical Field

[0001] This invention relates to the field of integrated circuit technology, and in particular to a multi-scale thermal parameter incremental update method and system for layout ECO. Background Technology

[0002] With the increasing prevalence of advanced processes and packaging technologies such as 3D stacking, back-side power supply, and hybrid bonding, the heat flow paths within chips are becoming increasingly complex in terms of materials, conduction direction, spatial scale, and interface characteristics. Referring to publicly available industrial data from leading companies like Cadence and Synopsys, it's evident that applications such as 3D integrated chips and multi-chip stacking place higher demands on thermal integrity analysis capabilities. Anisotropic thermal property modeling and the construction of thermal models in conjunction with layout design have become key technological directions for current industry development.

[0003] Based on the aforementioned cutting-edge technological development trends, the nature of chip thermal-related issues has also undergone a fundamental transformation. They are no longer marginal post-processing auxiliary issues in the chip design process, but rather core issues that directly determine chip performance, operational reliability, and design approval convergence efficiency. This has made multi-scale thermal parameter update technology, which is adapted to complex thermal scenarios, an indispensable key technology in the current chip thermal analysis process.

[0004] Existing multi-scale incremental update methods for thermal parameters mostly treat thermal parameter generation as a one-time batch processing task, rather than a data maintenance process that evolves continuously with each layout version. Whether it's importing 3D solutions from ECAD, online homogenization of local RVEs (Representative Volume Elements), or rapid prediction using machine learning, the typical workflow is to output the current parameters given the current layout, rarely discussing how to reuse the calculation results of the previous version when the next layout arrives. As a result, whenever an ECO (Engineering Change Order) occurs in the layout, the most common practice in engineering is often a global or large-scale recalculation, leading to a serious waste of computational resources. Summary of the Invention

[0005] This invention provides a multi-scale incremental update method and system for thermal parameters oriented to layout ECO, which solves the technical problem that existing multi-scale incremental update methods for thermal parameters lead to serious waste of computing resources.

[0006] The first aspect of this invention provides a multi-scale incremental update method for thermal parameters oriented towards layout ECO, comprising:

[0007] Acquire old layout files, new layout files, old multi-scale thermal parameter database, and PDK process data; and perform layout difference detection based on the old layout files and the new layout files, outputting layout change segment data.

[0008] Based on the map change fragment data and the window index table in the old multi-scale thermal parameter database, multi-scale window mapping is performed to output a dirty window candidate set.

[0009] Using preset influence propagation rules, thermal influence propagation analysis is performed on the dirty window candidate set based on the window index table in the old version of the multi-scale thermal parameter database, and the affected window set is output.

[0010] The affected window set is classified and judged according to the new map file, the map change fragment data and the window index table, thermal parameter table and version metadata table in the old multi-scale thermal parameter database, and the preset classification and judgment rules are used to output ordinary dirty windows and high-risk dirty windows.

[0011] A preset fast update model is used to update the thermal parameters of the ordinary dirty window and the high-risk dirty window respectively based on the new version of the drawing file, the old version of the multi-scale thermal parameter database and the PDK process data, and outputs the dirty window updated thermal parameter set.

[0012] Using preset aggregation rules and preset database replacement rules, cross-scale thermal parameter updates and compatibility encapsulation are performed based on the dirty window update thermal parameter set, the old version of the multi-scale thermal parameter database, and the window index table in the old version of the multi-scale thermal parameter database, and the new version of thermal parameter data in a compatible format is output.

[0013] Optionally, the step of performing map difference detection based on the old map file and the new map file, and outputting map change fragment data, includes:

[0014] By comparing the old map file with the new map file according to the map layer level, region or graphic element, the changed areas corresponding to multiple preset difference types are identified;

[0015] Extract the map change fragment data for each of the changed regions.

[0016] Optionally, the step of performing multi-scale window mapping based on the map change fragment data and the window index table in the old multi-scale thermal parameter database, and outputting a dirty window candidate set, includes:

[0017] Based on the window index table, locate the fine-scale window that intersects with each of the map change fragment data, and mark each fine-scale window as a direct change fine window;

[0018] Based on the parent-child index of the window index table, trace upwards to the parent window corresponding to the directly changing fine window and mark it as a candidate dirty parent window;

[0019] Based on the cross-layer association index of the window index table, locate the window that is directly changed and cross-layer associated with the fine window, and mark it as a candidate dirty cross-layer window;

[0020] Integrate the directly changing fine window, the candidate dirty parent window, and the candidate dirty cross-layer window to output a dirty window candidate set.

[0021] Optionally, the preset impact propagation rules include spatial neighborhood rules, cross-layer rules, scale rules, thermal channel rules, and risk propagation rules; the preset impact propagation rules are used to perform thermal impact propagation analysis on the dirty window candidate set according to the window index table in the old version of the multi-scale thermal parameter database, and the affected window set is output, including:

[0022] Based on the adjacent window list in the window index table, adjacent windows that conform to the spatial neighborhood rules are filtered and included in the propagation change window set;

[0023] Based on the cross-layer association index of the window index table, select the upper and lower layer hot path connection windows that meet the cross-layer rules and include them in the propagation change window set;

[0024] Based on the parent-child indexes of the window index table, parent windows and higher-level parent windows that meet the scaling rules are selected and included in the aggregated update window set;

[0025] Based on the local dominant hot channel association information in the window index table, connected windows that conform to the hot channel rules are selected and included in the propagation change window set;

[0026] Identify windows belonging to the preset high-heat-sensitive region in the dirty window candidate set, supplement and filter connected windows according to the risk expansion rule, and include them in the propagation change window set;

[0027] The dirty window candidate set, the propagated change window set, and the aggregated update window set are merged to obtain the merged window set;

[0028] Remove duplicate windows from the merged window set and output the affected window set.

[0029] Optionally, the step of using preset classification rules to perform dirty window classification processing on the affected window set based on the new map file, the map change fragment data, and the window index table, thermal parameter table, and version metadata table in the old multi-scale thermal parameter database, and outputting ordinary dirty windows and high-risk dirty windows, includes:

[0030] Based on the map change fragment data, extract the change intensity data corresponding to each affected window in the affected window set;

[0031] Extract the structural complexity data corresponding to each affected window based on the window index table and the local geometric data of the new map in the new map file;

[0032] Extract the thermal parameter confidence data corresponding to each affected window based on the thermal parameter table and the version metadata table;

[0033] Based on the preset classification and determination rules, the multiple affected windows are divided into ordinary dirty windows and high-risk dirty windows by using the change intensity, structural complexity, and thermal parameter confidence data corresponding to each affected window.

[0034] Optionally, the step of using a preset fast update model to update the thermal parameters of the ordinary dirty window and the high-risk dirty window respectively based on the new version drawing file, the old version multi-scale thermal parameter database, and the PDK process data, and outputting a dirty window updated thermal parameter set, includes:

[0035] Extract the local geometric difference feature data corresponding to the ordinary dirty window from the new version of the drawing file, and read the reference thermal parameter data corresponding to the ordinary dirty window from the old version of the multi-scale thermal parameter database;

[0036] The feature parameters are updated based on the local geometric difference feature data to obtain the updated feature parameters;

[0037] The updated feature parameters and the baseline thermal parameter data are input into the preset fast update model to calculate the new thermal parameter data corresponding to the ordinary dirty window.

[0038] Extract the local geometric data corresponding to the high-risk dirty window from the new drawing file, and construct a local three-dimensional high-precision model corresponding to the high-risk dirty window by combining it with the PDK process data;

[0039] Preset boundary conditions are applied to the local three-dimensional high-precision model, and the local heat flow response data is obtained by solving the problem.

[0040] Based on the local heat flow response data, the updated equivalent thermal conductivity tensor and associated thermal parameter data of the high-risk dirty window are calculated by inversion.

[0041] Integrate the new thermal parameter data of the ordinary dirty window, the updated equivalent thermal conductivity tensor and the corresponding thermal parameter data of the high-risk dirty window, and output the updated thermal parameter set of the dirty window.

[0042] Optionally, the step of using preset aggregation rules and preset database replacement rules to perform cross-scale thermal parameter updates and compatibility encapsulation based on the dirty window update thermal parameter set, the old version of the multi-scale thermal parameter database, and the window index table in the old version of the multi-scale thermal parameter database, and outputting new version thermal parameter data in a compatible format, includes:

[0043] Based on the parent-child index of the window index table and the set of dirty window update hot parameters, the parent window to be passively updated is determined.

[0044] Read the historical thermal parameter data of the unupdated child windows under the parent window to be passively updated from the old multi-scale thermal parameter database;

[0045] The historical thermal parameter data of the unupdated child window is combined with the new thermal parameter data of the updated child window in the dirty window update thermal parameter set, and the complete update thermal parameter data of the parent window to be passively updated is calculated using the preset aggregation rule.

[0046] According to the preset database replacement rules, the dirty window update thermal parameter set and the complete updated thermal parameter data are used to perform local replacement and parameter update on the corresponding window records in the old version of the multi-scale thermal parameter database to obtain the new version of the multi-scale thermal parameter database.

[0047] The complete data of the new multi-scale thermal parameter database is read according to the preset downstream compatible interface format, and the complete data is adapted and encapsulated to output the new thermal parameter data in a compatible format.

[0048] A second aspect of the present invention provides a multi-scale incremental update system for thermal parameters oriented to layout ECO, comprising:

[0049] The acquisition module is used to acquire old layout files, new layout files, old multi-scale thermal parameter databases, and PDK process data, and perform layout difference detection based on the old layout files and the new layout files, and output layout change segment data.

[0050] The mapping module is used to perform multi-scale window mapping based on the map change fragment data and the window index table in the old version of the multi-scale thermal parameter database, and output a dirty window candidate set.

[0051] The analysis module is used to perform thermal influence propagation analysis on the dirty window candidate set according to the window index table in the old multi-scale thermal parameter database using preset influence propagation rules, and output the set of affected windows;

[0052] The grading module is used to perform dirty window grading and judgment processing on the affected window set according to the new map file, the map change fragment data and the window index table, thermal parameter table and version metadata table in the old multi-scale thermal parameter database, and output ordinary dirty windows and high-risk dirty windows.

[0053] The update module is used to update the thermal parameters of the ordinary dirty window and the high-risk dirty window respectively according to the new version drawing file, the old version multi-scale thermal parameter database and the PDK process data using a preset fast update model, and output the dirty window updated thermal parameter set;

[0054] The encapsulation module is used to perform cross-scale thermal parameter updates and compatibility encapsulation based on the dirty window update thermal parameter set, the old version of the multi-scale thermal parameter database, and the window index table in the old version of the multi-scale thermal parameter database, using preset aggregation rules and preset database replacement rules, and output new version thermal parameter data in a compatible format.

[0055] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the multi-scale thermal parameter incremental update method for layout-oriented ECO as described above.

[0056] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, it implements the multi-scale thermal parameter incremental update method for layout-oriented ECO as described above.

[0057] As can be seen from the above technical solutions, the present invention has the following advantages:

[0058] The above-mentioned technical solution of the present invention provides a multi-scale incremental update method for thermal parameters for layout ECO. This method acquires old layout files, new layout files, an old multi-scale thermal parameter database, and PDK process data. Based on the old and new layout files, it performs layout difference detection and outputs layout change fragment data. Multi-scale window mapping is performed based on the layout change fragment data and the window index table in the old multi-scale thermal parameter database, outputting a dirty window candidate set. A preset influence propagation rule is used to perform thermal influence propagation analysis on the dirty window candidate set based on the window index table in the old multi-scale thermal parameter database, outputting an affected window set. A preset hierarchical judgment rule is used to classify the affected windows based on the new layout file, layout change fragment data, and the window index table, thermal parameter table, and version metadata table in the old multi-scale thermal parameter database. The system performs a dirty window classification process, outputting ordinary dirty windows and high-risk dirty windows. A pre-defined fast update model is used to update the thermal parameters of ordinary and high-risk dirty windows based on the new layout file, the old multi-scale thermal parameter database, and PDK process data, outputting a set of updated thermal parameters for dirty windows. Pre-defined aggregation rules and database replacement rules are used to perform cross-scale thermal parameter updates and compatibility encapsulation based on the updated thermal parameter set for dirty windows, the old multi-scale thermal parameter database, and the window index table in the old multi-scale thermal parameter database, outputting new thermal parameter data in a compatible format. Based on the above scheme, this invention first extracts layout change fragment data through layout difference detection to accurately locate the layout change area. Then, it combines the window index table to complete multi-scale window mapping, obtaining a candidate set of dirty windows, and only filtering out windows associated with layout changes. Subsequently, thermal impact propagation analysis is used for further filtering to determine the set of affected windows linked to layout changes, eliminating windows that do not require thermal parameter updates. This invention classifies affected windows into ordinary dirty windows and high-risk dirty windows, thereby matching corresponding thermal parameter update logic for different types of windows. In the thermal parameter update stage, existing benchmark data from the old multi-scale thermal parameter database is directly called, and incremental parameter updates are achieved based on the existing calculation results. Finally, cross-scale thermal parameter updates and compatibility encapsulation are carried out. Local thermal parameter aggregation is performed only for the changed window and its associated parent window. At the same time, the corresponding window records in the old multi-scale thermal parameter database are partially replaced, completely preserving the original content in the database that has not been changed, without the need for a complete reconstruction of the entire database. This stage also simultaneously completes data format adaptation and encapsulation, saving the additional computational overhead caused by subsequent separate format conversion, effectively reducing invalid computation tasks and overall computational load, thereby avoiding waste of computing resources. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating the steps of a multi-scale incremental update method for layout ECO provided in Embodiment 1 of the present invention.

[0061] Figure 2 This is a flowchart illustrating a multi-scale incremental update method for layout ECO provided in Embodiment 1 of the present invention.

[0062] Figure 3 This is an overall framework diagram of a multi-scale thermal parameter incremental update method for layout ECO provided in Embodiment 1 of the present invention;

[0063] Figure 4 This is a structural block diagram of a multi-scale thermal parameter incremental update system for layout ECO provided in Embodiment 2 of the present invention. Detailed Implementation

[0064] This invention provides a multi-scale incremental update method and system for thermal parameters oriented to layout ECO, which solves the technical problem that existing multi-scale incremental update methods for thermal parameters lead to serious waste of computing resources.

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of relevant departments, and in compliance with relevant laws, regulations, and standards. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0066] Terminology Explanation:

[0067] 1. GDS: GDS usually refers to GDSII, a commonly used data exchange format in integrated circuit layout design. It is used to describe geometric information such as polygons, paths, arrays, and layer numbers. GDS itself mainly stores two-dimensional layout graphics and does not directly contain complete three-dimensional layer thickness, elevation, and material thermal property information. The GDSII format does not include thickness and height information from the layer stack-up information; these need to be added separately after importing to form a three-dimensional structure.

[0068] 2. OASIS: OASIS is a compressed layout file format for large-scale layout data. Similar to GDSII, it also mainly describes the layout geometry and can be used for data representation in advanced chip and package designs.

[0069] 3. PDK: PDK stands for Process Design Kit, which is a complete set of process descriptions and design support information provided by process manufacturers to designers. It typically includes layer definitions, layer thicknesses, material types, design rules, device parameters, via definitions, and some parasitic or process constraint information. For thermal modeling, PDK can supplement the missing layer thickness, material, and stacking relationships in GDS / OASIS.

[0070] 4. BEOL: BEOL stands for Back End Of Line, commonly known as back-end interconnect technology. It usually refers to the multilayer metal interconnects, dielectric layers, vias, and other structural regions after transistor manufacturing is completed. The BEOL region has a complex material composition, strong directionality, and a large scale span, making it a key area for thermal modeling and thermal parameter extraction.

[0071] 5. RDL: RDL stands for Redistribution Layer, which is an important metal wiring layer in advanced packaging used to redistribute chip pads, bring out electrical connections, and interconnect packages. RDLs typically have strong planar orientation and local non-uniformity, thus significantly affecting local heat conduction.

[0072] 6. Via: Vias typically refer to conductive via structures between layers, used to achieve electrical connections between different metal layers. In thermal analysis, in addition to their electrical connection function, vias are usually also important vertical heat conduction paths. Their number, size, distribution, and connectivity all affect the local heat transfer capacity in the z-direction.

[0073] 7. ECO: ECO stands for Engineering Change Order, which refers to the process of modifying layouts, traces, vias, contacts, or local structures in the later stages of the design process. ECOs are typically characterized by locality, frequency, and iterativeness, thus incurring additional costs for maintaining thermal parameters and updating thermal analysis.

[0074] 8. Thermal Parameters: Thermal parameters refer to the set of relevant parameters used for thermal analysis, thermal verification, or thermal simulation. They typically include thermal conductivity, thermal conductivity tensor, equivalent interfacial thermal resistance, volumetric heat capacity, and thermal diffusivity. In this invention, thermal parameters are preferably organized and updated in windows.

[0075] 9. Thermal conductivity tensor: The thermal conductivity tensor is a tensor-form thermal parameter used to describe the differences in heat transfer capacity of a material or local structure in different directions. Unlike a single scalar thermal conductivity, the thermal conductivity tensor can simultaneously characterize the thermal conductivity in the x, y, and z directions, as well as possible directional coupling relationships, making it more suitable for describing anisotropic heat conduction in complex layout structures.

[0076] 10. Multi-scale: Multi-scale refers to establishing parameter representations with different spatial resolutions, such as coarse-scale, meso-scale, and fine-scale, for the same design area simultaneously. Coarse-scale is usually suitable for rapid global or module-level analysis, fine-scale is usually suitable for detailed analysis of hot spots and complex structural areas, and meso-scale is used to connect the results of coarse and fine-scale.

[0077] 11. Window: A window is a local analysis region divided within the layout space, serving as the basic organizational unit for extracting, updating, and storing thermal parameters. Windows can be divided using a fixed size or an adaptive method. Windows at different scales typically have a parent-child containment relationship.

[0078] 12. Dirty Window: A dirty window refers to a window that, after layout ECO (Enhanced Ecosystem Operation), is determined through difference detection and impact propagation analysis to require recalculation, reaggregation, or recalibration of thermal parameters. Dirty windows can be categorized into direct change windows, propagated change windows, and aggregation update windows.

[0079] 13. Incremental Update: Incremental update refers to an update method that, after a local change in the layout, does not recalculate the thermal parameters of the entire layout, but only performs local updates, cross-scale write-backs, and result replacements on the affected area and its associated windows. This method can reduce redundant calculations and improve design iteration efficiency.

[0080] 14. Impact propagation: Impact propagation refers to the fact that although a local change in the map occurs within a limited area, its thermal impact may propagate to other windows through spatial adjacency, cross-layer thermal channel relationships, parent-child scale relationships, or local dominant thermal paths. Therefore, it is necessary to expand the range of windows that need to be updated according to preset rules.

[0081] 15. Thermal Parameter Database: The thermal parameter database is a collection of data that organizes and stores thermal parameters and their metadata corresponding to different scale windows in a unified manner. In addition to the thermal parameters themselves, this database may also include information such as window coordinates, scale level, parent-child relationship, update time, version number, update method, and confidence level, to support subsequent thermal analysis calls and ECO incremental maintenance.

[0082] 16. Local High-Precision Update: Local high-precision update refers to the process of recalculating thermal parameters using higher-precision local modeling and solution methods for high-risk dirty windows. This process is typically used to handle regions with large variations, complex structures, or those sensitive to thermal results, in order to improve the accuracy of the updated results.

[0083] 17. Fast Update: Fast update refers to the process of updating hot parameters with lower computational cost for ordinary dirty windows by using methods such as local difference feature correction, lookup table mapping, lightweight proxy models, or simplified equivalent calculations. Fast update is mainly used to improve the overall update efficiency in ECO scenarios.

[0084] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a multi-scale incremental update method for layout ECO provided in Embodiment 1 of the present invention.

[0085] This invention provides a multi-scale incremental update method for thermal parameters oriented towards layout ECO, comprising:

[0086] Step 101: Obtain the old layout file, the new layout file, the old multi-scale thermal parameter database, and the PDK process data. Based on the old and new layout files, perform layout difference detection and output layout change segment data.

[0087] The old layout file refers to the complete layout data file corresponding to the chip before the engineering changes were performed. It fully records the layout graphics and overall layout information of various chip devices, wiring, vias, etc., and serves as the benchmark file for comparing the differences between the old and new layouts.

[0088] The new layout file refers to the complete layout data file generated after the chip undergoes engineering changes. It records the graphic structure, layout distribution, and other information after the layout is modified, and is used in conjunction with the old layout file to complete the difference identification.

[0089] The old version of the multi-scale thermal parameter database refers to a database that specifically stores thermal parameters, historical calculation results, and window association information at different spatial scales in the previous version of the chip, and can provide existing data support for incremental updates of thermal parameters in this solution.

[0090] PDK (Process Design Kit) process data refers to the relevant data corresponding to the chip process design kit, including process information such as the thermal properties of chip fabrication materials, device structural parameters, and interlayer interface characteristics. It is an essential basis for building thermal models and calculating thermal parameters.

[0091] Map change fragment data refers to the data set extracted after map difference detection, which records specific information related to map changes, such as the location, level, type of change, and size changes of the changed area.

[0092] It should be noted that the process involves acquiring old and new layout files, the old multi-scale thermal parameter database, and PDK process data. Based on the old and new layout files, layout difference detection is performed, and layout change fragment data is output. This step completes the unified collection of basic data sources and the comparative analysis of the old and new layouts, accurately identifying the areas of layout modification. The output layout change fragment data will directly provide core input for the subsequent multi-scale window mapping process.

[0093] Further, step 101 may include the following sub-steps:

[0094] S11. Compare the old and new map files according to the map layer level, region or graphic element, and identify the changed areas corresponding to multiple preset difference types.

[0095] S12. Extract the map change fragment data for each changed region.

[0096] Layout layer level refers to the layer in integrated circuit layout that uses independent physical / process layout layers as the comparison unit. Each layout layer corresponds to a photolithography process in chip manufacturing, specifically including layers with different functions such as metal layers (e.g., M1, M2, etc.), via layers (e.g., V1, V2, etc.), polysilicon layers, diffusion layers, contact layers, and bonding region layers. When comparing, a single layout layer is taken as the whole scope, which can identify all layout graphic changes within that layer, or changes in the relationship between that layer and other layers (e.g., the addition or removal of inter-layer connection structures), and is suitable for identifying overall changes at the layer level.

[0097] The region is a local spatial range in the integrated circuit layout, divided according to function, physical location, or preset coordinates. It is usually a rectangle or polygon, corresponding to a single functional module of the chip (such as a CPU core or memory cell block), a local wiring area, or a custom comparison area. When comparing, all layout layer graphics in the region are used as comparison objects to identify layout changes in the region. It is suitable for identifying changes at the level of local functional modules, which can narrow the comparison range and improve the efficiency of change identification.

[0098] Graphical units are the smallest independent graphic units that make up an integrated circuit layout. They are the basic components of layout data and include rectangles (such as metal blocks and contact holes), polygons (such as irregular wiring patterns), paths (such as long strip wiring), via cells, and instances (such as repeatedly called macro cells / IP cores). When comparing, each graphic unit is used as the comparison unit to identify changes in the position, shape, size, and parameters of the graphic unit. It is suitable for fine-grained identification of local graphic changes and can accurately locate changes in individual layout graphics.

[0099] Intra-layer geometric changes refer to changes in the geometry of the layout within the same integrated circuit process layer (such as a single metal layer, polysilicon layer, diffusion layer, etc.) that do not involve changes in cross-layer connections. Specifically, these include the addition, deletion, translation, rotation, scaling, merging, and segmentation of layout elements such as polygons, rectangles, and wiring paths within the layer, as well as local adjustments to the graphic outline. These are pure geometric or positional changes that occur only within a single layer and do not affect the connection structure with other process layers.

[0100] Metal boundary changes refer to local modifications to the edge contours of metal conductor patterns in the layout of integrated circuit metal layers. These include local adjustments to metal linewidths, offsets of metal pattern edges, chamfering modifications of metal corners, corrections to local protrusions or depressions in the metal contours, and increases or decreases in the length of metal pattern edges. These changes only involve adjustments to the boundary shape of the metal conductors and do not alter the overall affiliation of the metal pattern or its connection to other layers. They represent fine-tuning of the conductor contours within the metal layer.

[0101] The addition or removal of vias in an integrated circuit layout refers to the addition or removal of conductive vias used to connect different metal layers. Specifically, it includes adding via structures between two metal layers to establish new cross-layer connections and deleting existing via structures to disconnect existing cross-layer connections. It involves changes to cross-layer connection relationships and directly affects the electrical connection paths between metal layers, belonging to the cross-layer related layout difference type.

[0102] Contact area variation refers to changes in the contact area between conductor patterns (such as metal lines and vias) and other conductive structures, semiconductor active regions, or insulating structures in an integrated circuit layout. Specifically, this includes increasing or decreasing the size of contact holes, increasing or decreasing the number of via arrays, adjusting the area of ​​the contact pattern between metal conductors and active regions, and changing the area of ​​the overlapping region between metal lines and vias. These changes directly affect circuit performance parameters such as contact resistance and current carrying capacity, and belong to the type of layout difference directly related to circuit performance.

[0103] Changes in dummy patterns refer to modifications made to non-functional dummy patterns (such as dummy metal, dummy vias, dummy polysilicon, etc.) added to meet the density rules of the manufacturing process and reduce deviations in the chemical mechanical polishing (CMP) process in integrated circuit layouts. Specifically, these changes include the addition, deletion, position adjustment, size modification, and density adjustment of dummy patterns. These patterns do not participate in the electrical connection function of the circuit and only serve the manufacturing process. Their changes are non-functional layout process-related modifications.

[0104] Bonding area changes refer to modifications made to the bonding pads and related connection structures in the integrated circuit layout that enable electrical connections between the chip and external circuits. These modifications include the addition, deletion, position shifting, size adjustment, and shape modification of bonding pads, as well as adjustments to the linewidth and path of the pad leads. These modifications directly affect the chip's external connectivity and package compatibility, and fall under the category of layout differences related to chip input / output interfaces.

[0105] Changes in the hierarchical layout of integrated circuits refer to modifications made to instances at each level (such as module instances, macrocell instances, and standard cell instances). Specifically, these changes include adding, deleting, replacing, translating, rotating, and scaling instances, as well as modifying the versions of submodules referenced by instances and adding or removing instance arrays. These changes are at the level of the hierarchical design structure and not only involve individual primitives but also affect the hierarchical structure of the layout and the module reference relationships.

[0106] It should be noted that this invention reads the old map, the new map, and the multi-scale thermal parameter database corresponding to the old map. The old database can be formed by the previous full generation or accumulated from the previous several incremental updates. After reading the database, the system does not directly discard it entirely, but treats it as a baseline version to be maintained.

[0107] Next, layout difference detection is performed. The objects of difference detection preferably include: intra-layer geometric changes, metal boundary changes, additions or deletions of vias (conductive via structures used to connect different metal layers in an integrated circuit layout), contact area changes, fill pattern changes, bonding region changes, and layout hierarchy instance changes. The implementation of difference detection is not limited to a single algorithm and can be completed through Boolean operations, layer comparison, tile-level hash comparison, primitive signature comparison, or instance-level difference analysis. The system outputs several change fragments, each containing at least the location, layer identifier, change type, and change area or length measure.

[0108] Specifically, this step first loads the old and new layout files through the layout version input module. The files support mainstream integrated circuit layout formats such as GDSII and OASIS. Then, a full-domain comparison is performed on the two layout files according to three different comparison granularities: layer level, region, and element unit. In the layer level comparison stage, the independent process layout layer corresponding to the chip manufacturing is used as the basic unit of division. All layers, such as metal layer, via layer, polysilicon layer, diffusion layer, contact layer, and bonding region layer, are checked one by one to investigate changes in the graphics within a single layer and related changes in the connection structure between different layers. In the region comparison stage, multiple independent local regions are divided according to the chip functional modules and custom spatial coordinates. The layout graphics of all layout layers within each region are compared as a whole to locate layout changes at the functional module level. In the element unit comparison stage, the smallest basic layout elements, such as rectangles, polygons, wiring paths, via units, and macrocell instances, are compared. Using basic graphic elements as comparison objects, the changes in spatial location, outline, and dimensional parameters of each element are checked one by one. The entire comparison process can use algorithms such as Boolean operations, tile-level hash comparison, graphic element signature comparison, and instance-level difference analysis to perform difference determination, thereby identifying all change areas corresponding to preset difference types such as intra-layer geometric changes, metal boundary changes, via additions or deletions, contact area changes, fill graphic changes, bonding area changes, and layout layer instance changes. After completing the location of all change areas, for each identified change area, various information such as area attributes, layout attributes, and modification attributes are systematically collected and integrated to generate standardized layout change fragment data. This type of data will completely include information such as the spatial coordinate range of the change area, the layer identifier to which it belongs, the specific difference type, the modified area, the modified length, the associated graphic element number, and the functional area number to which it belongs, providing complete and accurate original change basis for the subsequent multi-scale window mapping process.

[0109] For example, taking the ECO modification scenario of the back-end interconnect (BEOL) area of ​​a 3D stacked chip as an example, both the old and new layout files use the industry-standard GDSII format. This engineering change mainly involves local adjustments to the chip's M2 metal layer and its corresponding V2 via layer. The first step is to perform multi-granularity layout comparison: First, a layer-level comparison is performed, verifying all process layout layers one by one to determine that there are layout changes in the M2 metal layer and V2 via layer; then, a region comparison is performed, delineating the rectangular area (X: 120μm~250μm, Y: 90μm~210μm) corresponding to the logic sub-module in the middle of the chip as the core change area based on the coordinates; finally, a primitive unit comparison is performed, identifying two types of primitive changes in the above area: minor adjustments to the four metal boundaries of the M2 metal layer and the addition of 15 sets of via conductive vias in the V2 via layer, corresponding to two preset difference types: metal boundary changes and via additions. The second step is to extract the map change segment data corresponding to the changed area. This segment data specifically includes the following: 1. Unique segment identifier: Segment number CP002, belonging to the functional area: Central Logic Submodule; 2. Spatial location parameters: Change area coordinate range X: 120μm~250μm, Y: 90μm~210μm, total area 1690μm. 2 3. Layer information: Involves two process layers: M2 metal layer and V2 via layer; 4. Difference type labeling: Metal boundary changes, new vias; 5. Quantitative parameters of changes: Total length of change in the metal boundary of layer M2 is 52.3μm, and 15 new vias are added in layer V2; 6. Associated primitive numbers: Primitive numbers for boundary adjustments in layer M2 are M2-01, M2-02, M2-03, and M2-04, and primitive numbers for new vias in layer V2 are V2-01 to V2-15. The above complete layout change fragment data can be directly imported into the multi-scale window mapping module for subsequent processing.

[0110] It is worth mentioning that, in terms of the implementation method of difference detection, this invention is not limited to primitive-level Boolean comparison. Title / tag-level signature comparison, local hash index comparison, instance-level difference comparison, or map segment summary comparison can also be used. As long as the local changes between the old and new maps can be identified and the changed segments can be output, the purpose of this invention can be achieved.

[0111] Furthermore, the identification of differences between the old and new layouts is crucial. These differences go beyond simple geometric variations and should include changes in intra-layer graphics, vias, local contact, boundaries, fill, and instance hierarchy. Next, the layout differences are mapped to a thermal parameter window system. This step is not a simple geometric mapping but rather transforms the layout changes into a problem of which windows need thermal parameter maintenance updates. Thirdly, there's the issue of impact propagation; some windows that haven't directly undergone geometric changes may be affected by thermal channels, adjacency relationships, parent-child scale relationships, or cross-layer connections, and therefore must also be included in the update set. Finally, rapid updates, high-precision local updates, cross-scale write-backs, and database replacements are performed on these windows. These actions together constitute a complete technical closed loop.

[0112] Step 102: Perform multi-scale window mapping based on the map change fragment data and the window index table in the old multi-scale thermal parameter database, and output a dirty window candidate set.

[0113] The window index table refers to the structured index table in the old version of the multi-scale thermal parameter database. Specifically, it contains information such as the unique identifier of thermal analysis windows at different spatial scales, the scale level (fine scale / medium scale / coarse scale), the spatial coordinate range, the parent-child hierarchical association index (parent window ID, child window ID), the cross-layer thermal path association index (associated layer window ID), and the list of adjacent windows.

[0114] It should be noted that this step takes the map change fragment data output by the previous map difference detection as input, combines the window index table in the old multi-scale thermal parameter database to carry out multi-scale window mapping, locates the windows associated with each map change fragment based on the window position and hierarchical association information recorded in the window index table, and integrates and outputs a dirty window candidate set.

[0115] Furthermore, step 102 may include the following sub-steps:

[0116] S21. Based on the window index table, locate the fine-scale window that intersects with the data of each map change segment, and mark each fine-scale window as a direct change fine window.

[0117] S22. Based on the parent-child index of the window index table, trace upwards to the parent window corresponding to the directly changed fine window and mark it as a candidate dirty parent window;

[0118] S23. Based on the cross-layer association index of the window index table, locate the window that directly changes the cross-layer association of the fine window and mark it as a candidate dirty cross-layer window;

[0119] S24. Integrate the directly changing fine window, the candidate dirty parent window, and the candidate dirty cross-layer window, and output the dirty window candidate set.

[0120] It should be noted that the change mapping performed in this invention is not a simple projection of the change fragment onto a single grid, but rather, based on a multi-scale window system, it simultaneously maps the same change fragment to fine-scale windows, medium-scale windows, and coarse-scale windows. Preferably, this invention first locates the fine-scale windows that intersect with the change fragment and marks them as first-type dirty windows; then, it marks their parent windows as candidate dirty windows by using parent-child indices; and finally, it extends to potentially affected adjacent windows through cross-layer relationships and spatial adjacency relationships.

[0121] Specifically, the input to this step is the map change fragment data output from the upstream map difference detection stage. It also requires reading the complete window index table from the old multi-scale thermal parameter database. This window index is a structured data table predefined and continuously maintained during the construction of the multi-scale thermal parameter system. Besides storing the unique identifier, spatial coordinates, and map layer to which each scale window belongs, the table also pre-constructs parent-child indexes based on the hierarchical relationship between multi-scale windows, as well as cross-layer association indexes based on the cross-layer association relationship between process layers. The parent-child indexes record the hierarchical relationship between fine-scale, medium-scale, and coarse-scale windows, and the cross-layer association indexes... The linked index records the relationships between windows in different process layers, formed by conductive vias and thermal channels. Both types of indexes can be directly retrieved and called without real-time generation, thus clarifying the source of the indexes and the connection between the input and output of the entire process. First, fine-scale window matching is performed based on this window index table. The structured metadata of each layout change segment is traversed, extracting the spatial coordinate range and the corresponding layout layer identifier. A bounding box intersection algorithm is used to compare the spatial region and layout layer information of each layout change segment with all fine-scale windows in the window index table. When the spatial ranges of the two segments intersect and the layout layers completely match, the fine-scale window is selected. The window is identified as the target window and uniformly marked as a directly changed fine window. Simultaneously, a unique number corresponding to the map change segment is bound to the window attributes. Then, a tracing operation is performed based on the parent-child index built into the window index table. Using the unique identifier of each marked directly changed fine window as the search keyword, the system searches upwards layer by layer according to the hierarchical order of fine scale → medium scale → coarse scale, identifying the first-level parent window and higher-level parent windows corresponding to the current window. All retrieved subordinate parent windows are marked as candidate dirty parent windows, and the hierarchical tracing path is recorded simultaneously. Next, the search continues based on the cross-layer association index built into the window index table, still focusing on directly changed fine windows. The unique identifier of the window is used as the search keyword. Other process layer windows that have conductive or thermal conductivity connections with the window are filtered out. These cross-layer associated windows are marked as candidate dirty cross-layer windows, and the specific cross-layer association type is labeled. Finally, all marked direct change sub-windows, candidate dirty parent windows, and candidate dirty cross-layer windows are summarized, and a global deduplication operation is performed to remove duplicate marked window entries. Then, the identifier, scale, associated fragments, source type, and other information of all windows are standardized according to a unified data format. Finally, a standardized dirty window candidate set is output, which will serve as dedicated input data for the downstream thermal impact propagation analysis.

[0122] For example, taking the ECO change scenario of the layout of the logic submodule in the middle of the chip as an example, the layout change segment obtained in the previous step is numbered CP002. Its layout change segment structured metadata is: spatial coordinates X: 120μm~250μm, Y: 90μm~210μm, involving two types of process layers: M2 metal layer and V2 via layer. The multi-scale window rule of this invention is fine-scale window 1μm×1μm, medium-scale window 10μm×10μm, and coarse-scale window 100μm×10μm. The window index table of the old multi-scale thermal parameter database has been pre-entered with complete data. Among them, the parent-child index clearly states that one fine-scale window belongs to one medium-scale window and one medium-scale window belongs to one coarse-scale window. The cross-layer association index indicates that the V2 via layer window has thermal and electrical conductivity association with the upper and lower M1 and M3 metal layer windows. The first step is to locate the directly changing fine windows: Retrieve all fine-scale window data from the window index table. Using a bounding box intersection algorithm, compare the coordinates and layer of the CP002 fragment with the fine-scale windows, filtering out 15851 intersecting windows, such as windows W1_M2_120_90 (coordinates 120μm~121μm, Y 90μm~91μm, M2 layer) and W1_V2_121_91 (coordinates 121μm~122μm, Y 91μm~92μm, V2 layer). Mark all these windows as directly changing fine windows and bind them to the fragment number CP002. The second step is to trace the candidate dirty parent windows: Search the parent-child index using the window ID of W1_M2_120_90 to find its first-order mesoscale parent window W2_M2_120_90 and its second-order coarse-scale parent window W3_GLOBAL_100_100. The process involved tracing 5851 directly changing fine windows sequentially, ultimately yielding 169 mesoscale windows and 1 coarse-scale window, all uniformly marked as candidate dirty parent windows. The third step involved locating candidate dirty cross-layer windows: using the ID of the V2 layer directly changing fine window W1_V2_121_91, the cross-layer association index was retrieved, matching the associated M1 layer window W1_M1_121_91 and M3 layer window W1_M3_121_91. After traversing all V2 layer fine windows, a total of 100 cross-layer associated windows were selected and marked as candidate dirty cross-layer windows. The fourth step involved integrating and generating a dirty window candidate set: summarizing 15851 directly changing fine windows, 169 candidate dirty parent windows, and 100 candidate dirty cross-layer windows, removing 3 duplicate windows caused by multiple associations, and standardizing the attribute information of all windows to output the final dirty window candidate set for subsequent thermal effect propagation analysis.

[0123] Among them, the direct change fine window is a fine-scale thermal analysis window that intersects with the spatial range of the map change fragment data. It is the basic window unit directly related to map changes in the multi-scale window mapping. Specifically, it includes attribute information such as the window's unique identifier, the map layer identifier to which it belongs, the spatial coordinate range, the associated map change fragment ID, the marking status (direct dirty window), and the thermal parameter update mark.

[0124] Candidate dirty parent windows are parent-child indices obtained by tracing upwards from directly changed fine windows in the window index table. They are medium-scale and coarse-scale parent windows that are hierarchically related to map changes and are to be marked. Specifically, they include the window's unique identifier, window hierarchy identifier (medium-scale / coarse-scale), map layer identifier, spatial coordinate range, list of IDs of subordinate directly changed fine windows, hierarchical tracing path (e.g., fine-scale → medium-scale → coarse-scale), marking status (candidate dirty parent window), and thermal parameter update markers.

[0125] Candidate dirty cross-layer windows are cross-layer thermal path associated windows located by direct change fine windows through the cross-layer association index of the window index table. They belong to the windows to be marked that have cross-layer thermal conduction association with the layout changes. Specifically, they include the window's unique identifier, the identifier of the layout layer to which it belongs, the ID of the associated direct change fine window, the cross-layer association type identifier (such as via connection type / adjacent layer thermal coupling type), the marking status (candidate dirty cross-layer window), thermal parameter update mark and other attribute information.

[0126] The parent-child index is a structured index information in the window index table that records the hierarchical relationship between thermal analysis windows at different levels. It is stored in key-value pairs. The key is a unique identifier for a fine-scale or mesoscale window, and the value is the association information between the corresponding parent and child windows. Specifically, it includes fields such as window ID, parent window ID (corresponding to a higher-level mesoscale / coarse-scale window), list of child window IDs (corresponding to a lower-level fine-scale / mesoscale window), hierarchical relationship identifier, and index version number. It supports bidirectional positioning, tracing back to the parent window from the child window and querying the child window from the parent window.

[0127] The cross-layer association index is a structured index information that records the thermal conduction association between thermal analysis windows of different layers in the window index table. It is stored in key-value pair format, where the key is the unique identifier of the window and the value is the corresponding cross-layer association window information, which specifically includes fields such as source window ID, target cross-layer window ID, layer identifier of the source window, layer identifier of the target window, association type identifier (such as via connection / adjacent layer coupling), thermal coupling strength level, and index version number. It is used to quickly locate cross-layer windows with thermal path associations.

[0128] The fine-scale window is the thermal analysis window with the highest spatial resolution and smallest coverage in the multi-scale thermal analysis system. It is the smallest unit for calculating thermal parameters. Its scale division is based on the size of the process feature and the local area of ​​the layout. The window side length is usually 4-16 times the size of the target process feature, or covers a single local wiring / device unit area in the layout (such as a rectangular / square window with a side length ≤10μm). Each fine-scale window contains only a small amount of layout graphics. Specifically, it includes attributes such as the window's unique identifier, the layer identifier of the layout to which it belongs, the spatial coordinate range, the parent window ID, the cross-layer associated window ID, historical thermal parameter data, associated layout graphic information, and thermal parameter update flags.

[0129] The mesoscale window is a thermal analysis window with medium spatial resolution and medium coverage in the multi-scale thermal analysis system. It is used for the aggregation calculation of thermal parameters within a medium range. Its scale division is based on the fine-scale window and functional sub-module. Typically, the window side length is 64-256 times the size of the target process feature, or it covers the area of ​​a single functional sub-module (such as a logic unit group or a memory array segment) (such as a rectangular / square window with a side length of 10-100μm). Each mesoscale window contains 4-16 fine-scale windows. Specifically, it includes attributes such as the window's unique identifier, the layer identifier to which it belongs, the spatial coordinate range, a list of lower-level fine-scale window IDs, the upper-level coarse-scale parent window ID, cross-layer associated window IDs, aggregated thermal parameter data, and thermal parameter update flags.

[0130] The coarse-scale window is the thermal analysis window with the lowest spatial resolution and the largest coverage in the multi-scale thermal analysis system. It is used for global thermal parameter aggregation calculation. Its scale division is based on the meso-scale window and functional module. Typically, the window side length is more than 512 times the size of the target process feature, or covers the entire functional module (such as a storage block) area (such as a rectangular / square window with a side length ≥100μm). Each coarse-scale window contains 4-16 meso-scale windows. Specifically, it includes attributes such as the window's unique identifier, the layer identifier to which it belongs, the spatial coordinate range, the list of IDs of the lower-level meso-scale windows, aggregated thermal parameter data, cross-layer association information, and thermal parameter update flags.

[0131] The dirty window candidate set is a structured set of all window units directly or indirectly affected by the current layout change, generated by a multi-scale hierarchical window mapping algorithm based on the structured metadata of the layout change fragments during the incremental update process of thermal parameters after an Engineering Change of Layout (ECO) in integrated circuits. All windows in the set are encapsulated in a standardized format, and each window element carries metadata such as a unique window identifier, window scale level, spatial coordinate range, process layer, associated layout change fragment ID, and window source type (direct intersection / parent window tracing / cross-layer association). This is used to accurately locate the window units whose thermal parameters need to be updated, avoiding the waste of computing power and inefficiency caused by full recalculation. The set contains three types of window elements strongly associated with map change segments: First, fine-scale windows that directly intersect with map change segments. These windows are the lowest-level basic units of multi-scale mapping, with a preset minimum resolution (e.g., a fine-scale window of 1μm×1μm). They partially intersect or completely contain the spatial coordinate range of the map change segment, and are directly affected by this change. They are the core basic component of the dirty window candidate set, and their window metadata is marked as "directly intersecting" for "window source type". Second, medium-scale and coarse-scale parent windows traced upwards from the directly changed fine windows. The multi-scale windows adopt a hierarchical pyramid structure design, with fine-scale windows corresponding to the bottom layer of the pyramid. Medium-scale (e.g., 10μm×10μm) and coarse-scale (e.g., 100μm×100μm) windows are the upper-level parent windows. There is a fixed hierarchical mapping relationship between parent and child windows (e.g., one medium-scale window corresponds to 10×10 fine-scale windows). The first type of window is the coarse-scale window, which corresponds to 10×10 medium-scale sub-windows. When any fine-scale sub-window is a directly changing fine window, all its corresponding upper-level medium-scale and coarse-scale parent windows are included in the dirty window candidate set. These parent windows are windows that indirectly affect the change and are used to ensure the consistent updating of upper-level multi-scale thermal parameters. Their window metadata is marked as "parent window traceability" for "window source type". The second type is windows that are directly changing fine windows and are associated across layers. These windows are based on the process layer connection relationship of the layout (such as the corresponding connection relationship between metal layers and via layers, the relationship between bonding regions and leads, the contact relationship between polysilicon layers and contact layers, etc.) and other process layer window units that have cross-layer electrical connections or structural relationships with the directly changing fine windows. These windows are windows that indirectly affect the change across layers and are used to avoid the problem of cross-layer thermal parameter mismatch caused by a single-level change. Their window metadata is marked as "cross-layer association" for "window source type".

[0132] For example, taking the layout change segment of the CP002 logic submodule in the middle of the chip as an example, this segment involves fine-tuning the boundary of the M2 metal layer and adding 15 vias to the V2 via layer. The spatial coordinate range is X: 120μm~250μm, Y: 90μm~210μm, and the process layers are the M2 metal layer and the V2 via layer. The multi-scale window pyramid structure used in this study is: fine-scale window 1μm×1μm, medium-scale window 10μm×10μm, and coarse-scale window 100μm×100μm. The hierarchical mapping relationship is that one medium-scale window corresponds to 10×10 fine-scale sub-windows, and one coarse-scale window corresponds to 10×10 medium-scale sub-windows. The specific composition of the generated dirty window candidate set is as follows:

[0133] Directly intersecting fine-scale windows: All 1μm×1μm fine-scale windows that intersect with the spatial range of the CP002 fragment, totaling 131 columns (X: 120~250μm)×121 rows (Y: 90~210μm) = 15851 windows, including metal boundary adjustment windows of the M2 metal layer and newly added via windows of the V2 via layer. A single window metadata example is: window IDW1_M2_120_90, scale level "fine-scale", spatial coordinates (120μm, 90μm)-(121μm, 91μm), process layer M2, associated fragment IDCP002, source type "directly intersecting";

[0134] The parent windows traced back to the intermediate and coarse-scale parent windows are as follows: The intermediate-scale parent windows corresponding to the above 15851 fine-scale windows consist of 13 columns × 13 rows = 169 10μm × 10μm windows, and one 100μm × 100μm coarse-scale parent window covering all intermediate-scale windows. An example of a intermediate-scale window is: Window IDW2_M2_120_90, scale level "Medium-scale", spatial coordinates (120μm, 90μm)-(130μm, 100μm), process layer M2, associated fragment IDCP002, source type "Parent Window Traceability"; an example of a coarse-scale window is: Window IDW3_GLOBAL_100_100, scale level "Coarse-scale", spatial coordinates (100μm, 100μm)-(200μm, 200μm), process layer "Global", associated fragment IDCP002, source type "Parent Window Traceability".

[0135] Cross-layer association windows: Based on the connection relationship between the V2 via layer and the upper and lower metal layers, there are 50 fine-scale windows each in the M1 and M3 metal layers that are electrically connected to the fine-scale windows that change directly with the V2 layer. An example is: window IDW1_M1_120_90, scale level "fine-scale", spatial coordinates (120μm,90μm)-(121μm,91μm), process layer M1, associated fragment IDCP002, source type "cross-layer association". These windows are included in the candidate set because their thermal parameters will be affected by the changes in the vias due to the electrical connection with the newly added vias.

[0136] The three types of windows mentioned above together constitute a complete dirty window candidate set. All windows carry standardized metadata, clearly distinguishing the source type, providing a precise processing range for subsequent incremental updates of hot parameters, and avoiding the resource consumption of full recalculation.

[0137] Step 103: Using preset influence propagation rules, perform thermal influence propagation analysis on the dirty window candidate set based on the window index table in the old multi-scale thermal parameter database, and output the affected window set.

[0138] The preset impact propagation rules are a complete set of pre-configured, programmable inference rules designed to adapt to the multi-scale, cross-layer, and multi-path thermal disturbance propagation characteristics of chips. These rules include five sub-rules: spatial neighborhood rules, cross-layer rules, scale rules, hot channel rules, and risk expansion rules. Based on the window index table storing basic information such as window spatial adjacency, cross-layer association, scale subordination, and hot channel connectivity, the rules comprehensively and multidimensionally infer the propagation boundaries and coverage of thermal disturbances caused by layout changes. Specifically, the spatial neighborhood rules filter windows that are spatially adjacent to dirty windows and have thermal conduction associations; the cross-layer rules filter associated windows with thermal path connectivity between different layout layers; the scale rules filter associated windows with parent-child subordination relationships at multiple scale levels; the hot channel rules filter conduction windows connected along the chip's dominant hot channels; and the risk expansion rules filter risk-related windows that can spread to highly thermally sensitive areas.

[0139] The affected window set is a structured window dataset formed after the dirty window candidate set has undergone comprehensive thermal impact propagation simulation, data merging and deduplication processing through preset impact propagation rules. It gathers all thermal analysis windows directly triggered and indirectly related to thermal disturbances caused by changes in map ECO. Specifically, it includes the original change-related windows of the dirty window candidate set, spatially adjacent heat conduction windows filtered by spatial neighborhood rules, inter-layer heat path connectivity windows filtered by cross-layer rules, multi-scale hierarchical aggregation windows filtered by scale rules, dominant heat channel connectivity windows filtered by heat channel rules, and high-heat-sensitive area risk diffusion windows filtered by risk expansion rules. Each window in the set corresponds to a unique window identifier, window scale type, map layer to which it belongs, type of thermal impact source, spatial coordinate range, and related map change information, etc. Complete attribute data have been stored, and duplicate and overlapping invalid window data have been removed, providing a comprehensive and non-redundant standardized data source of windows to be processed for subsequent dirty window classification and determination.

[0140] It should be noted that this step follows the previously generated dirty window candidate set. Based on the window association information recorded in the window index table of the old multi-scale thermal parameter database, it performs thermal disturbance diffusion simulation on the dirty window candidate set according to the preset influence propagation rules. It comprehensively explores the thermal influence associated windows generated by the map change, eliminates duplicate and redundant windows, summarizes all effective windows affected by thermal disturbance, and finally outputs a complete set of affected windows.

[0141] It's worth noting that the propagation rules don't necessarily rely solely on fixed-radius spatial neighborhood expansion. Graph propagation based on heatmaps, scoring methods for changes in training data, sensitive region expansion based on rule bases, or adaptive propagation based on local thermal gradient histories can also be used. Any method that allows the set of affected windows to be derived from the directly changing window can be considered an equivalent replacement.

[0142] Furthermore, step 103 may include the following sub-steps:

[0143] S31. Based on the window index table, filter the adjacent window list that conforms to the spatial neighborhood rules and include them in the propagation change window set;

[0144] S32. Based on the cross-layer association index of the window index table, filter the upper and lower layer hot path connection windows that meet the cross-layer rules and include them in the propagation change window set;

[0145] S33. Based on the parent-child index of the window index table, filter parent windows and high-level parent windows that meet the scaling rules and include them in the aggregated update window set;

[0146] S34. Based on the local dominant hot channel association information of the window index table, filter the connected windows that meet the hot channel rules and include them in the propagation change window set;

[0147] S35. Identify windows belonging to the preset high-heat-sensitive area in the dirty window candidate set, supplement and filter connected windows according to the risk expansion rule, and include them in the propagation change window set;

[0148] S36. Merge the dirty window candidate set, the propagated change window set, and the aggregated update window set to obtain the merged window set;

[0149] S37. Remove duplicate windows from the merged window set and output the set of affected windows.

[0150] It should be noted that the impact propagation analysis performed by this invention is preferably triggered by at least one of the following rules: spatial neighborhood propagation rule, i.e., if a window is adjacent to a changing window and the boundary distance is less than a preset threshold, then that neighboring window may also be affected by thermal parameters; cross-layer propagation rule, i.e., if the change occurs at a via, contact, or interface location affecting vertical heat conduction, then windows connected to its upper and lower layer thermal paths also enter the propagation range; scale propagation rule, i.e., if a fine-scale window is updated, its parent-scale window must enter the set to be aggregated and updated; thermal channel propagation rule, i.e., if the change is located on a locally dominant thermal channel, then related windows connected to that thermal channel also need to be updated; risk expansion rule, i.e., if the changed area belongs to a predefined high thermal sensitivity area, the propagation radius can be adaptively amplified. Through these rules, the system generates a complete set of affected windows, rather than being limited to areas of direct geometric change.

[0151] Specifically, the inputs to this step are the dirty window candidate set output from the previous multi-scale window mapping stage, and the complete window index table retrieved from the old multi-scale thermal parameter database. This window index table uniformly stores the adjacent window list, cross-layer association index, parent-child index, and local dominant thermal channel association information corresponding to each window. At the same time, the system pre-configures five types of influence propagation rules: spatial neighborhood rules, cross-layer rules, scale rules, thermal channel rules, and risk expansion rules, and delineates a preset high-thermal-sensitive area within the chip, which serves as the unified basis for the entire process of screening and judgment. First, the adjacent window list in the window index table is called, and the screening is carried out according to the spatial neighborhood rules. This rule uses the preset boundary distance as the judgment threshold, traverses each window in the dirty window candidate set, extracts its adjacent window information, and filters out all adjacent windows whose boundary distance from the target window is less than the threshold, and uniformly includes them in the propagation change window set. Next, the cross-layer association index of the window index table is called, and the window is screened according to the cross-layer rules. This rule determines that there are effective upper and lower layer thermal paths such as conductive vias and interlayer interfaces connected to the original window. The connected windows are considered as associated windows. The filtered windows are then further incorporated into the propagation change window set. Subsequently, the parent and child indexes of the window index table are retrieved, and filtering is performed according to the scale rule. This rule stipulates that when a fine-scale window changes, all its higher-level parent windows, such as the first-level mesoscale parent window and the second-level coarse-scale parent window, must be included in the update scope. All parent windows obtained from tracing are uniformly incorporated into the aggregated update window set. Next, the local dominant thermal channel association information in the window index table is read, and connected windows on the same heat conduction path are filtered according to the thermal channel rule. These windows are then added to the propagation change window set. Afterward, a judgment is performed based on the predefined preset high-heat-sensitive area. The filtering scope is expanded according to the risk expansion rule. For windows in the dirty window candidate set located in this area, their surrounding connected windows are additionally filtered and included in the propagation change window set. After all filtering is completed, the dirty window candidate set, the propagation change window set, and the aggregated update window set are merged to obtain a merged window set. Then, the unique window identifier is used as the deduplication criterion to remove duplicate records in the merged set, and finally, the affected window set is output.

[0152] For example, taking the ECO scenario of the chip's central logic submodule layout as an example, the layout change segment number is CP002, involving the M2 metal layer and the V2 via layer. The fine-scale window size is 1μm×1μm, the medium-scale window is 10μm×1μm, and the coarse-scale window is 100μm×100μm. The dirty window candidate set contains 15851 directly changed fine windows, 169 candidate dirty parent windows, and 100 candidate dirty cross-layer windows. The preset rule parameters are configured as follows: the spatial neighborhood rule boundary distance threshold is 5μm, the preset high heat sensitive area is the chip power device distribution area (part of the CP002 segment falls within this range), and all five influence propagation rules are enabled. The specific execution process is as follows:

[0153] Based on the adjacent window list and spatial neighborhood rule filtering: traverse all windows in the dirty window candidate set, retrieve the adjacent window list in the window index table, compare the window boundary distance one by one, and filter out 286 adjacent windows with a boundary distance of less than 5μm, and include them in the propagation change window set;

[0154] Based on cross-layer association index and cross-layer rule filtering: Based on the cross-layer association index, locate the thermal path connection windows of the M1 and M3 metal layers above and below the V2 via layer window, a total of 124, and add them all to the propagation change window set;

[0155] Based on parent-child index and scale rule filtering: trace upwards through the parent-child index to find all the high-level parent windows of the medium and coarse scales corresponding to the fine-scale windows, add 2 coarse-scale parent windows and 18 medium-scale parent windows, and include them in the aggregated update window set;

[0156] Based on the local dominant thermal channel association information and thermal channel rule filtering: the local dominant thermal channel to which the CP002 segment belongs was identified, and 72 connected windows within the channel were filtered out and added to the propagation change window set;

[0157] Based on the preset high heat-sensitive area and risk expansion rules, 210 windows in the dirty window candidate set were determined to be located in the preset high heat-sensitive area. According to the rules, the scope was expanded to select 95 surrounding connected windows, which were then included in the propagation change window set.

[0158] Set merging and deduplication: The original dirty window candidate set, the propagation change window set containing 677 windows, and the aggregated update window set containing 20 windows are merged to obtain a merged window set with a total of 16,648 records. Then, deduplication is performed based on the unique identifier of the window to remove 12 duplicate window records. Finally, the affected window set containing 16,636 windows is output, which is directly passed to the subsequent dirty window classification and determination stage.

[0159] Among them, the propagation change window set is a set of thermally affected windows obtained by screening based on spatial neighborhood rules, cross-layer rules, thermal channel rules, and risk expansion rules. It includes the spatial adjacency window of dirty window, the cross-layer thermal path connection window, the dominant thermal channel connection window, and the expansion window of high thermal sensitivity area. It is the set of windows directly affected by conduction in the thermal impact propagation simulation.

[0160] The aggregated update window set is a multi-scale hierarchical parent window set obtained based on scale rules. It includes the medium-scale and coarse-scale parent windows corresponding to dirty windows and is the set of windows to be processed for subsequent multi-scale thermal parameter aggregated updates.

[0161] The preset high thermal sensitivity area is a predefined set of areas on the chip with high thermal sensitivity. It typically includes critical areas such as clock networks, power devices, and I / O interfaces that are susceptible to thermal disturbances. In the risk propagation rules, the dirty window in this area can adaptively amplify the thermal impact propagation radius.

[0162] Step 104: Using preset classification and judgment rules, the affected window set is classified and judged as dirty windows based on the window index table, thermal parameter table and version metadata table in the new map file, map change fragment data and old multi-scale thermal parameter database, and the results are output as ordinary dirty windows and high-risk dirty windows.

[0163] The thermal parameter table is a structured table that stores historical thermal parameter data for each scale window in the old version of the multi-scale thermal parameter database. It includes fields such as unique window identifier, scale level, anisotropic thermal conductivity, equivalent thermal resistance, heat capacity, historical calculation results, and update records, providing historical thermal parameter reference data for dirty window classification.

[0164] The version metadata table is a structured table that records the calculation metadata of thermal parameters for each version in the old version of the multi-scale thermal parameter database. It includes fields such as version number, calculation time, layout version, process conditions, calculation parameter settings, and related change records, providing historical calculation background reference for dirty window classification and judgment.

[0165] Ordinary dirty windows are affected windows that have been assessed by preset classification rules and have a low risk level for updating thermal parameters and a small impact of changes on the overall heat distribution. They typically correspond to windows with weak impact from changes, minor heat conduction paths, and non-high heat-sensitive areas, and are suitable for low-complexity, fast update logic.

[0166] High-risk dirty windows are those affected by changes that have a high risk level for updating thermal parameters and a significant impact on the overall heat distribution, as assessed by preset classification rules. These windows typically correspond to windows with strong impact from changes, located in the dominant thermal channel, and belong to highly heat-sensitive areas, and are adapted to high-precision full update logic.

[0167] The pre-defined classification rules are a programmable, structured dirty window risk assessment and classification standard system customized for the chip multi-scale thermal parameter incremental update process. Each window in the affected window set is used as the assessment object, and the new layout file, layout change fragment data, and the window index table, thermal parameter table, and version metadata table of the old multi-scale thermal parameter database are used as inputs. A multi-dimensional quantitative assessment model outputs the thermal parameter update risk level judgment result for the window, providing a direct basis for subsequent differentiated thermal parameter update logic. This rule system is based on four core dimensions, relying on layout change fragment data and the window index table to assess the correlation type, correlation strength, and change type between the window and the layout change. Changes with direct intersection, high coupling strength, and high thermal conductivity paths will increase the risk score weight. The old multi-scale thermal parameter database's thermal parameter table is then retrieved to assess the window's historical thermal parameter fluctuation range, calculate confidence level, and other parameters. The system considers the update frequency and thermal parameter deviation from surrounding windows. Windows with large thermal parameter fluctuations, low confidence, and high deviation values ​​will have their risk score weight increased. Simultaneously, it combines the new map file and preset high thermal sensitivity area definitions to assess the thermal sensitivity level of the area to which the window belongs and the window's criticality in the heat conduction path. Windows in high thermal sensitivity areas and key heat conduction nodes will have their risk score weight increased. It also retrieves version metadata tables from the old multi-scale thermal parameter database to evaluate the window's past change records, thermal parameter update history, and anomaly records. Windows with frequent changes and historical anomaly records will have their risk score weight increased. The built-in quantitative scoring model sets configurable weights for each dimension. It calculates the thermal parameter update risk score for each window through weighted summation, and then, based on preset risk level thresholds, classifies windows with scores reaching or exceeding the high-risk threshold as high-risk dirty windows, and the remaining windows as ordinary dirty windows, achieving accurate classification of the two types of windows.

[0168] It should be noted that this step follows the previously generated set of affected windows, combines the new map file and map change fragment data, and retrieves the window index table, thermal parameter table and version metadata table from the old multi-scale thermal parameter database. Based on the preset classification judgment rules, the degree of change correlation and thermal parameter impact risk of each affected window are quantitatively assessed and classified. Finally, the affected windows are divided into two categories: ordinary dirty windows and high-risk dirty windows, providing a basis for judgment for subsequent differentiated thermal parameter updates.

[0169] Furthermore, step 104 may include the following sub-steps:

[0170] S41. Extract the change intensity data corresponding to each affected window in the set of affected windows based on the map change fragment data;

[0171] S42. Extract the structural complexity data corresponding to each affected window based on the window index table and the local geometric data of the new layout in the new layout file;

[0172] S43. Extract the thermal parameter confidence data for each affected window based on the thermal parameter table and the version metadata table;

[0173] S44. Based on the preset classification and judgment rules, the multiple affected windows are divided into ordinary dirty windows and high-risk dirty windows by using the change intensity, structural complexity and thermal parameter confidence data corresponding to each affected window.

[0174] It should be noted that after generating the set of affected windows, the system constructs a dirty window list. To control costs, the system further classifies dirty windows. Preferably, they are divided into ordinary dirty windows and high-risk dirty windows. Ordinary dirty windows can be updated using fast update strategies, such as calculating new parameters based on local geometric difference correction, lookup table correction, feature increment correction, or approximate re-estimation methods; high-risk dirty windows, on the other hand, employ local high-precision update strategies, such as local 3D reconstruction, local RVE re-solution, or local high-fidelity hot network solution.

[0175] Specifically, this step uses the previously output set of affected windows as the core input, while simultaneously retrieving the map change fragment data already loaded in the process, the window index table, thermal parameter table, version metadata table from the old multi-scale thermal parameter database, and the new map file. All data is precisely matched using the window's unique identifier as the primary key. First, for each window in the set of affected windows, its corresponding map change fragment data is matched using the window's unique identifier. From each fragment, three core quantitative indicators are extracted: the percentage of map change area within the window, the type of change, and the impact level of the heat conduction path. Then, according to preset parameters... The weighted algorithm integrates and calculates to obtain the change intensity data corresponding to each window. This data is used to quantify the direct impact of the current layout change on the current window. Then, based on the window index table, the spatial coordinates, process layer, and multi-scale layer association information of each affected window are queried. According to the queried coordinate range, the corresponding local geometric data of the new layout is extracted from the new layout file. The total number of layout primitives, wiring density, number of inter-layer vias, and geometric structure dispersion within this area are statistically analyzed. Combining the statistical results, the structural complexity data corresponding to each affected window is extracted to characterize the internal structure of the window. The complexity of the map structure is assessed. Then, using the unique window identifier as the search condition, the fluctuation range and parameter update deviation of the historical thermal parameters for that window are queried in the thermal parameter table, and the corresponding model calculation iterations, anomaly marker status, and historical change frequency are retrieved from the version metadata table. Multiple indicators are combined to complete the quantitative scoring, resulting in the thermal parameter confidence data for each affected window. This data directly reflects the reliability and stability of the window's historical thermal parameter results. Finally, a pre-configured classification rule is called. This rule assigns fixed weights to change intensity, structural complexity, and thermal parameter confidence, and divides the data into two levels of comprehensive risk score thresholds. First, the three data points for each window are weighted and summed to obtain a comprehensive risk score. Then, windows with a comprehensive risk score below the low threshold are uniformly classified as ordinary dirty windows, and windows with a comprehensive risk score reaching or exceeding the high threshold are classified as high-risk dirty windows. Finally, all affected windows are classified, and two sets of windows are output. Ordinary dirty windows can subsequently adopt a fast update strategy based on local geometric difference correction, table lookup correction, feature increment correction, or approximate reestimation. High-risk dirty windows can subsequently adopt local 3D reconstruction and local RVE (Representative Value Evaluator). A high-precision update strategy using Volume Element (RVE) re-solution or local high-fidelity hot network solution is adopted to control the overall computational cost and achieve differentiated updates.

[0176] For example, the layout change segment number is CP002, and the modification type includes M2 metal boundary fine-tuning and V2 via addition. The affected window set covers three scale windows: fine, medium and coarse. The preset classification judgment rules are configured: change intensity weight 0.4, structural complexity weight 0.3, and thermal parameter confidence weight 0.3. The full score of each of the three individual indicators is 100 points. The comprehensive risk score of 0~60 points is classified as ordinary dirty window, and 60 points and above are classified as high-risk dirty window. 1. Processing the fine-scale window W1_M2_120_90: Matching the CP002 layout change segment using the window's unique identifier, the changed area of ​​this window accounts for 100%, and the change is a routine minor adjustment of the metal boundary, with a low impact on the heat conduction path. The calculated change intensity is 26.15 points. Extracting the local geometric data of the new layout of this window, the number of primitives in the area is small, the wiring density is low, and there are no interlayer vias. After statistics, the structural complexity is 15 points. Searching the thermal parameter table and the version metadata table, the historical thermal parameter fluctuation range of this window is small, there are no abnormal markers, and the historical change frequency is low, with a thermal parameter confidence level of 90 points. The weighted comprehensive score is 26.15×0.4+15×0.3+90×0.3=40.96 points, which is lower than the 60-point threshold, and is judged as a normal dirty window. 2. Processing fine-scale window W1_V2_121_91: This window corresponds to the newly added area of ​​V2 vias, with a changed area ratio of 100%. The modification is a via addition related to the core hot channel, resulting in a high impact level on the heat conduction path, with a change intensity score of 88. The window contains a large number of via primitives, with high wiring density and geometric dispersion, resulting in a structural complexity score of 82. This window is located in a preset high thermal sensitivity area, with significant historical thermal parameter fluctuations, and a thermal parameter confidence score of 30. The weighted comprehensive score is 88×0.4+82×0.3+30×0.3=70.8, reaching the 60-point threshold, thus classifying it as a high-risk dirty window. By traversing all windows within the affected window set according to the above unified process, the classification and judgment of all windows can be completed in batches.

[0177] Among them, the change intensity data is a quantitative data used to characterize the degree of direct impact of the map change on the affected window. It consists of indicators such as the area ratio of the map change within the window, the type of change, and the level of influence of the heat conduction path. The higher the value, the stronger the direct impact of the map change on the window.

[0178] Structural complexity data is a quantitative data used to characterize the complexity of the geometric structure of the new layout within the affected window. It consists of indicators such as the number of graphic units within the window, wiring density, number of interlayer vias, and geometric dispersion. The higher the value, the more complex the layout structure within the window and the higher the uncertainty in the calculation of thermal parameters.

[0179] Thermal parameter confidence data is a quantitative data used to characterize the reliability and stability of historical thermal parameter data in the affected window. It consists of indicators such as the fluctuation range of historical thermal parameters in the window, the number of calculation iterations, deviation value records, and abnormal marking status. The lower the value, the worse the reliability of the historical thermal parameters and the higher the risk of updating.

[0180] It's worth noting that the rapid update module is not limited to using differential features and fast models. Local lookup table correction, template library matching, simplified thermal resistance network correction, or lightweight surrogate model recalculation can also be used. The core principle is that the ordinary dirty window does not perform a complete high-precision recalculation, but instead completes parameter updates at a low cost.

[0181] Furthermore, the use of finite element method or Recurrent Waveform (RVE) is not limited to updating high-risk windows. Finite volume method, boundary element method, local thermal network refinement, frequency domain heat propagation solution, or other high-fidelity local solution methods can be used instead. As long as the function is to update high-risk dirty windows with high accuracy, it conforms to the technical concept of this invention.

[0182] Step 105: Using a preset fast update model, update the thermal parameters of ordinary dirty windows and high-risk dirty windows respectively based on the new version of the drawing file, the old version of the multi-scale thermal parameter database and PDK process data, and output the dirty window updated thermal parameter set.

[0183] The preset fast update model is a programmable, multi-strategy fusion model built for the incremental update process of multi-scale thermal parameters of chips. It is divided into two independent modules: a fast correction sub-model and a high-precision solution sub-model. The two modules share a unified input interface and output format, and automatically route to the corresponding sub-model based on the dirty window hierarchical labeling. The fast correction sub-model is designed for ordinary dirty windows and integrates four lightweight algorithms: local geometric difference correction, table lookup correction, feature increment correction, and approximate reestimation. The input is the local geometric data of the new layout corresponding to the window, the old thermal parameter data, and the PDK process basic parameters. The output is the window equivalent thermal parameters obtained by fast correction. The high-precision solution sub-model is designed for high-risk dirty windows, integrating three types of high-precision algorithms: local 3D reconstruction, local RVE re-solution, and local high-fidelity thermal network solution. The inputs are the local 3D structural data of the new layout corresponding to the window, the old thermal parameter data, and the thermal properties and interface thermal resistance data of the PDK process material. The output is the window's equivalent thermal parameters obtained through high-precision solution. The model has built-in strategy matching logic, which can automatically select the appropriate algorithm to perform the update based on the dirty window's hierarchical label, window scale level, and associated change type. It also records the input data, execution steps, and intermediate verification values ​​of the update process in a versioned manner to ensure the traceability and consistency of the update results.

[0184] The dirty window updated thermal parameter set is a standardized dataset formed by organizing ordinary dirty windows and high-risk dirty windows into a structured format after differentiated thermal parameter updates. The set uses the unique window identifier as the index key. Each window entry includes fields such as the unique window identifier, window scale level, the map layer to which the window belongs, the updated thermal parameter values ​​(including core parameters such as anisotropic thermal conductivity, equivalent thermal resistance, and heat capacity), update strategy type marker (including fast correction / high-precision solution and specific algorithm name), update timestamp, associated map change record ID, thermal parameter confidence score, and update version number. The set is partitioned and stored according to window scale level and map layer to which it belongs, providing a structured and directly callable data source for subsequent multi-scale thermal parameter aggregation calculations and updates to the old multi-scale thermal parameter database.

[0185] It should be noted that this step follows the previously obtained classification of ordinary dirty windows and high-risk dirty windows. Relying on the new version of the drawing file, the old version of the multi-scale thermal parameter database and PDK process data, it calls the preset fast update model to match differentiated update strategies for the two types of windows, completes the incremental update of thermal parameters for each type of window, summarizes the update results of all windows, and outputs the set of updated thermal parameters for dirty windows.

[0186] Furthermore, step 105 may include the following sub-steps:

[0187] S51. Extract the local geometric difference feature data corresponding to the ordinary dirty window from the new version of the drawing file, and read the reference thermal parameter data corresponding to the ordinary dirty window from the old version of the multi-scale thermal parameter database.

[0188] S52. Update the feature parameters based on the local geometric difference feature data to obtain the updated feature parameters;

[0189] S53. Input the updated feature parameters and baseline thermal parameter data into the preset fast update model, and calculate the new thermal parameter data corresponding to the ordinary dirty window.

[0190] S54. Extract the local geometric data corresponding to the high-risk dirty window from the new drawing file, and construct a local three-dimensional high-precision model corresponding to the high-risk dirty window in combination with PDK process data.

[0191] S55. Apply preset boundary conditions to the local three-dimensional high-precision model and solve for the local heat flow response data.

[0192] S56. Based on the local heat flow response data, calculate the updated equivalent thermal conductivity tensor and supporting thermal parameter data of the high-risk dirty window.

[0193] S57. Integrate the new thermal parameter data of ordinary dirty windows, the updated equivalent thermal conductivity tensor and supporting thermal parameter data of high-risk dirty windows, and output the updated thermal parameter set of dirty windows.

[0194] It's important to note that in the fast update path, the system doesn't need to completely reconstruct all window features. Instead, it preferentially performs differential updates only on the locally changed features. For example, if a new set of vias is added to a fine-scale window, the system can update only the via count, vertical connectivity, local contact area ratio, and related interface parameters, and then obtain the updated thermal parameters based on a pre-trained rule model or fast mapping model (i.e., a preset fast update model). If the change only involves fine-tuning of local metal boundaries, the system can update only the metal area fraction, principal direction length density, and boundary truncation factor, without reprocessing irrelevant features. Through this differential feature update mechanism, the processing cost of ordinary dirty windows is significantly lower than that of full reconstruction.

[0195] In high-risk update paths, the system preferentially constructs a local high-precision model on the corresponding dirty window. This local high-precision model can be reconstructed based on the local geometry of the new map, layer thickness and material information in the PDK, interface thermal resistance parameters, and contact definitions. The system can apply unit temperature difference boundary conditions along the x, y, and z directions respectively to obtain the local heat flow response, and then inversely derive the updated equivalent thermal conductivity tensor. This tensor not only replaces the old parameters of the current dirty window, but can also be used to correct the approximate results of its surrounding propagation windows.

[0196] Specifically, this step takes the ordinary dirty windows and high-risk dirty windows obtained from the previous classification as the processing objects. First, it processes the ordinary dirty windows, extracting the local geometric difference feature data corresponding to each ordinary dirty window from the new version of the drawing file. If a new via is added within the window, the via count, vertical connectivity, local contact area ratio, and related interface parameters are extracted. If the window only involves local metal boundary fine-tuning, the metal area fraction, principal direction length density, and boundary truncation factor are extracted. At the same time, the baseline thermal parameter data corresponding to the ordinary dirty window is read from the old version of the multi-scale thermal parameter database. Then, based on the extracted local geometric difference feature data, only the changed local features are differentially updated to obtain the updated feature parameters. The updated feature parameters and the baseline thermal parameter data are then input into the fast correction sub-model in the preset fast update model. The calculation is performed through a pre-trained rule model or fast mapping model. The process involves calculating new thermal parameters for ordinary dirty windows, followed by processing high-risk dirty windows. Local geometric data for each high-risk dirty window is extracted from the new layout file and combined with layer thickness, material information, interface thermal resistance parameters, and contact definitions from the PDK process data to construct a local 3D high-precision model for the high-risk dirty window. Unit temperature difference boundary conditions are then applied to this local 3D high-precision model along the x, y, and z directions to obtain local heat flow response data. Based on this local heat flow response data, the updated equivalent thermal conductivity tensor and associated thermal parameters such as equivalent thermal resistance and heat capacity for the high-risk dirty window are calculated. Finally, the new thermal parameters for all ordinary dirty windows are integrated with the updated equivalent thermal conductivity tensor and associated thermal parameters for the high-risk dirty windows to form a structured set of updated thermal parameters for dirty windows, providing data support for subsequent multi-scale thermal parameter aggregation.

[0197] Among them, the local geometric difference feature data is a set of feature parameters extracted from the new layout file for ordinary dirty windows that are only related to local changes. When the change is the addition of a via, it includes via count, vertical connectivity, local contact area ratio and related interface parameters; when the change is the fine adjustment of local metal boundary, it includes metal area fraction, principal direction length density and boundary cutoff factor, which are used to characterize the impact of local layout changes within the window on heat conduction-related features.

[0198] The baseline thermal parameter data consists of historical equivalent thermal parameter data stored in the old version of the multi-scale thermal parameter database for ordinary dirty windows, including anisotropic thermal conductivity, equivalent thermal resistance, heat capacity, etc. It serves as the baseline data for rapid updates of thermal parameters for ordinary dirty windows.

[0199] The updated feature parameters are a set of parameters obtained by differentially updating the original heat conduction-related feature parameters of the ordinary dirty window based on local geometric difference feature data. It only contains the feature parameters that have changed and does not require processing of irrelevant features. It is used to drive the calculation of the fast correction sub-model.

[0200] The pre-trained rule model is a model within the fast correction sub-model for ordinary dirty windows in the preset fast update model. By pre-training, a mapping rule of "local geometric difference features - thermal parameter correction value" is established, which can quickly calculate the corresponding thermal parameter correction value based on the updated feature parameters.

[0201] The preset boundary conditions are the solution conditions set for the local three-dimensional high-precision model. The preferred boundary conditions are those that apply a unit temperature difference along the x, y, and z directions to simulate the heat conduction environment in different directions and obtain local heat flow response data.

[0202] Local heat flux response data is heat flux distribution data obtained by solving a local three-dimensional high-precision model under preset boundary conditions. It includes information such as heat flux density and temperature gradient in different directions and is the direct basis for inverting the equivalent thermal conductivity tensor.

[0203] The equivalent thermal conductivity tensor is a tensor data obtained by inversion based on local heat flow response data. It is used to characterize the anisotropic heat conduction properties of high-risk dirty windows. It can replace the old thermal parameters of the window and be used to correct the approximate results of the surrounding propagation windows.

[0204] The accompanying thermal parameter data consists of other parameter data that characterize the thermal properties of high-risk dirty windows, which are paired with the equivalent thermal conductivity tensor. These include equivalent thermal resistance, heat capacity, interfacial thermal resistance, etc. Together with the equivalent thermal conductivity tensor, they constitute the complete thermal parameter data of the window.

[0205] Step 106: Using preset aggregation rules and preset database replacement rules, cross-scale thermal parameter updates and compatibility encapsulation are performed based on the dirty window update thermal parameter set, the old version of the multi-scale thermal parameter database, and the window index table in the old version of the multi-scale thermal parameter database, and the new version of thermal parameter data in a compatible format is output.

[0206] The new compatible thermal parameter data is structured thermal parameter data generated after cross-scale aggregation updates and compatible encapsulation. It includes updated thermal parameters, version metadata, and index mapping relationships for all dirty windows and their corresponding parent windows, providing standardized thermal parameter support for subsequent chip thermal analysis. The new compatible thermal parameter data can be in a complete database format or an incremental encapsulation format for use by downstream thermal simulators or multiphysics analysis tools.

[0207] The preset aggregation rules are a programmable cross-scale thermal parameter aggregation calculation standard customized for the multi-scale thermal parameter incremental update process. It takes the updated thermal parameters of the fine-scale dirty window in the dirty window update thermal parameter set as input, and relies on the parent-child hierarchical relationship recorded in the window index table of the old multi-scale thermal parameter database. It completes the thermal parameter aggregation update of the parent window of the medium-scale and coarse-scale in the hierarchical order of "fine scale → medium scale → coarse scale". The rules incorporate trigger condition judgment logic, triggering the aggregation update of the parent window only when there is a dirty window marked for updating in the set of child windows of the parent window, thus avoiding meaningless duplicate calculations. For different thermal parameter types, the rules define corresponding aggregation calculation methods: the equivalent thermal conductivity tensor uses a weighted average algorithm based on the thermal conductivity, volume ratio, and heat flow direction weight of the child window; the equivalent thermal resistance uses a series-parallel equivalent algorithm of the thermal resistance of the child window; and the heat capacity uses a volume-weighted summation algorithm of the heat capacity of the child window. During the aggregation process, the rules incorporate a consistency verification step, which verifies the deviation between the aggregated parent window thermal parameters and the old historical data. Results with deviations exceeding a preset threshold are marked as pending verification, and metadata such as the list of child window IDs, aggregation algorithm version, and verification results involved in the aggregation process are recorded simultaneously.

[0208] The preset database replacement rules are a customized, programmable update and encapsulation rule system designed to achieve seamless compatibility between the new version of thermal parameter data and the old version of the multi-scale thermal parameter database. It includes three core modules: data format adaptation, index mapping synchronization, and version metadata update, covering the entire process requirements from data standardization to database structure compatibility. The data format adaptation module stipulates that the field definitions, data types, unit standards, and precision ranges of the new version of thermal parameter data must be completely consistent with the old version of the database, including the storage order of the anisotropic thermal conductivity tensor, the unit of equivalent thermal resistance, and the numerical precision of heat capacity. The index mapping synchronization module stipulates that all dirty windows and their corresponding parent windows must be updated according to the window index table, including the mapping information of window unique identifier, spatial coordinate range, hierarchical relationship, and cross-level relationship, and the version number and update flag of the index must be updated synchronously. The version metadata update module stipulates that complete metadata of the new version of the database must be generated, including fields such as version number, update time, associated map change record ID, updated window list, aggregation calculation log, and compatibility verification results, while retaining the historical metadata entries of the old version of the database. The rules also have built-in data integrity verification steps to ensure that only the relevant data of dirty windows and their corresponding parent windows are updated, the data of unaffected windows remain unchanged, and the overall structure and interface definition of the updated database are fully compatible with the old version.

[0209] It should be noted that the process takes over the dirty window update set generated above, relies on the old multi-scale thermal parameter database and its window index table, completes the cross-scale thermal parameter aggregation update of the mesoscale and coarse-scale parent windows according to the preset aggregation rules, and then follows the preset database replacement rules to encapsulate the updated thermal parameters in a way that is compatible with the format and index structure of the original database, generating a new version of thermal parameter data that can be directly called.

[0210] Furthermore, step 106 may include the following sub-steps:

[0211] S61. Based on the parent-child index of the window index table and the set of hot parameters for dirty window updates, determine the parent window to be passively updated;

[0212] S62. Read the historical thermal parameter data of the child windows that have not been updated under the parent window to be passively updated from the old multi-scale thermal parameter database;

[0213] S63. Combine the historical thermal parameter data of the unupdated child window with the new thermal parameter data of the updated child window in the dirty window update thermal parameter set, and use the preset aggregation rules to calculate the complete update thermal parameter data of the parent window to be passively updated.

[0214] S64. Based on the preset database replacement rules, use dirty windows to update the thermal parameter set and complete update the thermal parameter data. Perform partial replacement and parameter update on the corresponding window records in the old multi-scale thermal parameter database to obtain the new multi-scale thermal parameter database.

[0215] S65. Read the complete data of the new multi-scale thermal parameter database according to the preset downstream compatible interface format, and perform data format adaptation and encapsulation on the complete data to output the new thermal parameter data in a compatible format.

[0216] It should be noted that after completing the fine-scale window update, the system performs a cross-scale write-back. The write-back process preferably does not involve simple averaging, but rather recalculates the mesoscale and coarse-scale parameters according to preset aggregation rules. Aggregation rules may include: planar aggregation based on area or volume weighting; directional consistency correction based on the dominant thermal direction; vertical thermal conduction aggregation based on series-parallel equivalence relationships; coarse-scale parameter correction based on heat flux conservation; and bounded constraints on the eigenvalues ​​of the anisotropic tensor. Through this process, the system ensures that the new parameters of the parent scale window are consistent with the results of its child windows, avoiding inconsistencies between the new fine-scale parameters and the old coarse-scale parameters.

[0217] During the database replacement phase, the system only replaces the records of the dirty window and its passively updated parent window in the thermal parameter table and version metadata table, and writes new version numbers, update timestamps, and update source identifiers to these records. Records not included in the dirty window set are retained as is. Finally, the output interface module provides the new database to downstream thermal simulators in an interface compatible with the old version, allowing thermal analysis after ECO to directly use the updated parameters without changing the downstream tool interface.

[0218] Specifically, processing is carried out based on the window index table in the old multi-scale thermal parameter database. The parent-child indexes stored in this window index table were uniformly generated and permanently fixed during the overall construction phase of the multi-scale thermal parameter system, based on the inherent pyramid hierarchical relationship of fine-scale, meso-scale, and coarse-scale windows. This clearly records the hierarchical relationship mapping of windows at each level. This step directly calls this pre-stored index without real-time calculation. At the same time, the preset database replacement rules used in this step are standardized specifications that were pre-compiled and fixed in the early stage of solution deployment, combining the table structure, field format, storage logic of the old multi-scale thermal parameter database and the docking standards of downstream thermal simulation and multiphysics tools. The rules clearly define four execution requirements: First, limit the replacement scope, only allowing operations on window records where parameters have changed; second, limit the rewriting method, using new parameter data to overwrite the original parameter fields in situ; third, force synchronous updates of metadata fields such as version number, timestamp, and change source; and fourth, strictly retain existing data, without making any modifications to the overall database architecture, basic indexes, or window records that have not changed.In the overall process, the unique identifier of each updated fine-scale window in the dirty window update thermal parameter set is used as the primary key for retrieval. Based on the pre-stored parent-child index, the corresponding mesoscale and coarse-scale parent windows are traced upwards layer by layer. All parent windows at all levels that contain updated child windows are uniformly marked as parent windows to be passively updated, and a parent-child window association list is established simultaneously. Then, for each parent window to be passively updated, historical thermal parameter data such as thermal conductivity tensor, equivalent thermal resistance, and heat capacity of all child windows within the window's jurisdiction that are not included in the dirty window update thermal parameter set are read in batches from the thermal parameter table of the old multi-scale thermal parameter database, based on the window's unique identifier. Next, the historical thermal parameter data of unupdated child windows under the same parent window are integrated and grouped with the new thermal parameter data of updated child windows, and calculations are performed according to preset aggregation rules. Specifically, the equivalent thermal conductivity tensor uses a window volume-weighted average algorithm, the equivalent thermal resistance uses a series equivalent algorithm, and the heat capacity uses a volume summation algorithm. The complete updated thermal parameter data of the parent window to be passively updated is calculated by combining the area and volume ratio of the multi-scale window. Then, the database update operation is strictly performed according to the preset database replacement rules. According to the replacement range defined by the rules, only the database records corresponding to the dirty child window and the parent window to be passively updated are selected. The original parameter fields in the thermal parameter table are replaced with new thermal parameter data by in-situ overwrite. At the same time, according to the metadata update requirements, the version number, update timestamp, change source, update type and other auxiliary fields of the corresponding records are rewritten synchronously. All other window records, data table structures and original indexes in the database that have not been changed are completely preserved. This generates a new version of the multi-scale thermal parameter database. Finally, according to the preset downstream compatible interface format corresponding to downstream thermal simulation and multiphysics tools, all data of the new version of the multi-scale thermal parameter database is completely read. The data fields, numerical accuracy and communication protocol are uniformly calibrated. After format adaptation and standardized encapsulation, the new version of thermal parameter data in a compatible format is output.

[0219] For example, the fine-scale window size is 1μm×1μm, the mesoscale window is 10μm×1μm, and the coarse-scale window is 100μm×100μm. The parent-child index predefines one mesoscale window corresponding to 100 fine-scale windows, and one coarse-scale window corresponding to 100 mesoscale windows. Pre-defined database replacement rules are already fixed, clearly stating that only window changes can overwrite parameters in situ, version / timestamp / change source must be updated synchronously, and non-changed data and database structure are completely preserved. This dirty window update thermal parameter set contains 15851 fine-scale windows that have completed parameter updates. The first step, relying on the pre-stored parent-child index, traces upwards to obtain 169 mesoscale parent windows and 1 coarse-scale parent window, all marked as parent windows awaiting passive updates, and a parent-child relationship list is compiled. The second step selects a typical mesoscale parent window W2_120_90, and reads the historical thermal parameters of its 42 subordinate unupdated fine-scale windows from the old multi-scale thermal parameter database thermal parameter table. The historical average thermal conductivity of this group of windows is 0.85 W / ( The equivalent thermal resistance is 0.025 K / W; the third step integrates the thermal parameters of all new and old sub-windows under this window, and calculates them according to preset aggregation rules such as volume-weighted average and series equivalence to obtain the complete updated thermal parameters of this mesoscale parent window: comprehensive thermal conductivity 0.82 W / ( The first step is to strictly apply the preset database replacement rules to perform the update, operating only on the records of 15851 dirty fine windows, 169 mesoscale parent windows, and 1 coarse-scale parent window. The original thermal conductivity, equivalent thermal resistance, and other fields are overwritten in situ in the thermal parameter table, and the version number of these records is simultaneously updated from V1.0 to V2.0. The update timestamp and the source tag for this layout ECO change are added. The remaining tens of thousands of unchanged window records, data table structure, and original indexes in the database remain unchanged, ultimately generating a new version of the multi-scale thermal parameter database. The fifth step is to uniformly calibrate the data fields, numerical accuracy, and communication protocol according to the preset compatible interface specifications of the downstream thermal simulator, complete the standardized encapsulation, and output a new version of thermal parameter data in a compatible format that can be directly called by downstream tools.

[0220] Among them, the parent window to be passively updated is the mesoscale and coarse-scale parent window that needs to be synchronously updated across scales because it contains the updated fine-scale dirty window. It is the hierarchical window that needs to be passively updated synchronously after the fine-scale dirty window is updated.

[0221] The historical thermal parameter data of the child windows that have not been updated is the historical thermal parameter data stored in the old multi-scale thermal parameter database of the child windows under the parent window that are not passively updated, which have not entered the dirty window set and have not undergone thermal parameter updates. It is one of the basic data for the aggregation calculation of the parent window's thermal parameters.

[0222] The complete update of thermal parameter data is the final thermal parameter data of the parent window to be passively updated, calculated by combining the new thermal parameter data of the updated child windows and the historical thermal parameter data of the unupdated child windows through preset aggregation rules. This data maintains hierarchical consistency with the thermal parameters of all child windows under the parent window.

[0223] The update source identifier is the identification information written into the record of the new version of the multi-scale thermal parameter database. It is used to mark the update source of the thermal parameter record, including the associated map change record ID, update strategy type, update timestamp, etc., so as to facilitate tracing the source of data update.

[0224] The preset downstream compatible interface format is a predefined data format standard that is fully compatible with the downstream thermal simulator's calling interface. It specifies the field definitions, data structures, and calling interface protocols of the new version of thermal parameter data, ensuring that downstream tools can directly call the updated thermal parameter data without modification.

[0225] It is worth mentioning that the present invention preferably maintains the following core data structures.

[0226] One is the window index table, which records the unique identifier, spatial boundaries, scale, parent window identifier, list of child windows, list of adjacent windows, and cross-layer association index for each window. This table is used to support the influence propagation and write-back process.

[0227] The second is a thermal parameter table, used to record the thermal parameters corresponding to each window. Preferably, it includes at least thermal conductivity or thermal conductivity tensor, and may also include volume heat capacity, equivalent interfacial thermal resistance, thermal diffusivity, and the associated dominant thermal direction, anisotropy ratio, and confidence level.

[0228] Thirdly, there is the version metadata table, which records information such as the version number, update timestamp, update method, data source, and whether it is an actuarial result corresponding to the current parameters of each window.

[0229] Fourth is the dirty window table, which records the set of nodes that are marked as direct change windows, propagation change windows, or aggregate update windows in a certain ECO process, and can be labeled with change type and change intensity.

[0230] These data structures together constitute an incrementally maintainable thermal parameter database.

[0231] Furthermore, the cross-scale write-back method is not limited to simple weighted aggregation. Constrained optimization, Bayesian update, graph propagation write-back, tensor eigenvalue correction, or iterative aggregation based on conservation relationships can also be used. As long as the mechanism ensures that the results at the mesoscale and coarse scales remain consistent after the fine-scale window is updated, the objective of this invention can be achieved.

[0232] Furthermore, the database implementation is not limited to a specific data storage medium. Local file databases, relational databases, key-value databases, in-memory databases, graph databases, or sets of parameter files for thermal analysis tools can be used. The database is also not required to exist independently; it is sufficient that the system can maintain window-level parameter records, version numbers, and replacement relationships.

[0233] Finally, regarding layout input, although GDSII or OASIS is preferred, any layout representation capable of recovering local geometry and layer information can also be used, as long as it can still support ECO difference recognition and multi-scale window mapping, and should be considered within the scope of protection of this invention.

[0234] For example, such as Figure 2 As shown, for example, suppose a design already has a three-layer thermal parameter database (coarse, medium, and fine). In a certain ECO cycle, only one set of vias and a section of metal trace are modified. If the traditional method is used, the thermal parameters of the entire region or even the entire chip need to be extracted again. However, in this invention, the system only needs to identify a few affected fine-scale windows, and then update a small number of medium-scale and coarse-scale windows according to the parent-child relationship and propagation rules. Specific practical examples are as follows:

[0235] Assuming the old layout represents a multi-layered BEOL region, a multi-scale thermal parameter database has been established with fine-scale windows of 10μm side length, medium-scale windows of 40μm side length, and coarse-scale windows of 160μm side length. During an ECO (Environmental Change Operation), the designer added a set of vias near a local hotspot and adjusted the direction of the top-layer metal traces. After reading the old and new layouts, the system compares the layer differences to obtain two change segments: one representing the newly added region of the via array, and the other representing the adjusted region of the top-layer metal boundary. The system maps these two change segments to corresponding fine-scale windows, resulting in five directly changed fine windows. Subsequently, based on cross-layer propagation rules, three fine windows connected to the upper and lower thermal paths of the newly added vias are included in the propagation set; based on spatial neighborhood rules, two fine windows adjacent to the metal boundary adjustment window and within a distance of less than a threshold are included in the propagation set. This ultimately results in ten fine-scale dirty windows.

[0236] Then, the system categorizes the 10 fine windows based on the type of change. For the six windows with only minor boundary adjustments and low structural complexity, a fast update path is adopted, updating only the metal area fraction, principal direction length density, and boundary truncation factor, and using a fast mapping model to obtain new thermal parameters. For the four windows where the addition of vias causes abrupt changes in vertical thermal channels, the system classifies them as high-risk windows and enters a local high-precision solution path. After solving, the system obtains new thermal conductivity tensors for the four windows. Subsequently, the system aggregates the new results of the 10 fine windows into the corresponding three mesoscale windows and one coarse-scale window according to parent-child relationships, and updates the equivalent parameters of these parent windows. Finally, only the records of these 10 fine windows, three mesoscale windows, and one coarse-scale window are replaced in the database, while all other records remain unchanged. Compared with a full recalculation, the update cost is significantly reduced, while the impact of local changes on coarse-scale thermal analysis is simultaneously reflected.

[0237] In summary, the purpose of this invention is not to simply repeat existing layout and thermal parameter technologies, but to establish a complete incremental maintenance system around the high-frequency engineering scenario of ECO, from change detection, impact propagation, local recalculation, cross-scale write-back to database replacement.

[0238] For comparison of technical effectiveness, existing technologies can be used as a reference. Currently, there are roughly three main approaches to thermal modeling related to layout. The first approach involves importing the ECAD layout, reconstructing its 3D geometry, and then directly performing thermal analysis using a multiphysics solver. Current technical solutions can convert files such as GDS-II into 3D geometric models, but layer thickness and stackup information need to be supplemented. The advantage of this approach is that the thermal physics is intuitive and suitable for high-precision local analysis. However, its problem is that once the target layout is very large, has many layers, and has very detailed geometry, the modeling and solving costs increase rapidly.

[0239] The second approach is the multi-scale homogenization approach. The published paper, "A Multiscale Workflow for Thermal Analysis of 3DI Chip Stacks," proposes automatically constructing local Relative Thermal Vessels (RVEs) from GDSII / OASIS at required locations during macroscopic thermal analysis and solving for anisotropic homogenized thermal conductivity online. This demonstrates that automatically constructing local thermal structures from the layout and calculating anisotropic thermal parameters has a relatively clear public basis. The advantage of this approach is that it can balance the realism of local microstructures with the computability of macroscopic analysis; however, it focuses more on how to obtain parameters from local sources during a single analysis and does not address how to locally update parameters after an ECO as the main problem.

[0240] The third approach is based on rapid prediction using physical features and machine learning. A 2026 Microelectronics Journal paper, "Physical feature-based machine learning of BEOL thermal conductivity in 3D ICs," proposed transforming the local structure of the BEOL into physical features related to heat transfer, and then using machine learning to quickly predict thermal conductivity, claiming to maintain anisotropic trends and spatial distribution characteristics. This indicates that rapid prediction of local directional thermal parameters is no longer a gap in technology. This type of method is more suitable for rapid inference over a large scale, but the paper focuses on the mapping process from features to prediction, without fully addressing the core technical loop of how to compare multiple versions of the layout, which regions should be recalculated, and how to update across scales after recalculation.

[0241] Besides academic papers, patent publications have also revealed solutions related to post-layout thermal modeling. For example, Google Patents' US9721059B1 and US20170212978A1, which cover post-layout thermal-aware integrated circuit performance modeling, demonstrate that this general direction of post-layout thermal modeling has already been disclosed. Other published materials involve chip-scale electrothermal analysis, thermal network modeling, and cross-scale simulation, all of which primarily focus on heat extraction models from the layout or the thermal parameters generated by the layout itself.

[0242] Therefore, based on the prior art closest to this invention, it can be considered that the existing solutions already cover the following: first, thermal model construction based on layout data; second, obtaining anisotropic thermal parameters through local homogenization; third, rapidly predicting the directional thermal conductivity of BEOL through physical characteristics and data-driven methods; and fourth, combining post-layout thermal modeling with chip performance analysis or thermal signature. The existing technology has a fairly rich public foundation.

[0243] However, existing disclosures have not systematically presented the following complete closed loop: when the layout undergoes local changes due to ECO, how to first identify the changed units at the geometric level, then map the changes to thermal parameter windows, then expand the dirty window set according to the thermal effect propagation rules, then perform hierarchical updates on the dirty windows, then write back the update results across scales and replace the old records in the thermal parameter database, and finally continue to use the updated database for downstream thermal simulations. In other words, existing disclosures are more like answering how to generate thermal parameters, while this invention focuses on how to maintain the generated thermal parameters during ECO iterations. This is precisely the core of this invention.

[0244] Based on the above, the existing method has the following drawbacks:

[0245] (1) Most existing technologies treat thermal parameter generation as a one-time batch processing task, rather than a data maintenance process that evolves continuously with each layout version. Whether it is importing ECAD into 3D solution, local RVE online homogenization, or machine learning for rapid prediction, the typical process is to output the current parameters given the current layout, rarely discussing how to reuse the calculation results of the previous version when the next layout arrives. As a result, whenever an ECO occurs in the layout, the most common practice in engineering is often to recalculate globally or on a large scale, leading to serious waste of computing resources. This invention addresses this shortcoming by providing an incremental update method for thermal parameters oriented towards ECO, enabling the reuse of existing thermal parameter results without having to start from scratch after each change.

[0246] (2) Existing technologies typically lack mechanisms for change identification and impact propagation. In practice, when a layout changes, it's not just the geometrically directly altered regions that need updating. Some changes can spread to a larger range of thermal parameters through via chains, interface thermal resistance, local thermal channel connectivity, or window parent-child aggregation relationships. Updating only the geometrically altered regions will miss windows that are thermally affected but whose geometry has not directly changed; recalculating the entire layout is too crude. Existing technologies do not adequately discuss the differences between these altered regions and affected regions. Therefore, this invention proposes an impact propagation and dirty window expansion mechanism, the purpose of which is to more accurately control the update boundary.

[0247] (3) Existing technologies generally do not unify multi-scale organization and ECO update mechanisms. Existing multi-scale schemes mainly discuss how to calculate parameters at different scales, and rarely discuss how to passively update coarse-scale parameters after local changes at fine scales, and how to avoid conflicting states where old coarse parameters are matched with new fine parameters. One of the objectives of this invention is to establish a cross-scale write-back and consistency replacement mechanism so that the thermal parameter database can maintain overall consistency after local updates.

[0248] (4) While existing fast prediction methods are efficient, they lack sufficient confidence control over local anomalous structures, samples outside the training domain, and interface-dominated regions. Existing high-precision solution methods, while accurate, are too expensive to apply to all dirty windows after ECO. Current technologies generally do not combine rapid updates for ordinary change windows with fine updates for high-risk change windows into a dedicated ECO process. Therefore, this invention proposes a tiered update strategy, aiming to apply different update costs to change windows of different risk levels to optimize efficiency and accuracy.

[0249] (5) Existing technologies typically output static results, rather than database-based results that support version replacement, partial overlay, time tracking, and interface compatibility. The purpose of this invention is also to organize thermal parameter results into a traceable, replaceable, and incrementally maintainable data structure, enabling it to be embedded in the design process, rather than being a one-time analysis product.

[0250] As can be seen from the above, the technical problem to be solved by the present invention is not, in a general sense, how to calculate thermal parameters from a layout, but more specifically, for the ECO scenario, how to efficiently, correctly, and controllably update multi-scale thermal parameters locally after a local change in the layout.

[0251] First, in advanced chip and packaging design flows, the layout typically undergoes multiple rounds of ECOs before final approval. ECOs may involve metal trace adjustments, via array changes, local fill modifications, contact boundary corrections, macroblock edge shape adjustments, and even bond region relocation. While these changes are often geometrically local, their thermal impact is not necessarily confined to the geometrically altered area, as local heat conduction channels, cross-layer heat channels, and coarse-scale equivalent parameters may all change. Existing thermal modeling workflows mostly treat each thermal parameter extraction as an independent batch processing task, lacking a closed loop for ECO-oriented change identification—impact propagation—local recalculation—cross-scale write-back—database replacement. Therefore, the primary technical problem this invention addresses is: when the layout undergoes only local changes, how to accurately identify the range of thermal parameters that truly need updating, rather than performing indiscriminate recalculation of the entire layout.

[0252] Secondly, ECO not only affects windows at a single scale, but may also influence equivalent results at higher scales through spatial inclusion relationships and thermal conduction correlations. For example, a modification to the via array within a fine-scale window will change the local vertical thermal channel capability; if the fine-scale window belongs to a mesoscale or coarse-scale aggregation region, then the corresponding mesoscale and coarse-scale equivalent thermal parameters should also be updated synchronously. Although existing publicly available solutions have addressed multi-scale thermal analysis and local RVE homogenization, their focus is mainly on how to derive parameters from the layout or how to homogenize online, without systematically solving the problem of how local changes are propagated and updated within a multi-scale thermal parameter system. Therefore, the second technical problem this invention aims to solve is: how to establish a controllable propagation and consistent write-back mechanism for ECO changes across multiple scales (fine, meso, and coarse).

[0253] Furthermore, even if the changed region has been identified, applying uniform high-precision recalculation to all windows within that region could still be too costly; conversely, using only fast approximation replacement might result in a loss of accuracy. Therefore, this invention also addresses a third technical problem: how to implement a tiered update strategy for affected windows during the ECO incremental update process, enabling ordinary changed windows to use fast updates while triggering local high-precision corrections for high-risk changed windows, thereby achieving a balance between efficiency and accuracy.

[0254] Furthermore, many existing processes lack maintainability of results, often directly outputting static files after a single generation of hot parameters, which is detrimental to subsequent version iterations and updates. Faced with frequent ECOs, hot parameter results without version management, timestamps, window mapping, dirty marking, and replacement mechanisms are difficult to form sustainably maintainable data assets. Therefore, this invention also addresses a fourth technical problem: how to organize multi-scale hot parameters into a database structure that allows for incremental updates, partial replacements, and version tracking.

[0255] Finally, this invention also aims to address the fifth technical problem: how to achieve partial updates of upstream thermal parameters without altering downstream thermal simulation interface habits, ensuring that the updated database can still be directly accessed by existing thermal analyzers and multiphysics tools. In other words, this invention aims to provide not a one-time analysis script, but a thermal parameter incremental update technology that can be embedded in the design iteration process and is compatible with existing toolchains.

[0256] Specifically, this invention provides a method, system, and storage medium for incremental updates of multi-scale thermal parameters for ECO (Environmental Cost Allocation) of a map layout. Based on an existing multi-scale thermal parameter database corresponding to an older map layout, upon receiving a new map layout, instead of directly reconstructing all thermal parameters, it first establishes a difference mapping between the old and new maps. Then, it identifies directly changed regions based on the difference mapping, and subsequently obtains a set of affected windows according to thermal effect propagation rules, marking this set as a dirty window set. The system performs rapid updates or local high-precision updates on different windows in the dirty window set according to preset rules. The updated fine-scale results are then written back up to the corresponding mesoscale and coarse-scale nodes. Finally, only the corresponding window records in the database are replaced, forming a new multi-scale thermal parameter database, while the old records of unaffected windows continue to be reused.

[0257] In short, this invention does not generate a completely new heat map, but rather maintains a multi-scale thermal parameter asset library. The vast majority of content in this library is reused from the old version, with only necessary local areas and their scale correlation results updated. This approach is fundamentally different from existing one-time batch thermal modeling.

[0258] Furthermore, such as Figure 3As shown, the overall framework of this invention includes: a layout version input module, an old thermal parameter database reading module, a difference detection module, a change mapping module, an impact propagation module, a dirty window generation module, an update strategy determination module, a fast update module, a local high-precision update module, a cross-scale write-back module, a database replacement module, and an output interface module.

[0259] The map version input module receives old and new maps, preferably GDSII or OASIS files, but can also be other map representation files that can recover local geometric differences. The old thermal parameter database reading module loads the multi-scale thermal parameter database corresponding to the old map and reads information such as window coordinates, scale level, parent-child index, thermal parameter values, version number, and timestamp. The difference detection module performs structural difference comparison between the old and new maps, obtaining geometric change segments, layer change segments, via change segments, and contact change segments. The change mapping module maps these change segments to the thermal parameter window system, obtaining a set of directly changed windows. The influence propagation module expands the directly changed windows into a set of affected windows according to preset spatial neighborhood rules, cross-layer heat conduction rules, parent-child scale rules, and local thermal channel rules. The dirty window generation module deduplicates, classifies, and marks the affected window set, generating a dirty window list. The update strategy determination module determines whether a window should be updated quickly or locally with high precision based on the window category, change intensity, structural complexity, and existing confidence level. The rapid update module performs low-cost parameter correction, while the local high-precision update module performs high-precision recalculation. The cross-scale write-back module aggregates the updated fine-scale results upwards to the mesoscale and coarse-scale nodes, ensuring consistency. The database replacement module only replaces records related to dirty windows, leaving other records unchanged. The output interface module outputs the new database or compatible format files to the thermal simulator or multiphysics platform.

[0260] In summary, the key point and intended protection of this invention does not lie in the generalized derivation of thermal parameters from layout data. This invention is not a simple repetition of existing layout-to-thermal-parameter routes, but rather shifts the focus to the truly high-frequency, truly expensive, and most lacking in systematic solutions link in the design iteration process: the thermal parameter maintenance problem after ECO (Environmental Change Order). This invention establishes a complete chain—including legacy database reuse, difference detection, impact propagation, dirty window hierarchical updates, cross-scale write-back, and database replacement—transforming multi-scale thermal parameters from one-time calculation results into sustainably maintainable engineering data assets. Specifically, this is achieved through the following specific mechanisms.

[0261] First, an ECO incremental maintenance framework based on the collaboration of the old map, the new map, and the old thermal parameter database is proposed, which transforms the problem of generating thermal parameters during ECO into the problem of updating the thermal parameter database.

[0262] Second, the mapping mechanism of map difference detection results to a multi-scale window system, namely, the mechanism of generating a set of fine, medium and coarse multi-scale dirty windows from geometric change fragments.

[0263] Third, the influence propagation mechanism is based on spatial neighborhood, cross-layer thermal channels, scale parent-child relationship and thermal sensitivity rules, that is, the mechanism of expanding the set of affected windows from the directly changing window.

[0264] Fourth, a strategy mechanism for hierarchical updating of dirty windows, namely, using fast differential updates for ordinary dirty windows and local high-precision recalculation for high-risk dirty windows.

[0265] Fifth, a mechanism for consistent rewriting of mesoscale and coarse-scale parameters based on fine-scale update results.

[0266] Sixth, a partial database replacement mechanism based on window record replacement rather than overall database reconstruction.

[0267] Seventh, an incremental update output mechanism that maintains interface compatibility with downstream thermal analyzers.

[0268] Compared to existing layout-based thermal model generation schemes, the advantage of this invention lies not in faster initial thermal parameter generation or more accurate initial local solutions, but in its ability to significantly reduce redundant computations once multiple ECO iterations have begun. Existing solutions primarily address how to obtain the thermal parameters of the current layout, while this invention addresses how to maintain the library to the new version with minimal cost, given that the current layout's thermal parameters are already available and the next version only modifies certain local areas. Due to the frequent ECO iterations in advanced chips, this capability is more valuable in practical engineering than the speed of single-step modeling.

[0269] Furthermore, compared to existing rapid prediction or local homogenization schemes, this invention forms a complete closed loop by combining change detection, impact propagation, hierarchical updates, cross-scale write-back, and database replacement. Therefore, its output is not a one-time static result, but a traceable, replaceable, and version-maintainable multi-scale thermal parameter database. This makes this invention more suitable for embedding in EDA hot approval processes and multi-version design iteration processes, and it is less likely to be simply equated with ordinary thermal parameter extraction.

[0270] Furthermore, the innovation of this technology does not rely on the necessity of using finite element method, finite volume, RVE homogenization, or a certain type of neural network model, but rather on: how to determine which regions need to be updated, how to update them, how to write them back after updating, and how to replace the old results in an ECO scenario.

[0271] In this embodiment of the invention, a multi-scale incremental update method for thermal parameters oriented towards layout ECO is provided. This method acquires old layout files, new layout files, an old multi-scale thermal parameter database, and PDK process data. Based on the old and new layout files, it performs layout difference detection and outputs layout change fragment data. Multi-scale window mapping is performed based on the layout change fragment data and the window index table in the old multi-scale thermal parameter database, outputting a dirty window candidate set. A preset influence propagation rule is used to perform thermal influence propagation analysis on the dirty window candidate set based on the window index table in the old multi-scale thermal parameter database, outputting an affected window set. A preset hierarchical judgment rule is used to classify the affected windows based on the new layout file, the layout change fragment data, and the window index table, thermal parameter table, and version metadata table in the old multi-scale thermal parameter database. The invention employs a set of methods to classify and determine dirty windows, outputting ordinary dirty windows and high-risk dirty windows. A pre-defined fast update model is used to update the thermal parameters of ordinary and high-risk dirty windows based on the new layout file, the old multi-scale thermal parameter database, and PDK process data, outputting a set of updated thermal parameters for dirty windows. Pre-defined aggregation and database replacement rules are used to perform cross-scale thermal parameter updates and compatibility encapsulation based on the updated thermal parameter set for dirty windows, the old multi-scale thermal parameter database, and the window index table in the old multi-scale thermal parameter database, outputting new thermal parameter data in a compatible format. Based on the above scheme, this invention first extracts layout change fragment data through layout difference detection to accurately locate the layout change area. Then, it combines the window index table to complete multi-scale window mapping, obtaining a candidate set of dirty windows, and only filtering out windows associated with layout changes. Subsequently, thermal impact propagation analysis is used for further filtering to determine the set of affected windows linked to layout changes, eliminating windows that do not require thermal parameter updates. This invention classifies affected windows into ordinary dirty windows and high-risk dirty windows, thereby matching corresponding thermal parameter update logic for different types of windows. In the thermal parameter update stage, existing benchmark data from the old multi-scale thermal parameter database is directly called, and incremental parameter updates are achieved based on the existing calculation results. Finally, cross-scale thermal parameter updates and compatibility encapsulation are carried out. Local thermal parameter aggregation is performed only for the changed window and its associated parent window. At the same time, the corresponding window records in the old multi-scale thermal parameter database are partially replaced, completely preserving the original content in the database that has not been changed, without the need for a complete reconstruction of the entire database. This stage also simultaneously completes data format adaptation and encapsulation, saving the additional computational overhead caused by subsequent separate format conversion, effectively reducing invalid computation tasks and overall computational load, thereby avoiding waste of computing resources.

[0272] Please see Figure 4 , Figure 4 This is a structural block diagram of a multi-scale thermal parameter incremental update system for layout ECO provided in Embodiment 2 of the present invention.

[0273] This invention provides a multi-scale incremental update system for thermal parameters oriented to layout ECO, comprising:

[0274] The acquisition module 401 is used to acquire the old layout file, the new layout file, the old multi-scale thermal parameter database, and PDK process data, and to perform layout difference detection based on the old layout file and the new layout file, and output layout change segment data.

[0275] Mapping module 402 is used to perform multi-scale window mapping based on the map change fragment data and the window index table in the old multi-scale thermal parameter database, and output a dirty window candidate set.

[0276] Analysis module 403 is used to perform thermal influence propagation analysis on the dirty window candidate set according to the window index table in the old multi-scale thermal parameter database using preset influence propagation rules, and output the set of affected windows;

[0277] The grading module 404 is used to perform dirty window grading and judgment processing on the affected window set according to the new map file, map change fragment data and window index table, thermal parameter table and version metadata table in the old multi-scale thermal parameter database, and output ordinary dirty windows and high-risk dirty windows.

[0278] The update module 405 is used to update the thermal parameters of ordinary dirty windows and high-risk dirty windows respectively based on the new version of the drawing file, the old version of the multi-scale thermal parameter database and PDK process data using a preset fast update model, and output the dirty window updated thermal parameter set.

[0279] The encapsulation module 406 is used to perform cross-scale thermal parameter updates and compatibility encapsulation based on the dirty window update thermal parameter set, the old multi-scale thermal parameter database and the window index table in the old multi-scale thermal parameter database, using preset aggregation rules and preset database replacement rules, and output new thermal parameter data in a compatible format.

[0280] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0281] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the multi-scale thermal parameter incremental update method for layout-oriented ECO as described in the above embodiments.

[0282] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the multi-scale thermal parameter incremental update method for layout-oriented ECO as described in the above embodiments.

[0283] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0284] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0285] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0286] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0287] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-scale incremental update method for thermal parameters oriented towards layout ECO, characterized in that, include: Acquire old layout files, new layout files, old multi-scale thermal parameter database, and PDK process data; and perform layout difference detection based on the old layout files and the new layout files, outputting layout change segment data. Based on the map change fragment data and the window index table in the old multi-scale thermal parameter database, multi-scale window mapping is performed to output a dirty window candidate set. Using preset influence propagation rules, thermal influence propagation analysis is performed on the dirty window candidate set based on the window index table in the old version of the multi-scale thermal parameter database, and the affected window set is output. The affected window set is classified and judged according to the new map file, the map change fragment data and the window index table, thermal parameter table and version metadata table in the old multi-scale thermal parameter database, and the preset classification and judgment rules are used to output ordinary dirty windows and high-risk dirty windows. A preset fast update model is used to update the thermal parameters of the ordinary dirty window and the high-risk dirty window respectively based on the new version of the drawing file, the old version of the multi-scale thermal parameter database and the PDK process data, and outputs the dirty window updated thermal parameter set. Using preset aggregation rules and preset database replacement rules, cross-scale thermal parameter updates and compatibility encapsulation are performed based on the dirty window update thermal parameter set, the old version of the multi-scale thermal parameter database, and the window index table in the old version of the multi-scale thermal parameter database, and the new version of thermal parameter data in a compatible format is output.

2. The multi-scale incremental update method for thermal parameters oriented towards layout ECO according to claim 1, characterized in that, The step of performing map difference detection based on the old map file and the new map file, and outputting map change fragment data, includes: By comparing the old map file with the new map file according to the map layer level, region or graphic element, the changed areas corresponding to multiple preset difference types are identified; Extract the map change fragment data for each of the changed regions.

3. The multi-scale incremental update method for thermal parameters oriented towards layout ECO according to claim 1, characterized in that, The process involves performing multi-scale window mapping based on the map change fragment data and the window index table in the old multi-scale thermal parameter database, outputting a dirty window candidate set, including: Based on the window index table, locate the fine-scale window that intersects with each of the map change fragment data, and mark each fine-scale window as a direct change fine window; Based on the parent-child index of the window index table, trace upwards to the parent window corresponding to the directly changing fine window and mark it as a candidate dirty parent window; Based on the cross-layer association index of the window index table, locate the window that is directly changed and cross-layer associated with the fine window, and mark it as a candidate dirty cross-layer window; Integrate the directly changing fine window, the candidate dirty parent window, and the candidate dirty cross-layer window to output a dirty window candidate set.

4. The multi-scale incremental update method for thermal parameters oriented towards layout ECO according to claim 1, characterized in that, The preset impact propagation rules include spatial neighborhood rules, cross-layer rules, scale rules, thermal channel rules, and risk propagation rules; the preset impact propagation rules are used to perform thermal impact propagation analysis on the dirty window candidate set according to the window index table in the old version of the multi-scale thermal parameter database, and output the affected window set, including: Based on the adjacent window list in the window index table, adjacent windows that conform to the spatial neighborhood rules are filtered and included in the propagation change window set; Based on the cross-layer association index of the window index table, select the upper and lower layer hot path connection windows that meet the cross-layer rules and include them in the propagation change window set; Based on the parent-child indexes of the window index table, parent windows and higher-level parent windows that meet the scaling rules are selected and included in the aggregated update window set; Based on the local dominant hot channel association information in the window index table, connected windows that conform to the hot channel rules are selected and included in the propagation change window set; Identify windows belonging to the preset high-heat-sensitive region in the dirty window candidate set, supplement and filter connected windows according to the risk expansion rule, and include them in the propagation change window set; The dirty window candidate set, the propagated change window set, and the aggregated update window set are merged to obtain the merged window set; Remove duplicate windows from the merged window set and output the affected window set.

5. The multi-scale incremental update method for thermal parameters oriented towards layout ECO according to claim 1, characterized in that, The process employs preset classification rules to classify and classify the affected window set into dirty windows based on the new map file, the map change fragment data, and the window index table, thermal parameter table, and version metadata table in the old multi-scale thermal parameter database, outputting ordinary dirty windows and high-risk dirty windows, including: Based on the map change fragment data, extract the change intensity data corresponding to each affected window in the affected window set; Extract the structural complexity data corresponding to each affected window based on the window index table and the local geometric data of the new map in the new map file; Extract the thermal parameter confidence data corresponding to each affected window based on the thermal parameter table and the version metadata table; Based on the preset classification and determination rules, the multiple affected windows are divided into ordinary dirty windows and high-risk dirty windows by using the change intensity, structural complexity, and thermal parameter confidence data corresponding to each affected window.

6. The multi-scale incremental update method for thermal parameters oriented to layout ECO according to claim 1, characterized in that, The method employs a preset fast update model to update the thermal parameters of the ordinary dirty window and the high-risk dirty window based on the new version of the drawing file, the old version of the multi-scale thermal parameter database, and the PDK process data, respectively, and outputs a set of updated thermal parameters for the dirty window, including: Extract the local geometric difference feature data corresponding to the ordinary dirty window from the new version of the drawing file, and read the reference thermal parameter data corresponding to the ordinary dirty window from the old version of the multi-scale thermal parameter database; The feature parameters are updated based on the local geometric difference feature data to obtain the updated feature parameters; The updated feature parameters and the baseline thermal parameter data are input into the preset fast update model to calculate the new thermal parameter data corresponding to the ordinary dirty window. Extract the local geometric data corresponding to the high-risk dirty window from the new drawing file, and construct a local three-dimensional high-precision model corresponding to the high-risk dirty window by combining it with the PDK process data; Preset boundary conditions are applied to the local three-dimensional high-precision model, and the local heat flow response data is obtained by solving the problem. Based on the local heat flow response data, the updated equivalent thermal conductivity tensor and associated thermal parameter data of the high-risk dirty window are calculated by inversion. Integrate the new thermal parameter data of the ordinary dirty window, the updated equivalent thermal conductivity tensor and the corresponding thermal parameter data of the high-risk dirty window, and output the updated thermal parameter set of the dirty window.

7. The multi-scale incremental update method for thermal parameters oriented towards layout ECO according to claim 1, characterized in that, The process employs preset aggregation rules and preset database replacement rules to perform cross-scale thermal parameter updates and compatibility encapsulation based on the dirty window update thermal parameter set, the old multi-scale thermal parameter database, and the window index table in the old multi-scale thermal parameter database, outputting new thermal parameter data in a compatible format, including: Based on the parent-child index of the window index table and the set of dirty window update hot parameters, the parent window to be passively updated is determined. Read the historical thermal parameter data of the unupdated child windows under the parent window to be passively updated from the old multi-scale thermal parameter database; The historical thermal parameter data of the unupdated child window is combined with the new thermal parameter data of the updated child window in the dirty window update thermal parameter set, and the complete update thermal parameter data of the parent window to be passively updated is calculated using the preset aggregation rule. According to the preset database replacement rules, the dirty window update thermal parameter set and the complete updated thermal parameter data are used to perform local replacement and parameter update on the corresponding window records in the old version of the multi-scale thermal parameter database to obtain the new version of the multi-scale thermal parameter database. The complete data of the new multi-scale thermal parameter database is read according to the preset downstream compatible interface format, and the complete data is adapted and encapsulated to output the new thermal parameter data in a compatible format.

8. A multi-scale incremental update system for thermal parameters oriented to layout ECO, characterized in that, include: The acquisition module is used to acquire old layout files, new layout files, old multi-scale thermal parameter databases and PDK process data, and perform layout difference detection based on the old layout files and the new layout files, and output layout change segment data; The mapping module is used to perform multi-scale window mapping based on the map change fragment data and the window index table in the old version of the multi-scale thermal parameter database, and output a dirty window candidate set. The analysis module is used to perform thermal influence propagation analysis on the dirty window candidate set according to the window index table in the old multi-scale thermal parameter database using preset influence propagation rules, and output the set of affected windows; The grading module is used to perform dirty window grading and judgment processing on the affected window set according to the new map file, the map change fragment data and the window index table, thermal parameter table and version metadata table in the old multi-scale thermal parameter database, and output ordinary dirty windows and high-risk dirty windows. The update module is used to update the thermal parameters of the ordinary dirty window and the high-risk dirty window respectively according to the new version drawing file, the old version multi-scale thermal parameter database and the PDK process data using a preset fast update model, and output the dirty window updated thermal parameter set; The encapsulation module is used to perform cross-scale thermal parameter updates and compatibility encapsulation based on the dirty window update thermal parameter set, the old version of the multi-scale thermal parameter database, and the window index table in the old version of the multi-scale thermal parameter database, using preset aggregation rules and preset database replacement rules, and output new version thermal parameter data in a compatible format.

9. An electronic device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the multi-scale thermal parameter incremental update method for layout-oriented ECO as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the multi-scale thermal parameter incremental update method for layout-oriented ECO as described in any one of claims 1-7.

Images