Search

CN-121981050-A - EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching

CN121981050ACN 121981050 ACN121981050 ACN 121981050ACN-121981050-A

Abstract

The invention discloses a multiplexing method of EDA design object model and sample data based on dynamic and static separation attribute matching, and belongs to the technical field of circuit automation. The method realizes the efficient multiplexing of the design object model and the sample by dividing the design attribute into dynamic and static attribute and constructing an attribute package separated from the design identifier to perform encapsulation and de-encapsulation. The multiplexing process is mainly that topology and static attribute are precisely matched through map isomorphism and node mapping, then items with dynamic attribute meeting the range are screened, and a matched follow-up sample or model preferred method is provided. Meanwhile, the matched storage method realizes data standardization from the source. The method provides powerful support for the AI proxy model of any circuit or element automatically deployed by the user side, lays a foundation for the realization of native-AI EDA, has wide applicability, can be used for multiplexing samples and proxy models at the same time, and greatly improves the feasibility and application value of the scheme.

Inventors

  • SHE YUCHEN

Assignees

  • 南京波思芯软智能科技(个人独资)

Dates

Publication Date
20260505
Application Date
20260126

Claims (8)

  1. 1. The EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching is characterized by comprising the following steps of: S1, receiving user input, wherein the user input at least comprises element connection relation, design attribute and simulation condition of a target design object, the design attribute is divided into dynamic attribute which can be adjusted and static attribute which can be kept fixed in a design space exploration process by specific scene requirements, and the design space exploration process is a process of parameter optimization, design instance sample acquisition and proxy model construction based on the design object by a user; S2, constructing a reference attribute package of the current design object, wherein when the design instance is an element, the attribute package carries dynamic and static attribute information of the design object, and when the design instance is a circuit, the attribute package carries attribute package information of all element instances and connection relation information among the elements; s3, sequentially performing at least two levels of filtering on the candidate item set: The first stage, through the parallel data comparison of the attribute package of the reference attribute package and the attribute package of the candidate item, the item matched with the static attribute of the target design object in the candidate item is screened out, then all the static attributes in the attribute package of the candidate item are removed, and only the dynamic attribute is reserved; The second stage, from the filtering result of the first stage, judges and screens out the item of the candidate item attribute package data in the dynamic attribute range of the current design object in parallel; s4, giving the static attribute of the reference attribute kit to the items filtered out at the second stage, so that the electrical attribute of each item is completely expressed by the dynamic attribute and the static attribute of the current configuration.
  2. 2. The method for multiplexing EDA design object model and sample data based on dynamic and static separation attribute matching according to claim 1, wherein for the static attribute matching process, when the design object is a layout element, static attribute consistency check is directly performed, and when the design object is a circuit, the following steps are performed: s21, forming graph data by taking elements as points and connecting lines as edges according to the element connection relation, so as to generate first graph data representing the target circuit topological structure; s22, firstly, screening out the items of isomorphic second graph data and the first graph data from the candidate object library, and taking the items as preliminary candidate items, then, in the second graph data, finding out nodes with the same names as the elements of the first graph data, taking the nodes as anchor points, carrying out node mapping alignment of the two graphs, and finally, screening out items with the static attributes of the elements on the mapped nodes matched.
  3. 3. The method for multiplexing EDA design object model and sample data based on dynamic and static separation attribute matching according to claim 2, wherein the matching process is specifically implemented by the following steps: when the design object is a layout element, a reference attribute package of the current design instance is established, a candidate item attribute package is firstly unpacked, and then parallel static attribute consistency verification is carried out: S31, establishing a circuit standardized attribute package and an adjacency list, namely establishing an undirected graph based on a circuit connection relation to generate a universal port identifier irrelevant to design naming, extracting attributes of each element, adding a unique prefix of the identifier for each attribute of the element as a characteristic prefix of an attribute package of the element, and combining all the element attributes of the circuit to form a reference attribute package of the circuit; s32, checking a circuit structure, namely checking the isomorphism of the graph of the standardized adjacency list and the candidate entry adjacency list, and outputting isomorphic entries; S33, establishing a node mapping relation, namely finding out an element with the same name as the current circuit in isomorphic candidate item ports, aligning two undirected graphs by taking the node where the element is located as an anchor point, and establishing the mapping relation of the two graphs to each node; S34, checking the attribute names of the single elements, namely executing one element of the candidate entry circuit based on the mapping relation: a) Finding out a reference attribute packet of a corresponding element of the current circuit according to the mapping relation; b) Renaming each candidate item attribute prefix to be the characteristic prefix obtained in the S31 of the reference attribute kit; c) Performing static attribute name consistency verification; S35, checking attribute names, namely executing the step S34 on other elements of the candidate item circuit one by one, and immediately stopping and eliminating the candidate item only when all element checking passing parties judge that static attribute names are successfully matched and output; S36, unpacking the output result of S35, carrying out parallel static attribute consistency parallel data inspection on unpacked data, and outputting the items with consistent inspection as static attribute matching output.
  4. 4. The EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching of claim 1, further comprising a sample optimizing step after sample loading to further utilize loading content, wherein the specific steps are that samples in the sample library are filtered in a specified mode and specified intensity, and the specified mode comprises the following two modes, wherein each mode needs to be configured with a corresponding intensity coefficient: s41, performance filtration: Filtering based on Euclidean distance between sample performance and one or more design targets, wherein the method comprises the following specific steps: s411, weakening targets, namely converting each design target of the element example into a relatively loose performance range according to the intensity coefficient, wherein the larger the input intensity coefficient is, the smaller the weakened performance range is; S412, calculating the duty ratio, namely calculating the duty ratio of each sample performance data point falling into the corresponding weakening performance range in the loaded samples; S413, judging a threshold value, namely filtering out the sample when the duty ratio of the sample on all design targets exceeds a preset threshold value, wherein the larger the input intensity coefficient is, the higher the preset threshold value is; S42, filtering attributes: Filtering based on Euclidean distance between a sample dynamic attribute value and a current design object dynamic attribute value, wherein the method comprises the following specific steps: s421, constructing a neighborhood, namely constructing a hypercube neighborhood in a multidimensional space by taking a current dynamic attribute value of an element instance as a center and taking an intensity coefficient designated by a user as a radius, wherein the larger the input intensity coefficient is, the smaller the constructed neighborhood radius is; s422, interval screening, namely screening out samples with dynamic attribute values completely falling in the hypercube neighborhood from the loaded samples, and filtering out the samples.
  5. 5. The method for multiplexing EDA design object model and sample data based on dynamic and static separation attribute matching according to claim 1, further comprising a model preferential step after loading the model to further utilize loading content, wherein the method comprises the following specific steps: S51, loading model files of all items in the multiplexing loading result to obtain a group of candidate agent models; s52, sampling is carried out in the target range of the dynamic attribute, and a plurality of test samples are generated; s53, predicting the test sample by utilizing each candidate agent model, and grading according to a prediction result; s54, selecting the candidate agent model with the highest score as an optimal model.
  6. 6. An EDA design object holographic storage method based on design identification stripping is characterized in that the following steps are executed: S61, receiving user input, wherein the user input at least comprises element connection relation, design attribute and simulation condition of a target design object; S62, constructing an attribute package for the current design object, and packaging the current design object by taking the element as a unit, wherein the packaging method specifically comprises the steps of using an element class name as a prefix packaging attribute package if the design object is an element, and executing the establishment of a circuit standardized attribute package and an adjacent table if the design object is a circuit as follows: a) The method comprises the steps of establishing a circuit standardized attribute package and an adjacency list, constructing an undirected graph based on a circuit connection relation to generate a universal port identifier irrelevant to design naming, extracting attributes of each element, adding a unique prefix of the identifier for each attribute of the element as a characteristic prefix of the attribute package, merging all the element attributes of the circuit to form a reference attribute package of the circuit, and simultaneously, generating an independent file of the standardized adjacency list corresponding to the undirected graph of the current design circuit; b) The method comprises the steps of taking element characteristic prefixes as element attribute package prefixes one by one, packaging element attribute packages, combining all element attribute packages to be used as a whole circuit attribute package, enabling the number of formed circuit attribute packages to be equal to the number of circuit elements, and additionally adding a circuit type attribute for sea selection identification of stored data; s63, adding simulation conditions and simulation result file paths corresponding to each sample and corresponding attribute packages into the sample data structure; S64, storing the data structure as candidate item files, if the design object is a circuit, synchronously generating and storing independent files of the standardized adjacent undirected graph connection table obtained in the step of establishing the circuit standardized attribute package and the adjacent table, and binding the independent files with the candidate item files, wherein the binding relation is that when the samples are stored, all samples in the same storage batch share one connection table file, and when the models are stored, one candidate item file binds one adjacent table file.
  7. 7. The design object as claimed in any one of claims 1 to 6, wherein the layout element for the analog, radio frequency and millimeter wave circuit and the patch antenna as the independent radiating element are included at the element level, the analog, radio frequency and millimeter wave schematic circuit, the layout circuit, the hybrid design circuit comprising the schematic element and the layout element, and the complete antenna circuit comprising an array of a plurality of patch antenna elements in a specific topology or comprising a matching network are covered at the circuit level.
  8. 8. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the functions of the system or method of any of claims 1 to 6.

Description

EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching Technical Field The invention belongs to the technical field of circuit automation, and particularly relates to an EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching. Background The optimization technology based on the AI proxy model is a popular engineering problem solving method in recent years, mainly taking parameter data to be solved or optimized as sample characteristics, taking corresponding simulation or experiment result data as sample labels, establishing the AI proxy model, and then optimizing parameters (namely input variables of the proxy model) to be optimized based on the proxy model to obtain the optimal parameters as an optimal solution of the engineering problem. The model is different from a large predictive model in the conventional sense, but is a model built aiming at repeated simulation data, and is different from a large predictive model (LLM) in the form of a required dialogue text pair, wherein the sample characteristics of the model are a limited set formed by design properties of an optimized object and simulation conditions, the sample is often a data pair of input parameters (object properties or conditions) -simulation results, and the model is built with high causality consistency, namely, the result of simulation after the input of the parameter a is B under the condition of (a limited set), and the result of simulation after the input of the parameter c is d under the condition of (a limited set), the sample pairs (a, B) and (c, d) can not be used for training the same proxy model because the difference exists in the input-output relationship (namely, a physical response mechanism) mapped by the model due to the different simulation condition sets (A and B) to which the sample belongs. For analog/Radio Frequency (RF)/microwave circuit/patch antenna, the sample tag required by the proxy model is generally data obtained by electromagnetic simulation and circuit simulation, and the process often consumes a lot of time, so that the sample is not easy to obtain. If samples are generated in batches when the AI proxy model is built, a great deal of time is required to generate sample data labels, severely slowing down the overall optimization progress. Thus, one of the bottlenecks in AI proxy model optimization techniques is how to achieve efficient multiplexing of samples with models. Still further, even if mass, heterogeneous circuit/component samples and models can be stored in an EDA design system, it is still a challenge to manage them effectively and to enable them to be intelligently matched to the specific design needs of the user diversity. This makes it difficult for the optimization technique of the AI proxy model to fully meet the specific design requirements of the user, greatly reducing the flexibility and practicality of the technique in commercial applications. The core obstacle to design multiplexing stems mainly from two aspects: first, circuit design naming lacks standardization specifications and is disjoint from functional semantics. Traditional EDA platform relies on a designer custom naming system, and random and ambiguous naming often exists, so that the intrinsic topological characteristics and functional attributes of a design object are difficult to systematically reflect, and the system is difficult to effectively extract and store core characteristics with multiplexing value; and secondly, the circuit has high data dimension and high information density. The feature dimension of a complete design object is usually hundreds to thousands, and even if core features are successfully extracted, the high-dimensional data structure of the complete design object is difficult to support high-efficiency access and quick matching, so that the multiplexing efficiency of design knowledge is obviously restricted. Therefore, in view of the above technical problems, it is needed to provide an EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching. The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to provide an EDA design object model and sample data multiplexing method based on dynamic and static separation attribute matching, and the method is characterized in that design information is packaged and reconstructed by constructing a standardized attribute package. For elements or circuits, the method separates design attributes into dynamic and static categories, and packages the design attributes into structured attribute packages