DRC rule making method and device, equipment and storage medium

Abstract

The invention provides a DRC rule making method and device, equipment and a storage medium. The method comprises the steps of obtaining a DR description file and an old-version DRC configuration file; preprocessing the obtained DR description file and the old-version DRC configuration file to obtain basic data, and establishing a DRC field knowledge base based on the basic data; establishing a DRC formulating model by using the DRC domain knowledge base data, performing supervision and fine tuning on parameters of the DRC formulating model, enhancing the retrieval capability of the DRC formulating model on the DRC domain knowledge base, and optimizing an output result of the DRC formulating model through Few-shot Learning; and updating the old-version DRC configuration file based on different DR data so as to output a corresponding new-version DRC configuration file. According to the implementation of the method, the development quality of the DRC configuration file can be improved, the accuracy of DRC generation in a chip design verification link is ensured, and the DRC development time is shortened.

Description

Technical Field

[0001] The present invention relates to the collaborative field of chip R&D and design, and more specifically, to a DRC rule formulation method, device, equipment and storage medium. Background Art

[0002] In the field of integrated circuit design, as technology nodes continue to shrink, DTCO (Design Technology CoOptimization) has become a key strategy to ensure efficient coordination between design and process. The DTCO method allows circuit designers and process engineers to work closely together in the early stages of design to optimize the PPAC (Power, Performance, Area, Cost) of integrated circuits.

[0003] In the traditional design process, DRC (Design Rule Check) is an important step to ensure that the design meets the requirements of specific manufacturing processes. However, with the development of technology, the design of integrated circuits has become more and more complex, especially at advanced technology nodes, where design rules have become more stringent and numerous. In this case, later design modifications become more difficult and expensive. Therefore, how to quickly and accurately apply design rules in the early stages of the design process to prevent rule violations has become an urgent problem that the industry needs to solve.

[0004] Existing technologies have encountered bottlenecks in dealing with this challenge. Although some tools and methods can provide a certain degree of DRC support in the early stages of design, they often fail to meet the needs of fast iteration and efficient optimization. These tools are insufficient in handling complex designs, especially at advanced nodes, where their efficiency and accuracy issues are more obvious. In addition, with the continuous advancement of technology nodes, the complexity and number of design rules are also increasing, making it difficult for existing DRC tools to keep up with the rapidly evolving needs. Summary of the invention

[0005] To solve the above technical problems, the present invention provides a new methodology, which overcomes the shortcomings of the existing technology, improves the efficiency of the design process, shortens the time to market, and reduces costs through a more efficient design rule formulation and checking mechanism.

[0006] The core of the present invention is to integrate DRC into the early stage of design, so that potential design and process conflicts can be discovered and resolved earlier in the design process; by using advanced algorithms and automated scripting tools, the complexity of rules is simplified, and the number of rules that designers need to deal with is reduced, while maintaining the quality of the design and the compatibility of the process. When the design does not meet the rules, the design can be adjusted automatically, reducing manual intervention, improving efficiency, and providing a real-time feedback system, so that PDK designers can see the modified results immediately, speeding up the speed of design iteration, and promoting collaboration between circuit designers and process engineers, ensuring that both parties can understand each other's needs and limitations in a timely manner, and jointly promote the optimization of the design.

[0007] This method optimizes the DTCO process by improving the DRC-specific large language model, and realizes the rapid generation and iteration of design rules to meet the design requirements of advanced technology nodes. This method is particularly suitable for complex IC design. It can accelerate the DRC formulation and optimization process while ensuring design quality, improve design efficiency, and reduce violation risks and design costs.

[0008] The present invention is achieved through the following technical solutions.

[0009] The present invention provides a DRC rule formulation method, comprising:

[0010] Obtain the DR description file and the old version of the DRC configuration file. The DR description file includes DR rules.

[0011] Data preprocessing, including:

[0012] Using an automated script to pre-process the acquired DR description file and the old version DRC configuration file to obtain basic data,

[0013] Establishing a DRC domain knowledge base based on the basic data;

[0014] Using the DRC domain knowledge base data to establish a DRC formulation model, and fine-tuning the parameters of the DRC formulation model to obtain a DRC formulation model after fine-tuning.

[0015] Combining VSM technology and RAG technology, the retrieval capability of the supervised fine-tuned DRC formulation model for the DRC domain knowledge base is enhanced.

[0016] Optimizing the output of the supervised fine-tuned DRC formulation model through Few-shot Learning;

[0017] The old version of the DRC configuration file is updated based on different DR data to output a corresponding new version of the DRC configuration file.

[0018] Furthermore, the data preprocessing includes:

[0019] Data collection, which collects data from DR description files and legacy DRC configuration files through automated scripts;

[0020] Data analysis: analyze the collected data to determine the logical relationships and dependencies of the data and output the analysis results;

[0021] Rule connection: connect the analysis results with a single specific DR rule;

[0022] Data matching, matching DR rules in DR description files with old DRC configuration files through pattern recognition technology.

[0023] Furthermore, the DR rules in the DR description file and the old version of the DRC configuration file are matched by pattern recognition technology, which is implemented by the following formula:

[0024]

[0025] Where R = {R1, R2, ..., R i ,..,R n} is the DR rule set, n is the total number of DR rules, R i represents the i-th DR rule; C={C1,C2,...,C j ,...,C S} is the rule set in the old version of the DRC configuration file, S is the total number of rules in the rule set in the old version of the DRC configuration file, C j is the jth rule in the old DRC configuration file; U(f) is the utility function used to measure the mapping quality of mapping f, mapping f: R→C, for each R i ∈R, all have C j ∈C, satisfying M(R i ,C j )=1; matching function M:R×C→{0,1}, where if R i With C j Match, then M(R i ,C j )=1, otherwise M(R i ,C j )=0.

[0026] Furthermore, the supervised fine-tuning is performed in the following manner:

[0027]

[0028] Among them, m is the number of labeled samples, l is the loss function, which is used to measure the model prediction result f(X p;θ) and the actual label, X p Represents the feature vector of the pth labeled sample, y p represents the pth actual label, λ is the regularization parameter used to prevent overfitting; and w is the dimension of the feature. p The corresponding DRC rule label y p Correlation, where y p Indicates whether the sample meets the design rules of PDK. Using supervised learning algorithm, a classification model f(X p ; θ), the model aims to minimize the prediction error L(θ), where θ represents the parameters of the model.

[0029] Furthermore, the retrieval capability of the DRC formulation model for the DRC domain knowledge base is enhanced by:

[0030]

[0031] r = argmax r P(r|q,D retrieved )

[0032] in, For documents in the DRC domain knowledge base, is the input query document; the retrieved document set is , In a given query and the retrieved document collection Response under the conditions The probability of Training a parameterized model With parameters , the model receives a limited set of labeled sample combinations Then learn, E represents the combination set of labeled samples The total number of sample combinations, Represents a collection of labeled sample combinations The second sample combination of By adjusting the parameters To adapt to the new PDK design rules, this process is achieved through the loss function The loss function measures the model output With the true label The goal of model training is to minimize this loss function:

[0035] Few-shot learning works together with supervised fine-tuning to further improve the learning efficiency and accuracy of the model for new and specific design rules. Under this condition, the model parameters are divided into two parts: basic parameters θ B and the tuning parameter θ A , where θ B In a large amount of DRC data, θ A It is adjusted by Few-shot Learning, and the parameter update of the model is expressed as:

[0036]

[0037] The retrieval process is regarded as a query function Q, which retrieves relevant information V from the DRC domain knowledge base. The output of the model also includes the retrieved relevant information:

[0038] M(x i ;θ)=Generate(x,V,θ)

[0039] V=Q(x;K)

[0040] Among them, K represents the DRC domain knowledge base, the Generate function represents the process of model generation, and the Q function represents the process of retrieving information from the DRC domain knowledge base K based on the input x.

[0041] Furthermore, the specific process of outputting the corresponding new version of DRC configuration file based on different DR data includes the following steps: Assume that the model has a new DRC profile rule set from the new PDK , the i-th DRC profile rule Contains a corresponding design parameter , design parameters , for each i-th , extract its corresponding feature vector , through the feature extraction function To achieve, that is:

[0043] Apply clustering algorithm K to the feature vector set F = {f1,f2,...f n} Analyze to identify common patterns: W = K(F), where W represents the set of identified design rule patterns;

[0044] For the new PDK requirements, expressed as a set of design parameters P new , define an adaptation function ɑ to adjust the existing model to meet the new requirements: R new =ɑ(W,P new )

[0045] Finally, define an optimization O process to refine the rule R new , ensuring its effectiveness and optimality under the new PDK: R optimized =O(R new ).

[0046] The present invention also provides a DRC rule formulation device, comprising:

[0047] A data preprocessing module, used to obtain basic data obtained after preprocessing the DR description file and the DRC configuration file;

[0048] Establishing a DRC domain knowledge base module, for creating a DRC domain knowledge base from the basic data obtained after preprocessing the DR description file and the DRC configuration file;

[0049] Build a DRC formulation model module to train and optimize the DRC formulation model;

[0050] A module for outputting corresponding new versions of DRC configuration files based on different DR data sources is used to output corresponding new versions of DRC configuration files according to the DR data of a specific process node input.

[0051] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned DRC rule formulation methods when executing the program.

[0052] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements any of the above-mentioned DRC rule formulation methods when executed by a processor.

[0053] The beneficial effects of the present invention are:

[0054] (1) This method optimizes the DTCO process by improving the DRC-specific large language model, achieving rapid generation and iteration of design rules to meet the design requirements of advanced technology nodes. This method is particularly suitable for complex IC design (Integrated Circuit Design). It can accelerate the DRC formulation and optimization process while ensuring design quality, improve design efficiency, and reduce violation risks and design costs.

[0055] (2) The efficiency of DRC formulation and optimization is improved. By introducing automated tools and advanced algorithms, the efficiency of the DRC formulation and optimization process is significantly improved. Traditional DRC methods rely on manual adjustments and iterative processes, which are time-consuming and prone to errors. The method of the present invention uses automated algorithms to quickly and accurately identify potential design problems and provide optimization suggestions, reducing the need for human intervention. This increase in efficiency allows PDK designers to respond to changes in design rules more quickly, accelerate the design verification process, and shorten the time it takes for products to go from concept to market. At the same time, the automated DRC optimization process also reduces the workload of PDK designers, allowing them to focus more on innovation and other key aspects of design, thereby improving overall design productivity.

[0056] (3) Enhanced the adaptability and mobility of automatic generation of DRC configuration codes for evolving technology nodes. The modular design rule formulation method provides greater scalability and flexibility for design projects of different types and sizes. In the traditional DRC process, the formulation of DRC configuration codes is often fixed to a specific chip process, which limits the applicability of the rules between different projects. The method of the present invention allows design rules to exist in a modular form, which can be quickly adjusted and reorganized according to the specific needs of the project. This flexibility is not only suitable for diverse design projects, but also can adapt to rapidly changing manufacturing processes and technical standards.

[0057] (4) The potential risk of DRC violations is reduced through predictive analysis. By introducing an advanced large language prediction model, potential DRC conflict issues can be predicted in the early stages of the design process. This predictive analysis allows PDK designers to make necessary modifications before submitting the final design, thereby significantly reducing the possibility of violations. This not only reduces the number of reworks, but also shortens the overall design cycle and accelerates the time to market for products. In addition, this method also helps to identify and solve problems that may arise in the subsequent manufacturing process in advance, thereby reducing the risk of the entire project and ensuring the reliability and performance of the final product. Through such predictive analysis, the PDK design team can plan the design process more accurately, optimize resource allocation, and improve the efficiency of the entire design and manufacturing process. By improving the scalability and flexibility of automatic DRC code generation, the present invention supports a wider range of design innovations and promotes further improvement in PDK design efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is the overall flow chart of the present invention.

[0059] Figure 2 It is a block diagram of the design rule automatic generation module in the present invention.

[0060] Figure 3 A flow chart of a DRC formulation model in an embodiment of the present invention.

[0061] Figure 4 Automatically generate DRC configuration code content for inputting 40nm process node DR file content.

[0062] Figure 5 It is a schematic diagram of the human-computer interaction interface of the present invention.

[0063] Figure 6 This is a diagram of the rule verification and feedback mechanism in the present invention.

[0064] Figure 7 It is a structural schematic diagram of the DRC rule formulation device provided by the present invention;

[0065] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0066] The technical solution of the present invention is further described below, but the scope of protection claimed is not limited to the description.

[0067] A DRC rule formulation method comprises the following steps:

[0068] Acquire a DR description file and an old version of DRC configuration file, wherein the DR description file includes DR rules; perform data preprocessing, including using an automated script to preprocess the acquired DR description file and the old version of DRC configuration file to obtain basic data, and establish a DRC domain knowledge base based on the basic data; establish a DRC formulation model using the DRC domain knowledge base data, and perform supervised fine-tuning on the parameters of the DRC formulation model, and enhance the retrieval capability of the supervised fine-tuning DRC formulation model for the DRC domain knowledge base by combining VSM technology and RAG technology, and optimize the output result of the supervised fine-tuning DRC formulation model through Few-shot Learning; update the old version of the DRC configuration file based on different DR data to output the corresponding new version of the DRC configuration file.

[0069] By performing the same operations as above through the new version of the DRC configuration file, a new DRC domain knowledge base can be obtained, and then a new DRC formulation model can be obtained.

[0070] The data preprocessing includes: data collection, collecting data from DR description files and old version DRC configuration files through automated scripts; data analysis, analyzing the collected data to determine the logical relationships and dependencies of the data and outputting the analysis results; rule connection, connecting the analysis results with a single specific DR rule; data matching, matching the DR rules in the DR description file with the old version DRC configuration files through pattern recognition technology.

[0071] The present invention mainly uses automated scripts (such as Python or Shell scripts) to automatically parse the rules in the DR description file and intelligently match these rules with the existing DRC configuration files. This process significantly reduces manual intervention and improves the efficiency and accuracy of rule formulation. The established DRC domain knowledge base integrates the scattered design rules and configuration file information to provide high-quality data support for the subsequent training and retrieval of large language models.

[0072] R={R1,R2,...,R i ,..,R n} is the DR rule set, n is the total number of DR rules, R i represents the i-th DR rule; C={C1,C2,...,C j ,...,C S} is the rule set in the old version of the DRC configuration file, S is the total number of rules in the rule set in the old version of the DRC configuration file, C j is the jth rule in the old DRC configuration file; U(f) is a utility function used to measure the mapping quality of mapping f, which is the number of successfully matched rules or the accuracy of matching.

[0073] The goal of the automated script is to find a mapping f:R→C such that for every R i ∈R, all have C j ∈C, satisfying M(R i ,C j )=1; matching function M:R×C→{0,1}, where if R i With C j Match, then M(R i ,C j )=1, otherwise M(R i ,C j )=0.

[0074] Therefore, the task of the automation script can be formulated as the following optimization problem:

[0075]

[0076] Using this formula algorithm, data preprocessing attempts to maximize the matching utility while ensuring that each rule R i can find at least one matching configuration C j .

[0077] The parameters of the large language model are adjusted using the SFT (Supervised Fine-Tuning) technology. The retrieval mapping capability of the DRC domain knowledge base data is enhanced using the VSM (Vector Space Model) and RAG (Retrieval-Augmented Generation) technologies. The content of the DRC configuration file output by the model is optimized through Few-shot Learning.

[0078] Through SFT technology, the parameters of the large language model are adjusted according to the characteristics of the DRC configuration file to make it more suitable for specific application scenarios. Combining VSM and RAG technology, the model's retrieval and mapping capabilities for the DRC domain knowledge base are enhanced, improving the relevance and accuracy of the generated results. At the same time, through Few-shot Learning, the model quickly adjusts its parameters after receiving a limited number of annotated DRC rule samples, so that the model can still accurately adapt to and reflect the design rule requirements of a specific PDK when there is insufficient data. The combined application of these technologies enables the model to automatically generate highly accurate DRC configuration files based on the input DR rules, breaking through the bottleneck of traditional methods.

[0079] Extract the feature vector X of each labeled sample p , where i represents the sample index, X p ∈R w And w is the dimension of the feature. Each feature vector X p The corresponding DRC rule label y p Correlation, where y p Indicates whether the sample complies with the design rules of PDK. Using supervised learning algorithm, a classification model f(X p ; θ), the model aims to minimize the prediction error L(θ), where θ represents the parameters of the model. The optimization objective function L(θ) is defined by the following formula:

[0080]

[0081] Among them, m is the number of labeled samples, l is the loss function, which is used to measure the model prediction f(X p ;θ) and the actual label y p , λ is the regularization parameter used to prevent overfitting.

[0082] Through the iterative optimization algorithm, the model parameters θ are updated until they converge to the minimum error or reach the predetermined number of iterations. Through the above steps, the SFT technology can fine-tune the parameters according to the design rule requirements of a specific PDK, thereby improving the accuracy and efficiency of the automatic generation method in DRC rule detection.

[0083] VSM and RAG technologies combine the generation and retrieval capabilities of large language models to improve the efficiency and accuracy of queries on DRC domain knowledge base data. In this process, VSM first converts queries and documents into vectors. These vectors are represented in a high-dimensional space, where each dimension corresponds to a specific term. For document d and query q, their vector representations are and The similarity between a document and a query is usually calculated by taking the cosine similarity of their vectors:

[0084]

[0085] in, The dot product represents the inner product of two vectors. and is the Euclidean norm of the vector.

[0086] RAG technology combines vector space model-based retrieval and transformer-based language model generation. For a given input query q, RAG first uses VSM to retrieve relevant documents in the DRC domain knowledge base. This is done by calculating the similarity between the query vector and all document vectors in the knowledge base. The retrieved document set is D retrieved .

[0087] RAG then retrieves the document D retrieved As context information, it is input into the DRC specified model after SFT. The model combines the retrieved context and the original query to generate a response r. The response generated by RAG can be expressed as:

[0088] r = argmax r P(r|q,D retrieved )

[0089] Among them, P(r|q,D retrieved ) is the set of documents D retrieved from a given query q retrieved The probability of responding r under the condition of .

[0090] Finally, RAG adopts a fusion strategy to combine the retrieval and generation results to produce the final output.

[0091] Through this combined retrieval and generation approach, RAG is able to leverage the rich information in the DRC domain knowledge base to improve the relevance and accuracy of answers while maintaining the fluency and coherence of responses generated by the language model.

[0092] Using Few-shot Learning technology, the model can quickly adjust its parameters after receiving a limited number of annotated DRC rule samples, so that the model can accurately adapt to and reflect the design rule requirements of a specific PDK without the need for a large amount of training data.

[0093] Assume that a parameterized model M and parameter θ are trained. When the model receives a limited set of labeled sample combinations D = {(x1, y1), (x2, y2), ... (x E ,y E )}, E represents the total number of sample combinations in the labeled sample combination set D, (x2, y2) represents the second sample combination in the labeled sample combination set D. In the Few-shot Learning scenario, the model M can adapt to the new PDK design rules by adjusting the parameter θ. This process can be guided by the loss function L, which measures the model output M(x i ;θ) and the true label y i The goal of model training is to minimize this loss function:

[0094]

[0095] Few-shot learning works together with supervised fine-tuning to further improve the learning efficiency and accuracy of the model for new and specific design rules. Under this condition, the model parameters are divided into two parts: basic parameters θ B and the tuning parameter θ A , where θ B In a large amount of DRC data, θ A It is adjusted by Few-shot Learning, so the parameter update of the model can be expressed as:

[0096]

[0097] Combining VSM and RAG technology, Few-shot Learning enables the model to improve the efficiency and accuracy of queries through effective retrieval and generation operations when querying the DRC domain knowledge base, even for extremely rare rule information. The retrieval process is regarded as a query function Q, which retrieves relevant information V from the DRC domain knowledge base. Therefore, the output of the model also includes the retrieved relevant information:

[0098] M(x i;θ)=Generate(x,V,θ)

[0099] V=Q(x;K)

[0100] Among them, K represents the DRC domain knowledge base, the Generate function represents the process of model generation, and the Q function represents the process of retrieving information from the DRC domain knowledge base K based on the input x.

[0101] A fast DRC formulation method with strong portability based on different DR data sources is implemented.

[0102] After model training and optimization are completed, the proposed method has good portability and can adapt to DR data from different sources, realize the rapid formulation of DRC configuration files, and meet the development needs of different PDKs. This means that the method is not limited to a specific DR data source or PDK and can adapt to different design rules and process requirements. By analyzing and comparing DR data from different sources, the method can identify common design rule patterns and quickly adapt to new PDK requirements, thereby realizing the rapid formulation and migration of DRC rules in different design processes. This feature improves the applicability and practical value of the method.

[0103] Assuming the model has newly imported DRC profile rule set from the new PDK , the i-th DRC profile rule Contains a corresponding design parameter , design parameters , for each i-th , extract its corresponding feature vector , through the feature extraction function To achieve, that is:

[0104] Apply clustering algorithm K to the feature vector set F = {f1,f2,...f n} Analyze to identify common patterns: W = K(F), where W represents the set of identified design rule patterns;

[0105] For the new PDK requirements, expressed as a set of design parameters P new , define an adaptation function ɑ to adjust the existing model to meet the new requirements: R new =ɑ(W,P new )

[0106] Finally, define an optimization O process to refine the rule R new , ensuring its effectiveness and optimality under the new PDK: R optimized =O(R new ).

[0107] The basic step is to collect, analyze, connect and match the DR description file with the existing DRC configuration file through automated scripts to establish a knowledge base in the DRC field. This knowledge base provides the necessary data support for subsequent large language model training and reasoning.

[0108] Using the established knowledge base data, supervised fine-tuning (SFT) is performed on the large language model to adapt it to the DRC profile generation task. Combining VSM and RAG techniques, the model's retrieval ability for the knowledge base is enhanced, and Few-shot Learning is used to optimize the model output.

[0109] In the application phase of the method, after model training and optimization are completed, the trained model is used to quickly formulate DRC configuration files in actual applications. High mobility means that the method can adapt to different DR data sources and achieve rapid formulation of DRC across process nodes and different design rules.

[0110] Through the combined application of the above methods, the method for automatically generating a DRC configuration file proposed in the present invention not only improves the efficiency of generating the configuration file, but also ensures the high accuracy of the generated file, thereby solving the bottleneck problem of DRC configuration file development in the chip design verification link.

[0111] Taking the automatic generation of DR and corresponding DRC configuration content of the 40nm process node as an example, a DRC rule formulation method of the present invention is used to implement the process. Figure 1 As shown in the figure, an overall overview of the entire DRC rapid formulation process is presented, including data preprocessing, knowledge base construction, model training, human-computer interaction, and rule verification.

[0112] The process starts from "Start". First, data preprocessing is performed. This step includes operations such as automated data collection, data analysis, rule connection and rule matching. Then determine whether the data meets the knowledge base requirements. If not, the process ends. If it does, the PDK development kit DRC domain knowledge base is created. Subsequently, a DRC (design rule check) formulation model is built based on the large language model. This process includes SFT training, VSM+RAG data retrieval mapping and Few-shot Learning output optimization. After that, different DR data sources are used to quickly formulate DRC with strong mobility, and then check whether the DRC formulation model representation is good. If the model representation is not good, the process ends. If the representation is good, it enters the last step "End". The entire process ends smoothly when the model representation is good.

[0113] like Figure 2As shown in the figure, data preprocessing is performed on the DRC large language knowledge base, including sub-modules such as data collection, analysis, rule connection and matching. Data is collected from the existing design database and the old version of DRC configuration files through automated scripts. The collected data is analyzed to determine the logical relationship and dependency of the design rules. The analysis results are connected with specific DR rules to provide basic data for knowledge base construction. DR rules are matched with existing DRC configuration files through pattern recognition technology.

[0114] like Figure 3 As shown in the figure, the training and optimization process of the DRC formulation model is described in detail. The SFT technology is used to adjust some parameters of the large language model to adapt to the characteristics of the DRC configuration file. The VSM and RAG technologies are used to enhance the data retrieval capability. The Few-shot Learning solution is used to train the model based on a small number of samples to optimize the output configuration code content.

[0115] like Figure 4 As shown, the specific DR content of the 40nm process node is input into the trained DRC formulation model in text form, and the model automatically outputs the corresponding DRC configuration code according to the input DR rules.

[0116] like Figure 5 As shown in the figure, a schematic diagram of the human-computer interaction process is shown. Humans interact with the automated tool through a graphical interface to formulate and optimize design rules. Specific design parameters or rule requirements are input through the graphical interface. The system provides real-time feedback, and PDK designers adjust the input parameters and rules based on the feedback.

[0117] like Figure 6 The figure shows the process of design rules being verified and adjusted based on feedback. The system applies the generated DRC configuration file to the actual design for verification. Based on the verification results, the configuration file is adjusted and optimized to ensure its accuracy and effectiveness. A feedback link is set up to collect suggestions from PDK designers and the verification results of the system to continuously optimize the model.

[0118] A DRC rule formulation method of the present invention, a trained model can automatically generate a corresponding DRC configuration file according to the input DR rule content, reducing the error rate and workload of manually writing the configuration file. The present invention improves the development efficiency and quality of the DRC configuration file, reduces manual intervention, ensures the accuracy of DRC generation in the chip design verification link, and shortens the DRC development time. By improving the development efficiency and quality of the DRC configuration file, the present invention enhances the competitiveness of the entire chip design process, which is of great significance for shortening the time to market and reducing costs. The present invention solves the problems of efficiency and accuracy in the formulation of DRC configuration files in chip design through a series of technical innovations, and has high innovation and practical value.

[0119] The DRC rule formulation device provided by the present invention is described below. Figure 7 As shown, the DRC rule formulation device described below and the DRC rule formulation method described above can refer to each other.

[0120] A DRC rule formulation device, comprising:

[0121] A data preprocessing module 710 is used to obtain basic data obtained by preprocessing the DR description file and the DRC configuration file;

[0122] Establishing a DRC domain knowledge base module 720, for creating a DRC domain knowledge base from the basic data obtained after preprocessing the DR description file and the DRC configuration file;

[0123] Constructing a DRC formulation model module 730 for training and optimizing a DRC formulation model;

[0124] The module 740 for outputting a corresponding new version of the DRC configuration file based on different DR data sources is used to output a corresponding new version of the DRC configuration file according to the input DR data of a specific process node.

[0125] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the DRC rule formulation method, which includes:

[0126] Obtain DR description files and old version DRC configuration file data preprocessing, including, using automated scripts to preprocess the obtained DR description files and the DRC configuration files to obtain basic data, and establish a DRC domain knowledge base, including, using the basic data obtained in the data preprocessing step to establish the DRC domain knowledge base; construct a DRC formulation model, including, using the DRC domain knowledge base data to supervise and fine-tune the parameters of the DRC formulation model, combining VSM technology and RAG technology to enhance the DRC formulation model's retrieval ability for the DRC domain knowledge base, and optimizing the DRC formulation model output through Few-shot Learning; output a corresponding new version of the DRC configuration file based on different DR data sources, including, inputting DR data of a specific process node into the trained DRC formulation model, and the DRC formulation model automatically outputs a corresponding new version of the DRC configuration file according to the input DR data.

[0127] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the DRC rule formulation method provided by the above methods, the method comprising:

[0128] Obtain DR description files and old version DRC configuration file data preprocessing, including, using automated scripts to preprocess the obtained DR description files and the DRC configuration files to obtain basic data, and establish a DRC domain knowledge base, including, using the basic data obtained in the data preprocessing step to establish the DRC domain knowledge base; construct a DRC formulation model, including, using the DRC domain knowledge base data to supervise and fine-tune the parameters of the DRC formulation model, combining VSM technology and RAG technology to enhance the DRC formulation model's retrieval ability for the DRC domain knowledge base, and optimizing the DRC formulation model output through Few-shot Learning; output a corresponding new version of the DRC configuration file based on different DR data sources, including, inputting DR data of a specific process node into the trained DRC formulation model, and the DRC formulation model automatically outputs a corresponding new version of the DRC configuration file according to the input DR data.

[0129] The above description is only a preferred embodiment of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A DRC rule formulation method, characterized in that: include: Obtain the DR description file and the old version of the DRC configuration file. The DR description file includes DR rules. Data preprocessing, including: Using an automated script to pre-process the acquired DR description file and the old version DRC configuration file to obtain basic data, Establishing a DRC domain knowledge base based on the basic data; Using the DRC domain knowledge base data to establish a DRC formulation model, and fine-tuning the parameters of the DRC formulation model to obtain a DRC formulation model after fine-tuning. Combining VSM technology and RAG technology, the retrieval capability of the supervised fine-tuned DRC formulation model for the DRC domain knowledge base is enhanced. Optimizing the output of the supervised fine-tuned DRC formulation model through Few-shot Learning; The old version of the DRC configuration file is updated based on different DR data to output a corresponding new version of the DRC configuration file.

2. A DRC rule formulation method as claimed in claim 1, characterized in that: The data preprocessing includes: Data collection, which collects data from DR description files and legacy DRC configuration files through automated scripts; Data analysis: analyze the collected data to determine the logical relationships and dependencies of the data and output the analysis results; Rule connection: connect the analysis results with a single specific DR rule; Data matching, matching DR rules in DR description files with old DRC configuration files through pattern recognition technology.

3. A DRC rule formulation method as claimed in claim 2, characterized in that: The DR rules in the DR description file and the old version DRC configuration file are matched by pattern recognition technology, which is implemented by the following formula: Where R = {R1, R2, ..., R i ,..,R n } is the DR rule set, n is the total number of DR rules, R i represents the i-th DR rule; C={C1,C2,...,C j ,...,C S } is the rule set in the old version of the DRC configuration file, S is the total number of rules in the rule set in the old version of the DRC configuration file, C j is the jth rule in the old DRC configuration file; U(f) is the utility function used to measure the mapping quality of mapping f, mapping f: R→C, for each R i ∈R, all have C j ∈C, satisfying M(R i ,C j )=1; matching function M:R×C→{0,1}, where if R i With C j Match, then M(R i ,C j )=1, otherwise M(R i ,C j )=0.

4. A DRC rule formulation method as claimed in claim 1, characterized in that: The supervised fine-tuning is performed in the following way: Among them, m is the number of labeled samples, l is the loss function, which is used to measure the model prediction result f(X p ;θ) and the actual label, X p Represents the feature vector of the pth labeled sample, y p represents the pth actual label, λ is the regularization parameter used to prevent overfitting; X p ∈R w And w is the dimension of the feature. Each feature vector X p The corresponding DRC rule label y p Correlation, where y p Indicates whether the sample meets the design rules of PDK. Using supervised learning algorithm, a classification model f(X p ; θ), the model aims to minimize the prediction error L(θ), where θ represents the parameters of the model.

5. A DRC rule formulation method as claimed in claim 1, characterized in that: The retrieval capability of the DRC formulation model for the DRC domain knowledge base is enhanced by: r=argmax r P(r|q,D retrieved ) Where d is the document in the DRC domain knowledge base, q is the input query document; the retrieved document set is D retrieved ,P(r|q,D retrieved ) is the set of documents D retrieved from a given query q retrieved The probability of responding r under the condition of ; Train a parameterized model M and parameter θ. The model receives a limited set of labeled sample combinations D = {(x1, y1), (x2, y2), ... (x E ,y E )}, E represents the total number of sample combinations in the labeled sample combination set D, (x2, y2) represents the second sample combination in the labeled sample combination set D. In the Few-shot Learning scenario, the model M can adapt to the new PDK design rules by adjusting the parameter θ. This process is guided by the loss function L, which measures the model output M(x i ;θ) and the true label y i The goal of model training is to minimize this loss function: Few-shot learning works together with supervised fine-tuning to further improve the learning efficiency and accuracy of the model for new and specific design rules. Under this condition, the model parameters are divided into two parts: basic parameters θ B and the tuning parameter θ A , where θ B In a large amount of DRC data, θ S It is adjusted by Few-shot Learning, and the parameter update of the model is expressed as: The retrieval process is regarded as a query function Q, which retrieves relevant information V from the DRC domain knowledge base. The output of the model also includes the retrieved relevant information: M(x i ;θ)=Generate(x,V,θ) V=Q(x;K) Among them, K represents the DRC domain knowledge base, the Generate function represents the process of model generation, and the Q function represents the process of retrieving information from the DRC domain knowledge base K based on the input x.

6. A DRC rule formulation method as claimed in claim 1, characterized in that: The specific process of outputting the corresponding new version of DRC configuration file based on different DR data includes the following steps: Assume that the model has a new DRC profile rule set from the new PDK , the i-th DRC profile rule Contains a corresponding design parameter , design parameters , for each i-th , extract its corresponding feature vector , through the feature extraction function To achieve, that is: Apply clustering algorithms For the set of feature vectors Analysis was performed to identify common patterns: ,in represents the set of identified design rule patterns; For new PDK requirements, expressed as a set of design parameters , define an adaptation function ɑ to adjust the existing pattern to meet the new requirements: Finally, define an optimization O process to refine the rules , ensuring its effectiveness and optimality under the new PDK: .

7. A DRC rule formulation device, characterized in that: include A data preprocessing module, used to obtain basic data obtained after preprocessing the DR description file and the DRC configuration file; Establishing a DRC domain knowledge base module, for creating a DRC domain knowledge base from the basic data obtained after preprocessing the DR description file and the DRC configuration file; Build a DRC formulation model module to train and optimize the DRC formulation model; A module for outputting corresponding new versions of DRC configuration files based on different DR data sources is used to output corresponding new versions of DRC configuration files according to the DR data of a specific process node input.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the DRC rule formulation method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the DRC rule formulation method according to any one of claims 1 to 6 is implemented.

Images