Engineering state constrained multi-tool adaptation EDA tool question and answer method, apparatus, and media

Abstract

The present application relates to the technical field of electronic design automation, and particularly relates to a multi-tool adaptation EDA tool question and answer method, equipment and medium of engineering state constraint, the method comprises the following steps: each EDA tool constructs a tool knowledge base and a semantic vector library; receiving the log content selected by a user and inputting a question and vectorizing; collecting a regression full context to generate a structured context descriptor containing tool constraints; locking the target tool knowledge base according to the tool name to carry out semantic retrieval, and obtaining candidate knowledge entries; inputting the user question, the structured context descriptor and the candidate knowledge entries into a large language model, screening out matched knowledge entries and constraint rules as reasoning range; inputting the log content selected by the user, the user question, the structured context descriptor and the reasoning range into the large language model again to generate engineering operation guidance. The present application improves the solution efficiency and accuracy of EDA tool knowledge question and answer.

Description

Technical Field

[0001] This invention relates to the field of Electronic Design Automation (EDA) technology, and in particular to a multi-tool adaptation EDA tool question-and-answer method, device and medium for engineering state constraints. Background Technology

[0002] In the integrated circuit verification process, regression testing is a core step in ensuring the correctness of chip functionality. It requires the execution of massive numbers of test tasks to verify whether code modifications have introduced new errors. With the evolution of chip manufacturing processes to 7nm and below, the workload of a single regression test can reach tens of thousands, generating a huge amount of log files, configuration data, and execution status information. During regression testing, verification engineers need to locate problems generated by various EDA tools based on the log content. The solutions to these problems must be linked to the specific EDA tool type, tool version, task configuration parameters, log context, and historical operation records to provide accurate and executable solutions.

[0003] Currently, verification engineers primarily acquire tool usage knowledge and problem solutions through three methods: First, they consult static technical documentation of EDA tools. However, these documents are scattered and lack awareness of dynamic engineering conditions, making manual filtering time-consuming and inefficient. Second, they use general-purpose intelligent auxiliary tools based on large language models. However, these tools lack multi-tool-specific knowledge and verification task details, easily generating responses that do not match the actual engineering environment. Third, they utilize intelligent question-and-answer systems based on retrieval enhancement. However, these systems use a fixed document set for retrieval, failing to categorize by tool type or consider the dynamic running status of regression testing. This results in search results often containing invalid content from non-target tools, making it difficult to guarantee the relevance of answers and the executability of commands. These shortcomings cause verification engineers to spend an average of over 20 minutes locating and resolving tool operation problems during regression testing, severely slowing down verification efficiency and even delaying chip development cycles. Therefore, improving the efficiency and accuracy of EDA tool knowledge-based question-and-answer solutions has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-tool adaptation EDA tool question-answering method, device, and medium for engineering state constraints, thereby improving the efficiency and accuracy of EDA tool knowledge question answering.

[0005] According to a first aspect of the present invention, a multi-tool adaptation EDA tool question-and-answer method for engineering state constraints is provided, comprising:

[0006] Step S1: Construct a corresponding tool knowledge base and semantic vector library for each EDA tool. Each tool knowledge base includes multiple knowledge entries. Each knowledge entry includes a question description, the answer content corresponding to the question description, and a scene tag. The question description of each knowledge entry corresponds to a semantic vector in the corresponding semantic vector library.

[0007] Step S2: Receive the log content selected by the EDA tool user and the questions entered by the user, and convert the questions entered by the user into corresponding input question vectors;

[0008] Step S3: Based on the log content selected by the user, collect the full context of regression execution, and generate a structured context descriptor based on the collected full context of regression execution. The structured context descriptor includes tool constraints.

[0009] Step S4: Determine the target tool knowledge base based on the EDA tool name in the tool constraint of the structured context descriptor, and perform semantic similarity retrieval in the semantic vector library corresponding to the target tool knowledge base based on the input question vector to obtain a set of candidate knowledge entries;

[0010] Step S5: Input the user-input question, the structured context descriptor, and the candidate knowledge item set into the large language model. The large language model then selects matching knowledge items and their corresponding constraint rules from the candidate knowledge item set based on the user-input question and the structured context descriptor, which serve as the reasoning scope.

[0011] Step S6: Input the log content selected by the user, the question entered by the user, the structured context descriptor, and the inference scope back into the large language model. The large language model generates and outputs engineering operation guidelines, which are used to guide the user to perform corresponding operations in the EDA tool.

[0012] According to a second aspect of the present invention, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the method described in the first aspect of the present invention.

[0013] According to a third aspect of the present invention, a computer-readable storage medium is provided, storing computer-executable instructions for performing the method described in the first aspect of the present invention.

[0014] Compared with existing technologies, this invention has significant advantages and beneficial effects. Through the above technical solution, the multi-tool adaptation EDA tool question-and-answer method, device, and medium provided by this invention for engineering state constraints achieve considerable technological advancement and practicality, and have broad industrial application value. It has at least the following beneficial effects:

[0015] This invention achieves precise isolation and categorized retrieval of knowledge from multiple tools by constructing an independent knowledge base for each EDA tool and setting multi-dimensional scene tags. By collecting the full context of regression execution and generating structured context descriptors containing tool constraints, the system can perceive the engineering environment of the regression task in real time, overcoming the shortcomings of existing technologies where the lack of state awareness leads to mismatches between retrieval results and user needs, and inconsistencies between output content and the actual engineering environment. Furthermore, a two-stage inference architecture is adopted. First, a large language model filters knowledge entries that match the user's question and constraints, along with the constraint rules they satisfy, to form an inference scope. Then, based on this inference scope, engineering operation guidelines are generated. By precisely defining the knowledge boundary through the inference scope, the large language model effectively avoids generating response content detached from engineering basis. This invention effectively improves the efficiency and accuracy of knowledge question answering for EDA tools. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0017] Figure 1 Flowchart of the multi-tool adaptation EDA tool question-and-answer method for engineering state constraints provided in this embodiment of the invention;

[0018] Figure 2 This is a schematic diagram of the electronic device structure provided in an embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram of a computer-readable storage medium structure provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides a multi-tool adaptation EDA tool question-and-answer method for engineering state constraints, such as... Figure 1 As shown, it includes:

[0022] Step S1: Construct a corresponding tool knowledge base and semantic vector library for each EDA tool. Each tool knowledge base includes multiple knowledge entries. Each knowledge entry includes a question description, the answer content corresponding to the question description, and a scene tag. The question description of each knowledge entry corresponds to a semantic vector in the corresponding semantic vector library.

[0023] It should be noted that step S1 establishes an independent knowledge base for each EDA tool, with each knowledge base containing multiple knowledge entries. Each knowledge entry consists of three parts: a problem description, an answer content, and a scenario tag. Specifically, the problem description can be the typical problem addressed by the knowledge, the answer content can be the specific solution to the problem, and the scenario tag can be the tool name, version, functional module, task status, and operation type to which the knowledge is applicable. Simultaneously, the problem description of each knowledge entry is converted into a semantic vector using a semantic vectorization model and stored in a semantic vector library for subsequent rapid vector similarity retrieval. Step S1 lays the data foundation for this invention.

[0024] Step S2: Receive the log content selected by the EDA tool user and the questions entered by the user, and convert the questions entered by the user into corresponding input question vectors.

[0025] It should be noted that step S2 receives the log content selected by the EDA tool user as a clue in the regression test log, and at the same time receives the natural language question input by the user, and converts the question into vector form to facilitate subsequent similarity calculation with the vector in the knowledge base, thus completing the user interaction input and processing of the input information.

[0026] Step S3: Based on the log content selected by the user, collect the full regression execution context, and generate a structured context descriptor based on the collected full regression execution context. The structured context descriptor includes tool constraints.

[0027] It should be noted that step S3 automatically collects the full engineering context of the current regression task based on the log content selected by the user, including tool name, tool version, task attributes, configuration file content, log information, and historical operation records. This multi-dimensional information is then integrated into a standardized structured context descriptor, which at least includes tool constraints, namely the tool name and version used in the current task, serving as the basis for the engineering status of all subsequent retrieval and inference.

[0028] Step S4: Determine the target tool knowledge base based on the EDA tool name in the tool constraint of the structured context descriptor, and perform semantic similarity retrieval in the semantic vector library corresponding to the target tool knowledge base based on the input question vector to obtain a set of candidate knowledge entries.

[0029] It should be noted that step S4 extracts the name of the currently used EDA tool from the structured context descriptor and directly locates the corresponding tool knowledge base based on this name, effectively avoiding the retrieval of irrelevant knowledge from other tools. Then, based on the vector of the user's input question, it retrieves the several knowledge entries with the highest semantic similarity from the semantic vector library corresponding to the tool knowledge base, forming a set of candidate knowledge entries. Step S4 achieves accurate knowledge localization in multi-tool scenarios.

[0030] Step S5: Input the user-input question, the structured context descriptor, and the candidate knowledge item set into the large language model. The large language model then selects matching knowledge items and their corresponding constraint rules from the candidate knowledge item set based on the user-input question and the structured context descriptor, which serve as the reasoning scope.

[0031] It should be noted that in step S5, the user-inputted question, structured context descriptor, and candidate knowledge item set are input into the large language model. The large language model then precisely selects knowledge items from the candidate set that are both semantically relevant and satisfy the current engineering constraints, along with the constraint rules they satisfy, based on the semantics of the user's question and the engineering state constraints expressed by the structured context descriptor, thus forming the reasoning scope. Step S5 completes the first stage of intelligent reasoning, the core of which lies in accurately defining the knowledge boundaries.

[0032] Step S6: Input the log content selected by the user, the question entered by the user, the structured context descriptor, and the inference scope back into the large language model. The large language model generates and outputs engineering operation guidelines, which are used to guide the user to perform corresponding operations in the EDA tool.

[0033] It should be noted that in step S6, the user-selected log content, the user-input question, the structured context descriptor, and the inference scope generated in step S5 are input again into the large language model. The large language model then generates directly executable engineering operation guidelines based on the predefined knowledge scope. These guidelines are strictly matched to the current engineering environment and can be directly applied to the production environment. Step S6 completes the second stage of intelligent generation.

[0034] This embodiment achieves precise isolation and classification retrieval of knowledge from multiple tools by building an independent knowledge base for each EDA tool and setting multi-dimensional scene tags. By collecting the full context of regression execution and generating a structured context descriptor containing tool constraints, the system can perceive the engineering environment of the regression task in real time, overcoming the shortcomings of existing technologies where the lack of state awareness leads to mismatches between search results and user needs, and inconsistencies between output content and the actual engineering environment. At the same time, a two-stage inference architecture is adopted. First, the large language model filters knowledge items that match the user's problem and constraints and the constraint rules they satisfy to form an inference scope. Then, based on this inference scope, engineering operation guidelines are generated. By precisely defining the knowledge boundary through the inference scope, the large language model effectively avoids generating response content that is detached from engineering basis.

[0035] As one embodiment, step S1 includes:

[0036] Step S11: Obtain the original technical documentation for each EDA tool.

[0037] It should be noted that the original technical documentation for EDA tools may include technical manuals, user guides, troubleshooting case studies, etc. for various EDA tools.

[0038] Step S12: Perform format cleaning and text segmentation on the original technical document of each EDA tool. The text segmentation method is as follows: For the original technical document of each EDA tool, check the text format paragraph by paragraph. If the paragraph is found to have a title format, the title and the first sentence of the main text of the chapter corresponding to the title are used as the question description. All the content from the title to the next same-level title is used as the answer content corresponding to the question description, and an independent knowledge entry is generated.

[0039] It's important to note that format cleaning can be tailored to different document formats: for PDF documents, the PyPDF2 or pdfplumber libraries are used for plain text extraction; for HTML documents, the BeautifulSoup library is used to parse the document structure and remove HTML tags and script code; for Markdown documents, Markdown syntax marks are stripped to extract plain text information. Regarding text segmentation, the system checks the text format paragraph by paragraph. When a heading paragraph is found, the heading and the first sentence of the main text of the section containing that heading are combined to form a question description. The entire content from that heading to the next sibling heading is then used as the answer to that question description, thus generating an independent, self-contained knowledge entry. This segmentation method ensures that each knowledge entry is a complete question-answer pair.

[0040] Step S13: Add scene tags to each knowledge entry. The scene tags include tool name tag, tool version tag, function module tag, task status tag, and operation type tag.

[0041] It should be noted that step S13 adds multi-dimensional scenario tags to each knowledge entry, including tool name tags, tool version tags, functional module tags, task status tags, and operation type tags, enabling each knowledge entry to be retrieved and filtered from multiple dimensions. Specifically, the functional module tags may include regression scheduling function tags and log analysis function tags. The task status tags may include task running tags, task timeout tags, task failed tags, and task passed tags. The operation type tags may include log analysis operation tags, configuration modification operation tags, and command retry operation tags. Each knowledge entry can be stored in JSON format, including a unique knowledge identifier, a question description, answer content, and scenario tag fields.

[0042] Step S14: Convert the problem description in each knowledge entry into a semantic vector using a semantic vectorization model and store it in the corresponding semantic vector library.

[0043] It should be noted that step S14 converts the question description of each knowledge item into a semantic vector through a semantic vectorization model, and stores it in the corresponding semantic vector library, providing an index basis for subsequent semantic similarity retrieval.

[0044] Steps S11-S14 ensure standardized processing of original technical data by employing differentiated cleaning methods for different document formats; generate independent and complete question-answer knowledge units by accurately segmenting the title hierarchy of the document structure; enable precise retrieval and filtering of knowledge from multiple dimensions by adding multi-dimensional scene tags to each knowledge item; and provide a data foundation for subsequent efficient and accurate semantic similarity retrieval by vectorizing the question description and storing it in a vector library.

[0045] As one embodiment, step S3 includes:

[0046] Step S31: Collect the task attributes of regression testing, regression log directory address, task execution directory address, and executed operation records through the application programming interface of the verification platform running the EDA tool.

[0047] It should be noted that step S31 can specifically utilize the RESTful API interface provided by the verification platform to automatically collect regression test task attributes, log directory address, task execution directory address, and executed operation records. Task attributes may specifically include task ID, task status, and retries. For example, calling ` / api / v1 / tasks / {task_id}` retrieves the regression test task attributes, and calling ` / api / v1 / tasks / task_history / {history}` retrieves the executed operation records.

[0048] Step S32: Access the task execution directory to extract the contents of the configuration file, and access the regression log directory to extract the contents of the log file.

[0049] It should be noted that step S32 involves accessing the task execution directory to extract the configuration file, accessing the regression log directory to extract the complete log file, and obtaining all configuration parameters and log outputs during the task execution process.

[0050] Step S33: Use regular expressions to match the EDA tool name and tool version from the log file.

[0051] It should be noted that step S33 can use regular expressions to match the EDA tool name and version information used by the current task from the header of the log file.

[0052] Step S34: Generate tool constraints based on the collected EDA tool name and tool version, generate task attribute constraints based on the task attributes, generate configuration constraints based on the content of the configuration file, generate log constraints based on the content of the log file, generate operation record constraints based on the executed operation records, and integrate the tool constraints, task attribute constraints, configuration constraints, log constraints, and operation record constraints to generate the structured context descriptor.

[0053] It should be noted that step S34 converts the collected information into corresponding structured constraints: tool constraints are generated based on tool name and version, task attribute constraints are generated based on task attributes, configuration constraints are generated based on configuration file content, log constraints are generated based on log file content, and operation record constraints are generated based on executed operation records. Then, these five types of constraints are integrated into a unified structured context descriptor, which serves as the engineering state basis for all subsequent retrieval and reasoning steps.

[0054] Steps S31-S34 use a triple acquisition method—API interface, directory access, and regular expressions—to dynamically obtain full-dimensional engineering status information during the regression execution process. This information is then integrated according to a structured format based on five types of constraints, enabling the system to comprehensively and in real-time perceive the specific operating environment of the current regression task.

[0055] As one embodiment, step S33 includes:

[0056] Step S331: Extract the text of the preset number of lines at the beginning of the log file as the current matching range.

[0057] It should be noted that since the tool name and version information are usually recorded in the header of the log file, step S331 first extracts a preset number of lines from the header of the log file as the initial matching range.

[0058] Step S332: Using a regular expression that adapts to the EDA tool name and version format, perform a matching operation on each line of the text in the current matching range to extract the EDA tool name and tool version.

[0059] It should be noted that step S332 uses a pre-built regular expression adapted to the format of EDA tool names and versions to perform line-by-line matching on the text within the matching range to extract the EDA tool names and versions.

[0060] Step S333: If no valid EDA tool name or tool version is extracted, expand the number of lines to be extracted in the current matching range and return to step S332 to re-execute the matching operation.

[0061] It should be noted that in step S333, if a valid EDA tool name or tool version cannot be extracted within the current matching range, the number of lines to be extracted in the current matching range will be automatically expanded, and the process will return to step S332 to re-execute the matching operation. If no results are extracted after re-execution, the range can be expanded again and the matching operation can be executed again, forming a repeatable iterative backtracking and retry mechanism until a valid tool name and version information are successfully extracted or the preset maximum number of lines to be extracted is reached, ensuring that tool information can be successfully extracted from log files of different formats.

[0062] Steps S331-S333 employ an automatic fallback mechanism that involves truncating the header, performing regular expression matching, and iteratively expanding the scope for retrying upon failure. This mechanism adapts to the differences in log file formats of various EDA tools, ensuring accurate extraction of tool names and version information under various log formats and improving the robustness and versatility of context collection.

[0063] As one embodiment, step S4 includes:

[0064] Step S41: Extract the EDA tool name from the tool constraints of the structured context descriptor, and determine the EDA tool knowledge base corresponding to the EDA tool name as the target tool knowledge base.

[0065] It should be noted that step S41 extracts the name of the EDA tool used in the current task from the tool constraints of the structured context descriptor, and locates the corresponding tool knowledge base among multiple tool knowledge bases based on the name as the target tool knowledge base.

[0066] Step S42: Using the cosine similarity calculation method, calculate the similarity between each semantic vector and the input question vector in the semantic vector library corresponding to the target tool knowledge base.

[0067] It should be noted that step S42 uses the cosine similarity calculation method to calculate the similarity between the semantic vector of each knowledge item and the user input question vector in the semantic vector library corresponding to the located target tool knowledge base.

[0068] Step S43: Sort the semantic vectors from high to low similarity, select the knowledge entries corresponding to the first preset number of semantic vectors, and generate a candidate knowledge entry set.

[0069] It should be noted that in step S43, the knowledge items corresponding to the top-ranked semantic vectors are selected from the highest to the lowest similarity scores and used as a set of candidate knowledge items for further screening in subsequent steps.

[0070] Steps S41-S43 accurately locate the corresponding knowledge base by tool name, narrowing the search scope and avoiding cross-tool knowledge confusion and interference from irrelevant documents; using cosine similarity for semantic-level matching can identify technical expressions with similar semantics but different wording, improving the recall and accuracy of the search.

[0071] As one embodiment, step S5 includes:

[0072] Step S51: Input the user-inputted question, the structured context descriptor, and the candidate knowledge item set into the large language model through the first prompt word project. The first prompt word project includes four dimensions: role definition, input specifications, constraint mechanism, and output template.

[0073] It should be noted that step S51 inputs the user question, structured context descriptor, and candidate knowledge item set into the large language model through the first prompt word project. This prompt word project comprises four dimensions: role definition, input specifications, constraint mechanism, and output template. The role definition explicitly specifies the model's core capabilities in parsing YAML constraint rules and executing multi-dimensional condition matching. The input specifications systematically define a triplet data structure, including the user's natural language question, a structured context descriptor containing five types of constraint rules, and a candidate set of original knowledge units. The constraint mechanism implements a dynamic activation strategy, ensuring 100% effectiveness of version constraints, while intelligently triggering other constraint types based on question keywords, and employing a strict literal matching principle to ensure filtering accuracy. The output template specifies a YAML format response structure, requiring the retention of original knowledge fragment content and annotation of matching constraint paths. The first prompt word project, through precise context boundary control, effectively filters irrelevant knowledge fragments while maintaining the complete transmission of key constraint information, providing high-precision knowledge input for subsequent operation guidance generation.

[0074] Step S52: The large language model selects matching knowledge entries and the constraint rules they satisfy from the candidate knowledge entry set based on the user input question and the structured context descriptor.

[0075] It should be noted that in step S52, the large language model accurately selects knowledge items that are both semantically relevant and satisfy the current engineering constraints, along with the specific constraint rules they satisfy, from the candidate set based on the semantics of the user's question and the engineering state constraints in the structured context descriptor.

[0076] Step S53: The selected knowledge items and the constraint rules they satisfy are taken as the reasoning scope, wherein the reasoning scope is the knowledge items that conform to the constraint conditions described by the structured context descriptor and the constraint rules they satisfy.

[0077] It should be noted that step S53 defines the selected knowledge items and the constraint rules they satisfy as the reasoning scope, that is, the knowledge boundary defined by the system. All subsequent answer generation will be strictly based on the knowledge within this scope.

[0078] Steps S51-S53, through the structured design of the four-dimensional prompt word engineering, achieve refined screening of candidate knowledge items; by introducing the reasoning scope as an intermediate product, the generation space of the large language model is precisely limited, providing a high-precision knowledge input foundation for the generation of subsequent engineering operation guidance.

[0079] As one embodiment, step S6 includes:

[0080] Step S61: Input the log content selected by the user, the question entered by the user, the structured context descriptor, and the inference scope into the large language model through the second prompt word project. The second prompt word project includes four dimensions: role definition, input instructions, constraints, and output format. The constraints are used to force the answer to be limited to the use of only the provided knowledge, and to specify the accuracy of command parameters, path compliance, and security boundaries.

[0081] It should be noted that step S61 inputs the user-selected log content, user question, structured context descriptor, and inference scope into the large language model through the second prompt word engineering. This prompt word engineering includes four dimensions: role definition, input instructions, constraints, and output format, constructing a precise engineering instruction framework. Specifically, the role definition defines the agent's professional positioning and responsibility boundaries; the input instructions specify all available data sources, including the user's original question, structured context descriptors, and a constrained set of candidate knowledge units, defining the scope of knowledge input; the constraints set knowledge usage boundaries, limiting responses to be generated only based on provided candidate knowledge, excluding external knowledge references, and setting command parameter accuracy, path compliance, and security boundaries; the output format defines a standardized response structure, specifying an engineering guidance template containing elements such as root cause analysis, operational purpose, step-by-step verification process, risk warning, and knowledge tracing, so that the output content can be directly applied to the production environment. This prompt word engineering, through a multi-dimensional constraint mechanism, avoids the large language model generating response content that deviates from engineering basis, while ensuring that the generated operation instructions conform to verification engineering practice specifications, improving problem-solving efficiency and system security.

[0082] Step S62: Generate an engineering operation guide from the large language model. The engineering operation guide includes problem analysis, operation purpose, step-by-step process, core commands, precautions and knowledge sources. The knowledge sources include knowledge entries that conform to the structured context descriptor.

[0083] It should be noted that in step S62, the large language model generates an engineering operation guide that includes problem analysis, operation purpose, step-by-step process, core commands, precautions and knowledge tracing information within the knowledge boundary of the reasoning scope.

[0084] Step S63: Output the engineering operation guide, which is used to guide users to perform corresponding operations in the EDA tool.

[0085] It should be noted that step S63 outputs this engineering operation guide, which is used to directly guide users to perform corresponding troubleshooting and repair operations in EDA tools.

[0086] Steps S61-S63, through the structured design of the four-dimensional second prompt word engineering, force the large language model to answer only based on the provided knowledge, and stipulate engineering specifications such as the accuracy of command parameters, path compliance, and security boundaries, effectively preventing the model from generating response content out of context; at the same time, the output engineering operation guidance adopts a standardized structure to ensure that the output content can be directly applied to the actual engineering operation environment.

[0087] As one embodiment, the method further includes:

[0088] Step S7: Obtain user questions, knowledge items, and operation instructions generated during the question-and-answer interaction process, and conduct quality assessment and knowledge archiving.

[0089] Step S7 includes:

[0090] Step S71: Perform a quality assessment on the user questions, the knowledge entries, and the operation guidelines, and filter out the interactive data that meets the preset quality conditions.

[0091] It should be noted that step S71 performs a quality assessment on the user questions generated during the question-and-answer interaction, the adopted knowledge items, and the generated operation instructions, and filters out the interaction data that meets the preset quality conditions. The triggering methods for the preset quality conditions may include: engineers performing a confirmation operation through the interactive interface, or the same high-frequency question being asked multiple times with consistent answers, indicating that the question has a recurring characteristic.

[0092] Step S72: Extract user questions and operation instructions from the filtered interaction data and integrate them to generate new knowledge entries.

[0093] It should be noted that in step S72, user questions and operation instructions are extracted from the high-quality interactive data and integrated into new knowledge entries, forming a standardized question description and answer content structure.

[0094] Step S73: Attach the scene tag from the structured context descriptor to the new knowledge entry, associate it with the current EDA tool version and specify the applicable version range, store the new knowledge entry in the corresponding EDA tool knowledge base and update the corresponding semantic vector library synchronously.

[0095] It should be noted that step S73 involves attaching a scene tag from the structured context descriptor to the new knowledge entry, associating it with the current EDA tool version and specifying the applicable version range, and then storing the knowledge entry in the knowledge base of the corresponding tool and synchronously updating the semantic vector library to achieve incremental updates of knowledge.

[0096] Steps S71-S73 establish a self-evolving closed loop for acquiring and filtering question-and-answer interaction data, structural archiving, and incremental updates. This enables the knowledge base to continuously iterate and optimize with engineering practice, avoids redundant and ineffective knowledge, and allows the system's problem-solving capabilities and scenario adaptation accuracy to improve synchronously with usage frequency.

[0097] The method described in this invention achieves precise isolation and categorized retrieval of knowledge across multiple tools by constructing an independent knowledge base for each EDA tool and setting multi-dimensional scenario tags, avoiding cross-tool knowledge confusion and interference from irrelevant documents. It dynamically acquires multi-dimensional engineering status information of regression execution and generates structured context descriptors through a triple acquisition method of API interfaces, directory access, and regular expressions, enabling the system to perceive the engineering environment of the current regression task in real time. This overcomes the shortcomings of existing technologies, such as mismatch between search results and user needs, and inconsistency between output content and the actual engineering environment due to a lack of engineering status awareness. A two-stage inference architecture is adopted. First, a large language model filters knowledge entries that match the user's question and constraints, along with the constraint rules they satisfy, to form an inference scope. Then, based on this inference scope, engineering operation guidelines are generated. By precisely defining the knowledge boundary through the inference scope, the large language model effectively avoids generating response content detached from engineering basis, ensuring that the output operation guidelines strictly match the actual engineering environment and can be directly implemented. Furthermore, the self-evolving closed loop of question-and-answer interaction data acquisition and filtering, structured archiving, and incremental updates enables the knowledge base to continuously iterate and optimize with engineering practice. This invention effectively solves the problem of accuracy in knowledge retrieval under multi-tool environments, and significantly improves the efficiency and accuracy of problem localization and resolution in EDA tool regression testing scenarios.

[0098] It should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the steps as sequential processes, many of these steps can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the steps can be rearranged. A process can be terminated when its operation is complete, but it may also have additional steps not included in the figures. A process can correspond to a method, function, procedure, subroutine, subroutine, etc.

[0099] This invention also provides an electronic device, such as... Figure 2 As shown, it includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the method described in the embodiments of the present invention.

[0100] For example, the electronic device may be a mobile electronic device or a non-mobile electronic device. Further, a mobile electronic device may be any one of a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA); a non-mobile electronic device includes any one of a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine.

[0101] This invention also provides a computer-readable storage medium, such as... Figure 3 As shown, computer-executable instructions are stored thereon, which are used to execute the methods described in the embodiments of the present invention.

[0102] Examples of computer-readable storage media include: computer read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), compact disc read-only memory (CD-ROM), flash memory, magnetic disk, optical disk, portable computer disk, hard disk and / or solid-state drive, etc.

[0103] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-tool adaptation EDA tool question-and-answer method for engineering state constraints, characterized in that, include: Step S1: Construct a corresponding tool knowledge base and semantic vector library for each EDA tool. Each tool knowledge base includes multiple knowledge entries. Each knowledge entry includes a question description, the answer content corresponding to the question description, and a scene tag. The question description of each knowledge entry corresponds to a semantic vector in the corresponding semantic vector library. Step S2: Receive the log content selected by the EDA tool user and the questions entered by the user, and convert the questions entered by the user into corresponding input question vectors; Step S3: Based on the log content selected by the user, collect the full context of regression execution, and generate a structured context descriptor based on the collected full context of regression execution. The structured context descriptor includes tool constraints. Step S4: Determine the target tool knowledge base based on the EDA tool name in the tool constraint of the structured context descriptor, and perform semantic similarity retrieval in the semantic vector library corresponding to the target tool knowledge base based on the input question vector to obtain a set of candidate knowledge entries; Step S5: Input the user-input question, the structured context descriptor, and the candidate knowledge item set into the large language model. The large language model then selects matching knowledge items and their corresponding constraint rules from the candidate knowledge item set based on the user-input question and the structured context descriptor, which serve as the reasoning scope. Step S6: Input the log content selected by the user, the question entered by the user, the structured context descriptor, and the inference scope back into the large language model. The large language model generates and outputs engineering operation guidelines, which are used to guide the user to perform corresponding operations in the EDA tool.

2. The method according to claim 1, characterized in that, Step S1 includes: Step S11: Obtain the original technical documentation for each EDA tool; Step S12: Perform format cleaning and text segmentation on the original technical document of each EDA tool. The text segmentation method is as follows: For the original technical document of each EDA tool, check the text format paragraph by paragraph. If the paragraph is found to have a title format, the title and the first sentence of the main text of the chapter corresponding to the title are used as the question description. All the content from the title to the next same-level title is used as the answer content corresponding to the question description, and an independent knowledge entry is generated. Step S13: Add scene tags to each knowledge entry. The scene tags include tool name tag, tool version tag, functional module tag, task status tag, and operation type tag. Step S14: Convert the problem description in each knowledge entry into a semantic vector using a semantic vectorization model and store it in the corresponding semantic vector library.

3. The method according to claim 1, characterized in that, Step S3 includes: Step S31: Collect the task attributes of regression testing, regression log directory address, task execution directory address, and executed operation records through the application programming interface of the verification platform running the EDA tool; Step S32: Access the task execution directory to extract the contents of the configuration file, and access the regression log directory to extract the contents of the log file; Step S33: Use regular expressions to match the EDA tool name and tool version from the log file; Step S34: Generate tool constraints based on the collected EDA tool name and tool version, generate task attribute constraints based on the task attributes, generate configuration constraints based on the content of the configuration file, generate log constraints based on the content of the log file, generate operation record constraints based on the executed operation records, and integrate the tool constraints, task attribute constraints, configuration constraints, log constraints, and operation record constraints to generate the structured context descriptor.

4. The method according to claim 3, characterized in that, Step S33 includes: Step S331: Extract the text of a preset number of lines from the beginning of the log file as the current matching range; Step S332: Using a regular expression that adapts to the format of EDA tool name and version, perform a matching operation on each line of the text in the current matching range to extract the EDA tool name and tool version; Step S333: If no valid EDA tool name or tool version is extracted, expand the number of lines to be extracted in the current matching range, and then return to step S332 to re-execute the matching operation.

5. The method according to claim 1, characterized in that, Step S4 includes: Step S41: Extract the EDA tool name from the tool constraints of the structured context descriptor, and determine the EDA tool knowledge base corresponding to the EDA tool name as the target tool knowledge base; Step S42: Using the cosine similarity calculation method, calculate the similarity between each semantic vector and the input question vector in the semantic vector library corresponding to the target tool knowledge base; Step S43: Sort the semantic vectors from high to low similarity, select the knowledge entries corresponding to the first preset number of semantic vectors, and generate a candidate knowledge entry set.

6. The method according to claim 1, characterized in that, Step S5 includes: Step S51: Input the user-inputted question, the structured context descriptor, and the candidate knowledge item set into the large language model through the first prompt word project. The first prompt word project includes four dimensions: role definition, input specifications, constraint mechanism, and output template. Step S52: The large language model selects matching knowledge items and the constraint rules they satisfy from the candidate knowledge item set based on the user input question and the structured context descriptor. Step S53: The selected knowledge items and the constraint rules they satisfy are taken as the reasoning scope, wherein the reasoning scope is the knowledge items that conform to the constraint conditions described by the structured context descriptor and the constraint rules they satisfy.

7. The method according to claim 1, characterized in that, Step S6 includes: Step S61: Input the log content selected by the user, the question entered by the user, the structured context descriptor, and the inference scope into the large language model through the second prompt word engineering. The second prompt word engineering includes four dimensions: role definition, input instructions, constraints, and output format. The constraints are used to enforce that only the provided knowledge is used to answer the question, and to specify the accuracy of command parameters, path compliance, and security boundaries. Step S62: Generate an engineering operation guide from the large language model. The engineering operation guide includes problem analysis, operation purpose, step-by-step process, core commands, precautions and knowledge sources. The knowledge sources include knowledge entries that conform to the structured context descriptor. Step S63: Output the engineering operation guide, which is used to guide users to perform corresponding operations in the EDA tool.

8. The method according to claim 1, characterized in that, Also includes: Step S7: Obtain user questions, knowledge items, and operation instructions generated during the question-and-answer interaction process, and conduct quality assessment and knowledge archiving; Step S7 includes: Step S71: Perform a quality assessment on the user questions, the knowledge entries, and the operation instructions, and filter out the interactive data that meets the preset quality conditions; Step S72: Extract user questions and operation instructions from the filtered interaction data, and integrate them to generate new knowledge entries; Step S73: Attach the scene tag from the structured context descriptor to the new knowledge entry, associate it with the current EDA tool version and specify the applicable version range, store the new knowledge entry in the corresponding EDA tool knowledge base and update the corresponding semantic vector library synchronously.

9. An electronic device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that are executed by the at least one processor, the instructions being configured to perform the method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The device stores computer-executable instructions for performing the method as described in any one of claims 1-8.

Images