Design optimization system and method based on AI

Abstract

The embodiment of the invention relates to an AI-based design optimization system and method. The system comprises an AI module, a processing analysis module, a DOE module, an EDA module and an optimization module. And the AI module is connected with the processing analysis module, the DOE module, the optimization module and the EDA calling module, and is used for receiving the design task, performing disassembly, distribution and process planning on the design task, and optimizing the process and parameters of the design task based on a preset AI model. According to the technical scheme provided by the embodiment of the invention, the AI tool represented by the large language model is fused into the parameter optimization process, and full-autonomous decision and control of links such as parameter control, optimization control, experimental design, data analysis and application in the whole optimization process are realized by utilizing experience knowledge and professional knowledge built in the large language AI model, so that the optimization efficiency is improved. And the optimization process from'task input 'to'task output' without human intervention in a real sense is realized.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computer artificial intelligence data processing technology, and in particular to an AI-based design optimization system and method. Background Technology

[0002] Currently, the R&D and production models in various industries (such as the integrated circuit EDA field) heavily rely on traditional, manually-driven processes. The entire chain, from experimental design to data analysis, requires significant human intervention: the formulation of experimental plans depends on the accumulated experience of R&D personnel, and key aspects such as variable setting and sample selection are often affected by individual cognitive limitations; during experiments, data collection largely relies on manual recording or semi-automated equipment, which is not only time-consuming and labor-intensive but also prone to introducing data bias due to operational errors; even in the analysis phase, when using data statistical analysis software for parameter optimization and pattern summarization, R&D personnel still need to manually input data, set analysis model parameters, and judge the rationality of the results based on experience.

[0003] Current engineering optimization problems heavily rely on human mental labor. For example, the selection of optimization parameters, the determination of the range of optimization parameters, the use of optimization algorithms, and experimental design all require human experience and knowledge as a foundation, which severely limits the degree of automation and efficiency in the optimization process.

[0004] The drawbacks of this model are particularly prominent in complex R&D scenarios. On the one hand, it consumes enormous amounts of manpower and resources, and requires a high level of experience and knowledge from personnel: a large number of repetitive experiments require significant amounts of laboratory equipment, raw materials, and manpower, especially in fields such as pharmaceuticals, new materials, and high-end manufacturing, where the long-term accumulation of resource waste is significant. On the other hand, R&D efficiency is low and the cycle is lengthy. In the traditional model, determining an optimization parameter often requires several weeks or even months of iteration, making it difficult to adapt to the rapidly iterating market demands. More importantly, reliance on experience leads to limitations in R&D results: human judgment is easily influenced by fixed mindsets, making it difficult to break through existing knowledge frameworks to discover potential patterns. For example, in complex systems with multiple coupled variables, manual parameter tuning often only finds local optimal solutions rather than global optimal solutions, which greatly restricts the depth and breadth of R&D innovation.

[0005] More specifically, in EDA applications, such as device simulation, SPICE parameter extraction, and design, a large number of simulation experiments are needed to find the optimal design parameters. Facing complex physical mechanisms and stringent mathematical solution conditions, a lot of time and manpower are required. Finally, the design is verified through experimental fabrication, resulting in a relatively long R&D design cycle. Summary of the Invention

[0006] Based on the above-mentioned situation of the prior art, the purpose of this invention is to provide an AI-based design optimization system and method that integrates AI large language models into the design system to achieve fully autonomous decision-making and control over parameter control, optimization control, experimental design, data analysis and application in the entire design process.

[0007] To achieve the above objectives, according to one aspect of the present invention, an AI-based design optimization system is provided, the system comprising an AI module, a processing and analysis module, a DOE module, an EDA module, and an optimization module;

[0008] The AI ​​module is interconnected with the processing and analysis module, DOE module, optimization module and EDA calling module, and is used to receive design tasks, decompose and allocate design tasks and plan processes, and optimize the process and parameters of design tasks based on a preset AI model.

[0009] The processing and analysis module is used to analyze and optimize the data information of the design task based on the analysis commands of the AI ​​module;

[0010] The DOE module is used to invoke optimization parameter data based on the construction command of the AI ​​module, and to construct the DOE based on the optimization parameter data;

[0011] The EDA module is used to perform simulation based on the call commands from the AI ​​module and DOE data.

[0012] The optimization module is used to model and solve for the optimal parameter data based on the optimization commands, optimization parameter data, and simulation result data from the AI ​​module.

[0013] Furthermore, the system also includes a data module, which is interconnected with the AI ​​module, processing and analysis module, DOE module, EDA module and optimization module;

[0014] The AI ​​module is used to write the design task data information into the data module;

[0015] The data module is used to receive the result data generated by each module and to provide each module with the data required for operation.

[0016] Furthermore, the EDA module is also used to perform simulation operations based on the calling commands and optimal parameter data of the AI ​​module;

[0017] The AI ​​module is also used to verify the simulation results based on the preset AI model.

[0018] Furthermore, the AI ​​module verifies the simulation results based on the preset AI model, including:

[0019] If the simulation results meet the preset design objectives, a design report is generated;

[0020] If the simulation results do not meet the preset design goals, the optimization will be re-executed.

[0021] Furthermore, the re-execution of optimization includes any one of the following: returning to the processing and analysis module to analyze and optimize the data information of the design task, returning to the DOE module to construct the DOE, and returning to the optimization module to model and solve for the optimal parameter data.

[0022] Furthermore, the optimization module models and solves for optimal parameter data based on the optimization commands, optimization parameter data, and simulation results data from the AI ​​module, including:

[0023] The solution domain is obtained by fitting the optimized parameter data and the simulation results data to the data relationship.

[0024] In the solution domain, the optimal parameter data is solved with the minimum deviation from the preset target curve as the objective constraint.

[0025] According to another aspect of the present invention, an AI-based design optimization method is provided, the method being applied to an AI module based on an optimization system as described in the first aspect of the present invention, comprising the steps of:

[0026] Receive a design task, break down the design task, and extract the data information from the design task;

[0027] An analysis command is sent to the processing and analysis module so that the processing and analysis module can analyze and optimize the data information of the design task based on the analysis command;

[0028] A build command is sent to the DOE module, so that the DOE module calls the optimization parameter data based on the build command and builds the DOE based on the optimization parameter data;

[0029] Send a call command to the EDA module so that the EDA module can perform simulation based on the call command and DOE data;

[0030] An optimization command is sent to the optimization module so that the optimization module can model and solve for the optimal parameter data based on the optimization command, optimization parameter data, and simulation result data.

[0031] Furthermore, it also includes the following steps:

[0032] Send a call command to the EDA module so that the EDA module can perform simulation based on the call command and optimal parameter data;

[0033] The simulation results are verified based on a pre-set AI model.

[0034] The simulation results are verified based on the preset AI model, including the following steps:

[0035] If the simulation results meet the preset design objectives, a design report is generated;

[0036] If the simulation results do not meet the preset design goals, the optimization will be re-executed.

[0037] The optimization module models and solves for the optimal parameter data based on the optimization commands, optimization parameter data, and simulation results data from the AI ​​module, including the following steps:

[0038] The solution domain is obtained by fitting the optimized parameter data and the simulation results data to the data relationship.

[0039] In the solution domain, the optimal parameter data is solved with the objective constraint of minimizing the deviation from the preset target curve.

[0040] In summary, this invention provides an AI-based design optimization system and method. The system includes an AI module, a processing and analysis module, a Design of Engineering (DOE) module, an Electronic Design Automation (EDA) module, and an optimization module. The AI ​​module is interconnected with the processing and analysis module, the DOE module, the optimization module, and the EDA module, and is used to receive design tasks, decompose and allocate these tasks, and plan the process. It also optimizes the process and parameters of the design tasks based on a preset AI model. The processing and analysis module analyzes and optimizes the data information of the design tasks based on the analysis commands from the AI ​​module. The DOE module calls the optimized parameter data based on the construction commands from the AI ​​module and constructs a Design of Engineering (DOE) based on the optimized parameter data. The EDA module performs simulation based on the calling commands from the AI ​​module and the DOE data. The optimization module models and solves for optimal parameter data based on the optimization commands from the AI ​​module, the optimized parameter data, and the simulation results. The technical solution provided by this invention integrates AI tools, represented by large language models, into the parameter optimization process. By utilizing the built-in experience and expertise of the large language AI model, it achieves fully autonomous decision-making and control over all aspects of the optimization process, including parameter control, optimization control, experimental design, data analysis, and application. This replaces the aspects in existing technologies that require human decision-making, significantly reducing the consumption and dependence on human mental labor, and realizing a truly unmanned optimization process from "task input" to "task output". Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the overall structure of the AI-based design optimization system provided in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the function curve fitted to the YX relationship;

[0043] Figure 3 This is a schematic diagram illustrating the solution based on the solution space and constraints of ModelY;

[0044] Figure 4 This is a schematic diagram illustrating the final optimized effect using the embodiments of the present invention;

[0045] Figure 5 This is a flowchart of the AI-based design optimization method provided in the embodiments of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0047] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0048] Currently, the R&D and production models across various industries heavily rely on traditional, manually-driven processes. From experimental design to data analysis, the entire chain requires significant human intervention, especially in Electronic Design Automation (EDA) applications. For example, device simulation, SPICE parameter extraction, and circuit design optimization necessitate numerous simulation experiments to find optimal design parameters. In the parameter optimization process of the BSIM-CMG model for SPICE parameter extraction, the BSIM-CMG device model has at least 150 parameters. Even for simpler models like BSIM3 and BSIM4, there are at least 50 parameters. During a single model fitting and parameter extraction process, the number of selectable and optimizeable parameters is enormous, offering a vast optimization space. Typically, the optimization of a device model requires optimizing 10-30 relevant parameters to ensure that the simulated characteristic curves closely match the real-world curves, achieving model calibration. The task description in this example is as follows: Optimize BSIM-CMG model parameters. Provide the netlist and model parameters. Use an EDA tool to modify the model parameters to improve the fitting accuracy between the IdVg curve and the real-world curve. The goal is to minimize the fitting error to less than 0.01. Keep the parameters within a reasonable range. During parameter parsing, extract relevant parameters as much as possible to avoid affecting the optimization effect later due to missing parameters. The maximum number of iterations during optimization is 10, and the maximum number of experiments per round is 15. As can be seen from the above task description, the optimized parameters include almost all of the BSIM-CMG parameters. However, many parameters are merely functional switches unrelated to performance optimization, yet they are all submitted to the tool indiscriminately by the user. It is particularly noteworthy that the user did not differentiate, distinguish, or provide instructions regarding the optimized parameters. Furthermore, the user did not indicate the range of the parameters. In existing actual parameter tuning processes, it is necessary to have a very clear understanding of the physical meaning of each parameter. Furthermore, since full-factor experiments are extremely time-consuming, it requires extensive experience in selecting key parameters from a large pool of data for Design of Experiments (DOE). This necessitates first identifying the influence patterns of the parameter (group) and then fine-tuning the local parameters. This raises several critical issues: 1. Simulation professionals need a thorough understanding of the physical mechanisms; 2. They need experience in selecting key parameters; 3. A large number of experiments are required; 4. It requires repeated local optimization. Improper handling of any of these factors can easily lead to the failure of the target task. Moreover, the entire process requires repeated switching between different scenarios, posing a significant challenge to the task management and control of the professionals involved.Embodiments of the present invention provide an AI-based design optimization system and method. Utilizing the built-in understanding capabilities of a large language AI model, it interprets all user input, automatically summarizes and generates key steps for task execution, and, with the help of the large language model's internal experience and understanding, filters optimization parameters, eliminating those irrelevant or weakly relevant to the task. Based on the optimization progress and situation, it dynamically adjusts optimization parameters (groups) and dynamically sets the parameter optimization space, thereby overcoming the drawbacks of relying primarily on human experience (processes, analysis) and greatly improving the efficiency of research and development design.

[0049] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. An embodiment of the present invention provides an AI-based design optimization system. Figure 1 The diagram shows the overall structure of the AI-based design optimization system provided in an embodiment of the present invention. Figure 1 As shown, the AI-based design optimization system of this invention includes an AI module, a processing and analysis module, a DOE module, an EDA module, and an optimization module.

[0050] The AI ​​module, interconnected with the processing and analysis module, DOE module, optimization module, and EDA invocation module, receives design tasks, breaks them down and allocates them, and optimizes the process and parameters of the design tasks based on a pre-set AI model. The AI ​​module primarily receives design tasks, confirms task objectives based on the design task, breaks down and allocates the task, plans the design process, monitors the entire process, and reviews the results. The pre-set AI model can be trained using historical data using specific models; no specific limitations are specified here. Design tasks include, for example, chip design tasks and electronic circuit design tasks. After receiving the design task, the AI ​​module uses the analytical and reasoning capabilities of the large model to analyze the current task information, breaks down the main task into multiple sub-tasks, plans the execution order of these sub-tasks, and identifies the corresponding modules for each sub-task. Simultaneously, it generates a Flow.Json information file (log) for current task information descriptions, module calls, and module flow connections.

[0051] The processing and analysis module is used to analyze and optimize the data information of the design task based on the analysis commands of the AI ​​module. The processing and analysis module analyzes the background information and data of the design task, providing optimized parameter information and theoretical support.

[0052] The DOE module is used to call construction commands from the AI ​​module to optimize parameter data, and then construct a DOE based on that optimized parameter data. Internally, the DOE module can use DOE methods for large language models, or DOE algorithms based on orthogonal experiments, randomized experiments, Latin squares, hyper-Latin squares, etc. The DOE module is used for DOE design and analysis based on optimized parameter data.

[0053] The EDA module is used to perform simulations based on commands from the AI ​​module and DOE data, and to generate simulation results data, serving as a "verification" and "confirmation" tool for the design. The EDA module is primarily used as a communication window to connect to and invoke other applications, including but not limited to operations such as running, writing, reading, and logging data from third-party software.

[0054] The optimization module is used to model and solve for optimal parameter data based on the optimization commands, optimization parameter data, and simulation results data from the AI ​​module. The optimization module is used for analyzing, modeling, optimizing, and solving for optimization parameter data. The optimization module embeds optimization algorithms, including but not limited to heuristic algorithms, gradient descent, Bayesian algorithms, genetic algorithms, ant colony algorithms, machine learning methods, and neural networks. It models and solves for optimal parameter data based on optimization parameter data, simulation results data, and a preset target curve. The modeling and solving for optimal parameter data can be performed according to the following steps:

[0055] The solution domain is obtained by fitting the optimized parameter data and simulation results data to the data relationship.

[0056] The optimal parameter data is solved in the solution domain with the minimum deviation from the preset target curve as the objective constraint.

[0057] As is well known, the current development of AI demonstrates its incredibly powerful functions and potential. However, probabilistic text generation is currently unable to quantitatively solve complex mathematical relationships. Its so-called "illusion" further highlights its limitations in large-scale and complex engineering applications. AI can accurately describe the influence trend of each parameter, but how each parameter specifically affects the other (for example, as x increases, y also increases, but the amount by which x increases corresponding to the increase in y is unknown) and the interactive effects of multiple parameters are all unknown. For example, Table 1 shows a relationship table fitted using AI to the YX relationship. Figure 2 The diagram shows a schematic of the function curve fitted to the YX relationship.

[0058] Table 1

[0059] Parameter X <![CDATA[X1]]> <![CDATA[X2]]> <![CDATA[X3]]> <![CDATA[X n ]]> <![CDATA[Result Y1]]> ↑ ↑ ↓ ↑ <![CDATA[Result Y m > ↑ ↓ ↓ ↓

[0060] Combining Table 1 and Figure 2 As shown, AI knows that X1 is a gain with respect to Y1, and AI knows that within this parameter range, Y1 increases as X1 increases. However, from a strictly mathematical perspective, AI does not know if it is according to... Figure 2The increase of the red dashed line is not determined by whether it increases according to the increase of the black dashed line, so in practical engineering applications, AI cannot provide an accurate solution. In this process, AI gets stuck in a local loop. The above example simplifies the problem. In real-world applications, there are multiple parameters interacting and conflicting ranges, leading to unsolvable problems. AI is often even less able to determine the exact solution. In this embodiment of the invention, a "solution boundary" (i.e., a preset target curve) is provided for AI to solve the problem. More specifically, AI knows whether the effect of each parameter on the result is gain or suppression, but it cannot know the rate of change of gain and suppression. Therefore, in practical applications, it will get stuck in a loop within the local range of parameters until the loop is completed without obtaining the optimal solution. The "solution boundary" (i.e., the preset target curve) provided by this embodiment of the invention can help AI clarify the rate of change and comprehensive relationship in mathematical relationships, thereby making up for the current shortcomings of AI. The preset target curve in this embodiment of the invention includes, but is not limited to, obtaining it through the following methods: actual engineering test curves (e.g., IdVg), target curves that need to be achieved in the actual design process, curves in literature, etc. Expanding on this preset target curve can also be used as a feature extraction feature of the curve, for example, the threshold voltage V. th Turn-on current I on Turn-off current I off Subthreshold swing (SS), breakdown voltage (BV), etc. Specific examples are provided below:

[0061] The optimization module targets the optimization parameter data (i.e., DOE parameters X1, X2, X3, X4, X5... X...). n ) and EDA simulation results data (Y1, Y2, Y3...Y m By fitting the data relationships, the following mathematical relationships were obtained:

[0062]

[0063] The multi-parameter, multi-objective model, Model Y, represents the solution boundary conditions described above. The constraints are as follows:

[0064] Min(ModelY-TargetY)

[0065] That is, the constraint condition is that the deviation between ModelY and the preset target curve TargetY is minimized. Figure 3 The diagram above illustrates the solution process, as shown below. Figure 3As shown, based on the solution space and constraints of ModelY, the closer the green curve is to the target red curve, the better. Furthermore, if we follow specific circuits, devices, or other specialized fields, as the number of training iterations increases, we are generating a physical and mathematical AI model specific to that field. This model will be more suitable for optimizing and solving specific problems, making the AI ​​more professional and accurate, improving the efficiency of the optimization process, and avoiding the AI ​​falling into a vicious cycle of repetitive optimization.

[0066] According to some optional embodiments, the EDA module is further used to perform simulation operations based on the invocation commands and optimal parameter data of the AI ​​module; the AI ​​module is further used to verify the simulation results based on a preset AI model. The verification of the simulation results by the AI ​​module based on the preset AI model includes, for example, the following steps:

[0067] If the simulation results meet the preset design objectives, a design report will be generated;

[0068] If the simulation results do not meet the preset design goals, the optimization process is re-executed. Re-execution includes any one of the following: returning to the processing and analysis module to analyze and optimize the data information of the design task; returning to the DOE module to construct the DOE; or returning to the optimization module to model and solve for the optimal parameter data. The starting point and process of re-execution can be flexibly adjusted and adapted according to the actual data situation and task information. For example, if the optimal solution can be found during the processing stage of the DOE module, the process can be stopped manually. Alternatively, if a large amount of experimental data is available and DOE is not required, modeling and solving for the optimal parameter data can be performed directly through the optimization module. Other optional processes can also be selected according to actual needs, which will not be elaborated here.

[0069] According to certain optional embodiments, the system may also include a data module, such as Figure 1As shown, the data module is interconnected with the AI ​​module, processing and analysis module, DOE module, EDA module, and optimization module. The AI ​​module also writes design task data into the data module; the data module receives the results data generated by each module and provides the necessary data to each module. The data module acts as a data transfer intermediary between modules throughout the system, including but not limited to reading, writing, calling, tagging, and logging. The configuration of the data module ensures the stability of the optimization process and greatly guarantees the accuracy of data transmission and storage during optimization. The data calls and transfers between various modules of the system are as follows: The AI ​​module calls the processing and analysis module, analyzes and provides the parameters that need optimization, and writes the optimized parameter data (parameter.Json) into the data module; the AI ​​module calls the DOE module and passes in the optimized parameter data (parameter.Json), generates a new DOE, and writes the newly generated DOE data into the data module (DOE.Json); the AI ​​module calls the EDA module, writes the DOE data (DOE.Json) into the EDA module and starts the simulation run, and writes the simulation result data (Optimizer.Json) from the completed EDA module into the data module; the AI ​​module calls the optimization module and the data module... The optimization parameter data and simulation result data (Optimizer.Json) in the block begin algorithm modeling and solve for the optimal parameters (verification.Json) based on the task objective information. The algorithm's running information and parameter information are then written into the data module. The algorithm can be a traditional algorithm, a heuristic algorithm, or a genetic, machine learning, neural network, etc., and can be used for modeling and analysis of multiple parameters and multiple objectives. The AI ​​module calls the EDA module to write the newly generated optimal parameter data (verification.Json) into the EDA module and starts the simulation. The AI ​​module reviews whether the results meet the objectives. If they do, the process ends and a report is generated; otherwise, a new loop begins.

[0070] According to certain optional embodiments, the processing and analysis module, DOE module, optimization module, and data module in this system can all synchronously embed sub-AI modules. Each module is matched with existing business processes, forming a human-like project management work mode. The embedded sub-AI modules interact with other modules of the system and also with the outside world. They are used for analyzing and processing the tasks of the current module and for further planning of the current tasks. They can also exchange information with the AI ​​module. In the embodiments of this invention, the AI ​​(including the AI ​​module and sub-AI modules) can explicitly present process results and interact with humans, allowing for human judgment and intervention.

[0071] According to certain optional embodiments, in the collaborative work of multiple AIs, besides the AI ​​module, at least one of the other modules uses a second AI to handle phased tasks and works collaboratively with the AI ​​module. The module's core functionality is implemented by programming code, and the AI's information transmission and interaction are executed in the code through prompts that form parameters.

[0072] According to some optional embodiments, the system may also include other modules, such as a user interaction module, a graphical interface module, a log management module, and a result storage module, to implement conventional functions such as user interaction, process and result display.

[0073] Current large language AI models often suffer from "illusion" and generation errors, leading to inconsistent accuracy in simple applications. This invention addresses these issues by combining large language AI models with EDA algorithms, using EDA tools or algorithms to verify the generated content, thereby improving accuracy. Specifically, the EDA module connects to third-party simulation tools for verification, resolving issues like illusion and unreliability in large language models. The optimization module addresses the limitations of large language AI models in computation and data analysis by introducing optimization algorithms. The DOE module addresses the weaknesses of large oracle models in experimental planning and computation. The data module implements database functionality and a multi-AI collaborative working mechanism, enabling multi-level data processing and temporary storage, thus resolving the limitation of large language model prompts. Finally, the AI ​​module handles task understanding and decomposition, parameter selection, parameter optimization range determination, optimization result analysis, and optimization process control, eliminating reliance on human mental labor in current optimization methods and achieving fully automated operation.

[0074] Regarding the example provided above (parameter optimization of BSIM-CMG), the system provided in this embodiment of the invention is used. The AI ​​module receives the task, formulates the optimization process and strategy according to the task content, can call the database to read historical data, and can also call the EDA module to conduct specific experimental calculations. It can also perform algorithm modeling based on the results and parameter information to assist in finding the optimal solution. All tool calls and optimization processes are completed by the AI ​​module and its collaborating AIs. First, according to the task requirements, the AI ​​eliminates irrelevant parameters (including switch setting parameters, setting parameters unrelated to performance, and parameters with weak relation to the optimization objective), and selects 15 parameters as optimization targets from all parameters in the model. The principles for parameter selection include: key parameters that directly affect the Ids-Vgs curve; parameters that play a dominant role in the model; and parameters that have a significant impact on threshold voltage, mobility, saturation current, etc., based on literature and engineering experience. In the first round of optimization, the AI, based on experience, selects 5 key parameters as optimization targets for this round, serving as the parameters for the first round of experiments, while other parameters are executed according to default values. For example, the parameters selected for the first round of experiments are: RDSW (source-drain current parameter), VSAT (saturation velocity), U0 (low-field mobility), ETA0 (DIBL coefficient), and DVT0 (short-channel effect coefficient), with the optimization range for each parameter given. After these parameters and ranges are passed into the DOE module, the DOE module's algorithm designs a certain number of experiments. These determined experimental parameters are written to the EDA module via API calls and executed. After a set of experiments is run, the results are extracted and displayed, and the fitting error (or other performance indicators related to the task objective) is output. Alternatively, it can be submitted to a data analysis AI to summarize task-related evaluative text. After each round of DOE module execution, the optimization module is called to perform algorithmic modeling on the running data and find the optimal solution through the algorithm. If the target is not reached, iterative optimization is performed according to the task requirements. The above information and key information of the entire running process can be displayed in real time in the progress window of the graphical interface module. When the final optimization effect is achieved, an overall project report is output. The final optimization effect of this embodiment is shown in [link to example]. Figure 4 The vertical axis represents the drain current Id, and the horizontal axis represents the gate voltage Vg. The blue dots represent experimental data points, and the red curves represent the curves obtained from EDA module simulations. Figure 4 As shown, the error between the curve obtained from the Ids-Vgs simulation and the curve at the experimental data points is 0.07627, proving the feasibility of the technical solution provided by this invention. The aforementioned data information is simultaneously written into the data module, which can serve as the database for the entire process.

[0075] Embodiments of the present invention also provide an AI-based design optimization method, which is based on the AI-based design optimization system provided in the above embodiments and is applied to an AI module. Figure 5 The flowchart of the optimization method of this embodiment of the present invention is shown in the figure. Figure 5 As shown, the method includes the following steps:

[0076] S202. Receive the design task, break down the design task, and extract the data information from the design task.

[0077] S204. Send an analysis command to the processing and analysis module so that the processing and analysis module can analyze and optimize the data information of the design task based on the analysis command.

[0078] S206. Send a build command to the DOE module so that the DOE module can call the optimization parameter data based on the build command and build the DOE based on the optimization parameter data.

[0079] S208. Send a call command to the EDA module so that the EDA module can perform simulation based on the call command and DOE data.

[0080] S210. Send an optimization command to the optimization module, so that the optimization module can model and solve for the optimal parameter data based on the optimization command, optimization parameter data, and simulation result data. Modeling and solving for the optimal parameter data includes the following steps:

[0081] S2101. Fit the data relationship between the optimized parameter data and the simulation result data to obtain the solution domain.

[0082] S2102. Solve for the optimal parameter data in the solution domain with the minimum deviation from the preset target curve as the objective constraint.

[0083] According to some optional embodiments, the method further includes the step of:

[0084] S212. Send a call command to the EDA module so that the EDA module can perform simulation based on the call command and optimal parameter data.

[0085] S214. Verify the simulation results based on a preset AI model. This verification may include, for example, the following steps:

[0086] S2141. If the simulation results meet the preset design objectives, a design report will be generated.

[0087] S2142. If the simulation results do not meet the preset design goals, then the optimization will be re-executed.

[0088] The above optimization steps are generalized steps, and the actual execution can be dynamically adjusted and executed according to the current task information. For example, the task execution may include, but is not limited to: executing steps S202 to S208 for a simple DOE module plus EDA module simulation task; directly inputting existing EDA result data and parameter information for modeling, skipping the step of calling the EDA module in S208, and finally selecting the simulation tool to call the EDA module or manually inputting the EDA module based on the optimal parameter data in step S2102 to verify the results.

[0089] The specific implementation of each step in the AI-based design optimization method of this embodiment of the present invention is the same as the process of each mode of the AI-based design optimization system of the above embodiments of the present invention realizing its function, and its repeated description will be omitted here.

[0090] In summary, this invention relates to an AI-based design optimization system and method. The system includes an AI module, a processing and analysis module, a Design of Engineering (DOE) module, an Electronic Design Automation (EDA) module, and an optimization module. The AI ​​module is interconnected with the processing and analysis module, the DOE module, the optimization module, and the EDA module, and is used to receive design tasks, decompose and allocate design tasks, and optimize the process and parameters of the design tasks based on a preset AI model. The processing and analysis module is used to analyze and optimize the data information of the design tasks based on the analysis commands of the AI ​​module. The DOE module is used to call the optimized parameter data based on the construction commands of the AI ​​module and construct a Design of Engineering (DOE) based on the optimized parameter data. The EDA module is used to perform simulation operation based on the call commands of the AI ​​module and the DOE data. The optimization module is used to model and solve for the optimal parameter data based on the optimization commands of the AI ​​module, the optimized parameter data, and the simulation operation result data. This embodiment fully utilizes the excellent analytical and reasoning capabilities of AI to provide key parameters free from human experience interference, greatly compensating for the deficiencies of insufficient human experience. It can flexibly integrate and plan, run, and optimize different processes, third-party tools, and task objectives. Whether from the perspective of process or data analysis optimization, it has changed the drawbacks of relying mainly on human experience (process, analysis), greatly improved the efficiency of R&D design, and saved R&D costs.

[0091] It should be understood that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of this invention, technical features of the above embodiments or different embodiments can also be combined, steps can be implemented in any order, and many other variations exist regarding different aspects of one or more embodiments of the invention as described above, which are not provided in the details for the sake of brevity. The specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. An AI-based design optimization system, characterized in that, The system includes an AI module, a processing and analysis module, a DOE module, an EDA module, and an optimization module; The AI ​​module is interconnected with the processing and analysis module, DOE module, optimization module and EDA calling module, and is used to receive design tasks, decompose and allocate design tasks and plan processes, and optimize the process and parameters of design tasks based on a preset AI model. The processing and analysis module is used to analyze and optimize the data information of the design task based on the analysis commands of the AI ​​module; The DOE module is used to invoke optimization parameter data based on the construction command of the AI ​​module, and to construct the DOE based on the optimization parameter data; The EDA module is used to perform simulation based on the call commands from the AI ​​module and DOE data. The optimization module is used to model and solve for the optimal parameter data based on the optimization commands, optimization parameter data, and simulation result data from the AI ​​module.

2. The system according to claim 1, characterized in that, The system also includes a data module, which is interconnected with the AI ​​module, processing and analysis module, DOE module, EDA module and optimization module; The AI ​​module is used to write the design task data information into the data module; The data module is used to receive the result data generated by each module and to provide each module with the data required for operation.

3. The system according to claim 1, characterized in that, The EDA module is also used to perform simulation operations based on the call commands and optimal parameter data of the AI ​​module; The AI ​​module is also used to verify the simulation results based on the preset AI model.

4. The system according to any one of claims 1-3, characterized in that, The AI ​​module verifies the simulation results based on the preset AI model, including: If the simulation results meet the preset design objectives, a design report is generated; If the simulation results do not meet the preset design goals, the optimization will be re-executed.

5. The system according to claim 4, characterized in that, The re-execution of optimization includes any one of the following: returning to the processing and analysis module to analyze and optimize the data information of the design task, returning to the DOE module to construct the DOE, and returning to the optimization module to model and solve for the optimal parameter data.

6. The system according to claim 5, characterized in that, The optimization module models and solves for the optimal parameter data based on the optimization commands, optimization parameter data, and simulation results data from the AI ​​module, including: The solution domain is obtained by fitting the optimized parameter data and the simulation results data to the data relationship. In the solution domain, the optimal parameter data is solved with the objective constraint of minimizing the deviation from the preset target curve.

7. An AI-based design optimization method, characterized in that, The method, based on the optimization system as described in any one of claims 1-6, is applied to an AI module and includes the following steps: Receive a design task, break down the design task, and extract the data information from the design task; An analysis command is sent to the processing and analysis module so that the processing and analysis module can analyze and optimize the data information of the design task based on the analysis command; A build command is sent to the DOE module, so that the DOE module calls the optimization parameter data based on the build command and builds the DOE based on the optimization parameter data; Send a call command to the EDA module so that the EDA module can perform simulation based on the call command and DOE data; An optimization command is sent to the optimization module so that the optimization module can model and solve for the optimal parameter data based on the optimization command, optimization parameter data, and simulation result data.

8. The method according to claim 7, characterized in that, It also includes the following steps: Send a call command to the EDA module so that the EDA module can perform simulation based on the call command and optimal parameter data; The simulation results are verified based on a pre-set AI model.

9. The method according to claim 8, characterized in that, The simulation results are verified based on the preset AI model, including the following steps: If the simulation results meet the preset design objectives, a design report is generated; If the simulation results do not meet the preset design goals, the optimization will be re-executed.

10. The method according to claim 9, characterized in that, The optimization module models and solves for the optimal parameter data based on the optimization commands, optimization parameter data, and simulation results data from the AI ​​module, including the following steps: The solution domain is obtained by fitting the optimized parameter data and the simulation results data to the data relationship. In the solution domain, the optimal parameter data is solved with the objective constraint of minimizing the deviation from the preset target curve.

Images