Intelligent auxiliary design optimization system and method for military power module

Abstract

This invention provides an intelligent assisted design optimization system and method for military power modules. The system includes an intelligent design and manufacturing database for storing production line big data of packaged power modules. This production line big data includes successful design data that meets design requirements, manufacturing processes and their corresponding measured performance data, as well as negative sample data. A design module, trained based on the successful design data in the intelligent design and manufacturing database, receives target electrical performance parameters and outputs preliminary circuit design parameters and initial manufacturing process parameters. A collaborative optimization module receives the circuit design parameters and initial manufacturing process parameters and outputs predicted electrical performance parameters. The output of the design module is connected to the input of the collaborative optimization module to form a design and verification closed loop. This invention solves the technical problems of long design cycles, high verification costs, and difficulty in collaborative optimization of design and manufacturing in the development of military packaged power modules.

Description

Technical Field

[0001] This invention relates to the technical fields of artificial intelligence and power module design, and in particular to an intelligent assisted design optimization system and method for military power modules. Background Technology

[0002] In critical fields such as military ships and aerospace, packaged power modules are the core units for realizing the functions of electronic systems. Their miniaturization and modularization have long been core research and development directions in this field. Due to international technology embargoes on modules, the complete domestic substitution of packaged power modules and their component modules has become an urgent strategic need. To cope with the large number of R&D tasks related to the design, manufacturing, and verification of power modules under the background of complete domestic substitution, improving the R&D iteration efficiency of military-grade packaged power modules by relying on artificial intelligence (AI) technology has become an important research direction in this field.

[0003] Compared to traditional on-board power systems, military-grade in-package power module systems offer advantages such as high integration and strong power stability. Compared to on-chip power systems, military-grade in-package power module systems feature a wider voltage range, stronger current drive, and a circuit architecture that can be flexibly adjusted according to different application scenarios. Currently, the design and development of military-grade in-package power modules faces the following technical challenges: First, it places stringent requirements on the accuracy and adaptability of component selection, component and process optimization, key electrical virtual testing, and overall circuit architecture design and optimization. Second, the mainstream design scheme for in-package power modules still employs a manual simulation design process. Designers must manually perform circuit topology design, component selection, and parameter adjustment, and verify the design effect through multiple costly and time-consuming manufacturing tape-outs, resulting in long design cycles, high verification costs, and slow technological updates.

[0004] Therefore, there is an urgent need for an intelligent assisted design optimization system and method for military power modules that can solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent auxiliary design optimization system and method for military power modules, which solves the technical problems of long design cycle, high verification cost, and difficulty in coordinating design and manufacturing optimization in the research and development of military packaged power modules.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides an intelligent assisted design optimization system for military power modules, comprising: an intelligent design and manufacturing database for structured storage of production line big data of power modules within packages, the production line big data including successful design data that meets design requirements, manufacturing processes and their corresponding measured performance data, and negative sample data containing design defects or process defects; a design module, trained based on the successful design data in the intelligent design and manufacturing database, for receiving target electrical performance parameters and outputting preliminary circuit design parameters and corresponding initial manufacturing process parameters; and a collaborative optimization module, comprising a generative adversarial network submodule and a genetic algorithm optimization submodule, trained based on the intelligent design and manufacturing database, for receiving the preliminary circuit design parameters and corresponding initial manufacturing process parameters and outputting predicted electrical performance parameters; wherein the output of the design module is connected to the input of the collaborative optimization module to form a design and verification closed loop.

[0007] Furthermore, the generative adversarial network submodule is used to simulate the impact of manufacturing processes on circuit performance; the genetic algorithm optimization submodule is used to collaboratively optimize the network parameters of the generative adversarial network submodule and the input manufacturing process parameters.

[0008] Furthermore, the generative adversarial network submodule includes a generator and a discriminator; the generator generates simulated performance parameters based on input circuit design parameters and process parameters; the discriminator is used to determine the authenticity of the simulated performance parameters and the measured performance parameters in the intelligent design and manufacturing database; wherein, the genetic algorithm optimization submodule adjusts the network structure parameters of the generator and discriminator through iterative optimization, and simultaneously optimizes the input process parameters to minimize the difference between the simulated performance parameters and the target performance parameters.

[0009] Furthermore, the design module is constructed using a deep convolutional neural network to realize the mapping from the performance parameter space to the circuit design parameter space; the basic prediction framework of the collaborative optimization module is constructed using a deep graph neural network to model the topological relationship between circuit elements, process parameters and electrical performance.

[0010] Furthermore, the system also includes a data preprocessing module, which performs multi-level normalization preprocessing on the data in the intelligent design and manufacturing database; the preprocessing performs standardization and equalization processing on the data distribution of different component parameters, process parameters and performance parameters respectively.

[0011] Furthermore, the system also includes an iterative control module, used to: compare the predicted electrical performance parameters output by the collaborative optimization module with the target electrical performance parameters; when the difference between the two exceeds a preset threshold, generate a feedback signal containing performance difference information and feed it back to the design module and the collaborative optimization module to start a new round of iterative optimization; when the difference between the two is less than or equal to the preset threshold, output the current circuit design parameters and the optimized manufacturing process parameters as the final solution.

[0012] Furthermore, the system also includes a prototype verification module, which is used to compare the performance of the final circuit design generated by the iterative control module with a benchmark military power module product that did not participate in the database construction and training.

[0013] Furthermore, both the target electrical performance parameters and the predicted electrical performance parameters include specific reliability indicators of the power module, which include at least one of the following: the drift rate of the output voltage at extreme temperatures and the stability parameter under mechanical stress.

[0014] This invention also provides an intelligent assisted design optimization method for military power modules, comprising the following steps: S1. Constructing and maintaining an intelligent design and manufacturing database, and performing multi-level normalization preprocessing on production line big data; S2. Based on the preprocessed data, training and obtaining a design module and a collaborative optimization module respectively; S3. Receiving target electrical performance parameters as input, and generating initial circuit design parameters and corresponding initial manufacturing process parameters through the design module; S4. Inputting the initial circuit design parameters and initial manufacturing process parameters into the collaborative optimization module, and outputting optimized manufacturing process parameters and corresponding predicted electrical performance parameters through the synergistic effect of its internal generative adversarial network submodule and genetic algorithm optimization submodule; S5. Comparing the predicted electrical performance parameters with the target electrical performance parameters; if the difference is greater than a preset threshold, generating a feedback signal containing performance difference information and feeding it back to step S3 for iterative optimization; if the difference is less than or equal to the preset threshold, outputting the final circuit design parameters and collaboratively optimized manufacturing process parameters that meet the requirements.

[0015] Furthermore, the synergistic effect of the Generative Adversarial Network (GAN) submodule and the Genetic Algorithm (GA) optimization submodule includes: the generator of the GAN submodule generates simulated performance based on the current circuit design parameters and process parameters; the discriminator of the GAN submodule evaluates the fidelity of the simulated performance; and the GA optimization submodule, based on the evaluation results and the difference between the simulated performance and the target performance, optimizes the simulation performance to improve its fidelity and approximate the target performance, while simultaneously encoding, selecting, crossing over, and mutating the network structure parameters of the generator and discriminator, as well as the input process parameters, to generate the next generation of optimized parameters.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: This invention employs an intelligent design and manufacturing database containing negative samples, enabling the trained design module to avoid common design flaws and improve the success rate of initial design schemes. Combined with a design module based on deep convolutional neural networks, it achieves automated mapping from military target electrical performance parameters to circuit design parameters and manufacturing process parameters, replacing the traditional manual simulation and repeated trial and error that relies on engineers' experience. This shortens the initial design cycle of power modules within military packages, while structured big data support makes the design scheme more in line with the stringent design requirements of the military field, improving the accuracy of the initial design.

[0017] This invention introduces a collaborative optimization module integrating Generative Adversarial Networks (GANs) and Genetic Algorithms. This module can perform high-fidelity virtual performance prediction and process co-optimization on the solutions output by the design module, forming a rapid closed loop of "design-virtual verification". It replaces the traditional expensive and time-consuming physical chip fabrication verification with low-cost and high-efficiency digital simulation, enabling the design to be rapidly iterated and optimized multiple times in the digital domain, reducing the cost of a single verification and accelerating the overall optimization process. The Genetic Algorithm achieves global optimization of network parameters and process parameters through iterative optimization, solving the stability problem of simple algorithm application, reducing verification costs, and improving the process adaptability and manufacturability of the design solution.

[0018] This invention, through the establishment of an iterative control module and a prototype verification module, combined with a full-process intelligent assisted design optimization method, achieves full-process automation and standardization of the design, verification, and optimization of military-grade packaged power modules. It forms a self-driven iterative optimization system that can complete the iterative adjustment of design schemes without manual intervention, thereby improving the R&D iteration speed of military-grade packaged power modules. It can efficiently respond to a large number of R&D tasks related to the design, manufacturing, and verification of power modules under the background of domestic substitution, and is suitable for the R&D direction and strategic needs of miniaturization and modularization of packaged power modules in the military shipbuilding and aerospace fields. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the intelligent assisted design optimization system provided in this embodiment; Figure 2A flowchart of the intelligent assisted design optimization method provided in this embodiment; Figure 3 This is a schematic diagram of the structure of the intelligent design and manufacturing database in the intelligent assisted design optimization system provided in this embodiment; Figure 4 This is a technical roadmap for the collaborative optimization module in the intelligent assisted design optimization system provided in this embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0023] This embodiment provides an intelligent assisted design optimization system for military power modules, such as... Figure 1 As shown, the system includes: an intelligent design and manufacturing database for structured storage of production line big data of power modules within the package. This production line big data includes successful design data that meets design requirements, manufacturing processes and their corresponding measured performance data, as well as negative sample data containing design or process defects, achieving comprehensive structured management of production line data; a design module, trained based on successful design data in the intelligent design and manufacturing database, for receiving externally input target electrical performance parameters, and outputting preliminary circuit design parameters and corresponding initial manufacturing process parameters after module computation, achieving direct mapping from performance requirements to design / process parameters; and a collaborative optimization module, including a Generative Adversarial Network (GAN) submodule and a genetic algorithm optimization submodule, trained based on the intelligent design and manufacturing database, for receiving the circuit design parameters and corresponding initial manufacturing process parameters, and outputting predicted electrical performance parameters. The output of the design module is connected to the input of the collaborative optimization module to form a design and verification closed loop.

[0024] Specifically, based on the big data accumulated from long-term R&D and production of internal power modules, an intelligent design and manufacturing database will be established.

[0025] This solution constructs a full-link architecture of "data support - automated design - collaborative verification" for military power modules, breaking through the technical pain points of scattered data without negative sample support and the disconnect between design and manufacturing processes in traditional military power module design. By integrating positive and negative samples through a structured database, it provides comprehensive data support for module training. The module design combining GAN and genetic algorithm is adapted to the core requirements of collaborative design and manufacturing of military power modules. The closed-loop design and verification replaces the traditional chip fabrication trial and error mode, shortening the design cycle of military power modules and solving the technical problems of long design cycle and high verification cost of traditional design methods.

[0026] In this embodiment, the core sub-modules of the collaborative optimization module are defined as follows: the GAN sub-module is responsible for establishing a complex "process-performance" simulation module, which can simulate the impact of subtle changes in the manufacturing process on the final electrical characteristics with high fidelity; the genetic algorithm optimization sub-module is used to collaboratively optimize the network parameters (such as the number of layers and nodes) of the GAN sub-module and the input manufacturing process parameters.

[0027] like Figure 4 As shown in the diagram, the technical roadmap of the co-optimization module in the intelligent assisted design optimization system provided in this embodiment includes a Design and Technology Co-Optimization (DTCO) artificial intelligence module. This module utilizes GAN (Generative Artificial Intelligence) to simulate the impact of electronic component manufacturing processes on the performance of the power module system, performing co-optimization of the power module's design and manufacturing processes to maximize its performance. GAN can learn the complex relationship between semiconductor process parameter combinations and power module performance based on a large amount of historical data, providing a data-driven method for optimizing the combination of electronic component manufacturing process parameters within the power module system. Simultaneously, the GAN tool can simulate various possible combinations of process parameters, providing a rich pool of candidate solutions for the design of the power module and the co-optimization (DTCO) of related semiconductor module manufacturing processes.

[0028] In this scheme, the GAN submodule is responsible for accurate simulation, while the genetic algorithm is responsible for efficient optimization. This solves the problems of low accuracy, high parameter tuning difficulty, and single optimization dimension in military power supply process-performance simulation optimization using a single algorithm. The GAN submodule can accurately capture the complex nonlinear relationship between manufacturing process and circuit performance, improving the realism of simulation results. The genetic algorithm makes up for the shortcomings of GAN manual parameter tuning, such as low efficiency and blind parameter optimization. At the same time, it achieves dual optimization of module network parameters and manufacturing process parameters, so that the optimized process parameters are compatible with the module's simulation logic, improving the accuracy of process simulation and the actual adaptability of process parameters, and ensuring that the performance of the design scheme can be realized through process implementation.

[0029] In this embodiment, the internal composition of the GAN submodule is detailed: it includes a generator and a discriminator; the generator generates simulated performance parameters based on the input circuit design parameters and process parameters; the discriminator is used to determine the authenticity of the simulated performance parameters and the measured performance parameters in the intelligent design and manufacturing database; based on this, the genetic algorithm optimization submodule iteratively optimizes the network structure parameters of the generator and discriminator to improve the effect of adversarial training, and simultaneously optimizes the input manufacturing process parameters to minimize the difference between the simulated performance parameters and the target performance parameters.

[0030] This solution achieves precise and targeted simulation and parameter optimization of military power supply performance by disassembling the internal structure of GAN and clarifying the linkage rules with the genetic algorithm. The adversarial training between the generator and the discriminator improves the realism of the simulated performance parameters, making the simulation results more consistent with the actual production test data. The iterative optimization of the genetic algorithm submodule enables the synchronous tuning of network structure parameters and manufacturing process parameters, making the optimization process precisely target the goal of minimizing performance differences. This meets the stringent requirements of military power supplies for the accuracy of performance parameters and replaces the traditional time-consuming and labor-intensive tape-out verification process.

[0031] In this embodiment, the algorithm architecture and core functions of two core modules are defined: the design module is constructed using a deep convolutional neural network (CNN) to realize the mapping from the performance parameter space to the circuit design parameter space, and to complete the transformation from target performance to design parameters; the basic prediction framework of the collaborative optimization module is constructed using a deep graph neural network to model the topological relationship between circuit components, process parameters and electrical performance, and to provide a module foundation that fits the actual circuit for performance simulation.

[0032] Specifically, deep convolutional neural networks (DCNNs), as a high-precision classification and recognition algorithm, can achieve high-precision design parameter prediction, providing multiple alternative solutions for power module systems that flexibly meet design requirements. Based on an intelligent design and manufacturing database, an AI-based design module for in-package power module systems is built and trained, enabling automated forward design by inferring design and manufacturing process parameters from target electrical performance parameters. A collaborative optimization module employs deep graph neural networks (DNNs) to build and train an AI-based collaborative optimization module for the design and manufacturing of in-package power module systems, enabling prediction and verification from design and manufacturing process parameters to electrical performance. The design module and the DTCO AI module for in-package power module systems achieve bidirectional prediction "from performance to design" and "from design to performance," enabling high-speed, high-efficiency, and low-cost design and iterative optimization of in-package power modules.

[0033] This solution overcomes the problems of poor adaptability and high training redundancy of traditional general-purpose algorithm architectures by precisely matching the algorithm architecture with the module functions. Deep convolutional neural networks can efficiently extract key features of target electrical performance parameters, filter invalid information, and improve the accuracy and efficiency of mapping from the performance parameter space to the circuit design parameter space, making the output design parameters more in line with the performance requirements of military power supplies. Deep graph neural networks make up for the deficiency of traditional algorithms in accurately modeling circuit topology relationships, and can clearly capture the causal relationship between circuit components, process parameters and electrical performance, making the performance simulation results more in line with the actual working characteristics of the circuit. The targeted selection of the two algorithm architectures improves the prediction accuracy and training efficiency of the two major modules, reduces the redundancy of module training, and makes the modules more suitable for the design characteristics of military power supply modules.

[0034] In this embodiment, a data preprocessing module is added to the system. This module performs multi-level normalization preprocessing on the raw production line big data in the intelligent design and manufacturing database. The preprocessing performs standardization and equalization processing on the data distribution of different component parameters, process parameters and performance parameters respectively.

[0035] Specifically, the logic of multi-level normalization preprocessing is targeted layered processing: standardization and equalization are performed separately for three different types of data—component parameters (e.g., resistance, capacitance), manufacturing process parameters (e.g., temperature, time), and electrical performance parameters (e.g., voltage, efficiency)—to eliminate inherent defects in different data types. Those skilled in the art can design suitable processing flows based on existing data standardization (e.g., Z-score standardization, min-max standardization) and data equalization (e.g., oversampling, undersampling) algorithms, tailored to the distribution characteristics of different parameter types. This multi-level data preprocessing logic can be implemented through programming. For example, data from different sources can be converted to the same numerical scale, and sparse or unevenly distributed data can be enhanced or balanced.

[0036] This solution addresses the core issues of inconsistent parameter dimensions, uneven data distribution, and inconsistent data quality in the big data of military power supply production lines. These issues are key factors leading to low training accuracy and unstable prediction results in subsequent artificial intelligence modules. By employing multi-level targeted preprocessing for different types of parameters, the solution improves the quality and stability of subsequent module training, avoids module training bias or convergence difficulties caused by inconsistent data scales, ensures that the knowledge in the database can be fully and evenly learned by the modules, and especially enables the effective utilization of "defective data," thereby enhancing the reliability of the final design solution.

[0037] In this embodiment, an iterative control module is added to the system. This module provides autonomous control logic for the design and verification closed loop. Its core function is performance parameter comparison and optimization command triggering: First, the predicted electrical performance parameters output by the collaborative optimization module are compared with the target electrical performance parameters. If the difference exceeds a preset threshold, a feedback signal containing performance difference information is immediately generated and fed back to the design module and the collaborative optimization module, initiating a new round of iterative optimization. If the difference is less than or equal to the preset threshold, the current circuit design parameters and optimized manufacturing process parameters are directly output as the final design scheme. Those skilled in the art can set reasonable performance parameter comparison thresholds based on existing closed-loop control programming logic and combined with the performance standards of military power supplies, clarify the command generation rules and process optimization triggering conditions, and realize the full-process autonomous control of this module through programming.

[0038] This solution enables the system to autonomously iterate and optimize. The targeted feedback mechanism ensures that each iteration accurately targets performance bottlenecks, avoiding ineffective iterations. Dual optimization (module tuning + process optimization) improves iteration efficiency and the performance compliance rate of the final solution, reducing the risks of subsequent production and verification. It achieves a fully automated iterative optimization closed loop, seamlessly connecting the design, verification, and optimization stages. It can autonomously determine the optimization status and decide the iteration direction without human intervention, and coordinate the linkage between the front-end design and back-end optimization modules. It transforms the discrete "design-verification-modification" steps into a continuous, self-driven optimization process, shortening the time for manual decision-making and waiting.

[0039] In this embodiment, a prototype verification module is added, which is used to compare the performance of the final circuit design scheme generated by the iterative control module with a benchmark military power module product that did not participate in the database construction and training.

[0040] Specifically, after the system outputs the final circuit design through iterative optimization, this module does not immediately end the process. Instead, it compares this AI-generated design with a known reliable benchmark military power module product that was not used to build the current database and for training. The comparison focuses on various performance parameters to verify whether the performance of the AI ​​design reaches or approaches the level of existing mature products.

[0041] This solution provides external, objective performance benchmarking and verification of AI design results, solving the problem that AI design solutions often encounter: "performance meets standards but lacks adaptability to practical applications." It provides empirical support for the military application of the design solution: by comparing the performance with mature benchmark military power module products, the actual performance of the design solution can be directly verified, ensuring that the design solution not only meets the performance requirements in theory, but also meets the real needs of military scenarios in actual production and application. At the same time, the results of the performance comparison can be fed back into the intelligent design and manufacturing database, providing new measured data support for the subsequent training and optimization of the module, allowing the system's design capabilities to continuously improve with the accumulation of data.

[0042] In this embodiment, the core components of the target electrical performance parameters and the predicted electrical performance parameters in the system are clearly defined. Both include specific reliability indicators of the power module. These specific reliability indicators include at least one of the following: the drift rate of the output voltage under extreme temperatures and the stability parameter under mechanical stress. Those skilled in the art can, based on the actual application scenarios of military power modules (such as ships, aerospace), define the quantitative testing standards and methods for these specific reliability indicators and incorporate them into the system's performance parameter input and output system to achieve the parameter limitation.

[0043] This solution highlights the system's specific application to military power modules from a performance parameter perspective, clearly differentiating it from general-purpose circuit design systems. It addresses the issue of general-purpose design systems failing to cover reliability indicators under extreme military conditions: by incorporating core reliability indicators specific to military applications, such as extreme temperatures and mechanical stress, into the parameter system. This ensures that the entire process of system design, simulation, and optimization revolves around the core performance requirements of military power supplies, preventing design schemes from deviating from the strong environmental adaptability requirements of military products. Simultaneously, this parameter constraint makes module training and performance verification more targeted, ensuring that the designed power modules can operate stably under extreme military conditions, thus improving the environmental adaptability and actual operational reliability of military power modules.

[0044] This invention also provides an intelligent assisted design optimization method for military power modules, such as... Figure 2As shown, the process includes the following steps: S1. Constructing and maintaining an intelligent design and manufacturing database, performing multi-level normalization preprocessing on production line big data to provide high-quality data for module training; S2. Based on the preprocessed data, training and obtaining the design module and the collaborative optimization module respectively; S3. Receiving externally inputted target electrical performance parameters, generating initial circuit design parameters and corresponding initial manufacturing process parameters through the design module; S4. Inputting the initial circuit design parameters and initial manufacturing process parameters into the collaborative optimization module, outputting optimized manufacturing process parameters and corresponding predicted electrical performance parameters through the synergistic effect of its internal GAN ​​sub-module and genetic algorithm optimization sub-module; S5. Comparing the predicted electrical performance parameters with the target electrical performance parameters; if the difference is greater than a preset threshold, generating a feedback signal containing performance difference information and feeding it back to step S3 for iterative optimization; if the difference is less than or equal to the preset threshold, outputting the final circuit design parameters and collaboratively optimized manufacturing process parameters that meet the requirements.

[0045] This solution transforms the architecture of the intelligent assisted design optimization system into a standardized, implementable, and automated method, freeing the design process of military power modules from reliance on human experience. It achieves full-process standardization from data processing and module training to design generation, optimization verification, and result output, improving the standardization of the design process and the consistency of design solutions. This method replaces the traditional design mode that relies on human experience, shortens the overall design cycle of military power modules, and reduces R&D and trial-and-error costs. At the same time, the iterative optimization steps ensure that the design solution continuously approaches military performance targets, improving the performance compliance rate of the design solution and making the design process more adaptable to the R&D requirements of military products.

[0046] In this embodiment, a refined collaborative process between the Generative Adversarial Network (GAN) submodule and the Genetic Algorithm (GA) optimization submodule is clearly defined. First, the generator of the GAN submodule generates simulated performance based on the current circuit design parameters and process parameters. Then, the discriminator of the GAN submodule evaluates the fidelity of the simulated performance. The GA optimization submodule, based on the evaluation results and the difference between the simulated performance and the target performance, aims to improve the fidelity of the simulated performance and make it closer to the target performance. Simultaneously, it performs encoding, selection, crossover, and mutation operations on the network structure parameters of the generator and discriminator, as well as the input process parameters, to generate the next generation of optimized parameters for the next round of simulation optimization.

[0047] This solution constructs a closed-loop collaborative process of "simulation-evaluation-optimization," achieving precision and efficiency in the simulation optimization of military power supply processes and performance. It incorporates the difference between simulated and target performance into the optimization criteria of the genetic algorithm, making the optimization objective more precise and avoiding the problem of deviating from the design goal by focusing solely on realism. The encoding, selection, crossover, and mutation operations of the genetic algorithm enable global optimization of network structure parameters and manufacturing process parameters, avoiding the problem of getting trapped in local optima with a single optimization method, and ensuring that a globally optimal parameter combination suitable for military performance requirements can be found. Simultaneously, this automated collaborative optimization process requires no manual intervention, reducing the workload of designers, minimizing human error, and improving the efficiency and accuracy of military power supply process parameter optimization, making the simulation optimization results more closely aligned with the actual production and performance requirements of military power supplies.

[0048] The present invention also provides the following specific embodiments: Prototype verification will be conducted based on the large database of the in-package DC-DC power module system production line. By adapting and adjusting to different databases, the technology will be promoted to be applicable to the research and development of other in-package power module systems.

[0049] 1. Construction of an intelligent design and manufacturing database for packaged power module systems.

[0050] To enable the intelligent design module of the in-package power module system to better learn and fit, and to support the offline and online training of the artificial intelligence design and optimization algorithm module, various data accumulated from the design and manufacturing of in-package power module systems are organized and divided into two categories: design and manufacturing data that meets the design requirements of the in-package power module system and can achieve the performance indicators well; and design and manufacturing data that does not meet the design requirements of the in-package power module system and contains design defects.

[0051] The construction of this database primarily involves the selection of power system topologies and components. Data meeting packaging requirements is compiled based on different power ratings, output voltages, input voltages, number of output paths, and control logics. This involves selecting and organizing corresponding power system topologies, and identifying the optimal existing components in the product based on performance indicators, such as the selection of primary-side power MOSFETs and secondary-side rectifiers. This process requires continuous supplementation from a large number of products and engineer-verified solutions, while adhering to existing design specifications and procedures. Data not meeting packaging requirements focuses on distinguishing and comparing design flaws encountered during the design process and manufacturing defects encountered during production. This allows for more accurate module training, ensuring bidirectional prediction "from performance to design" and "from design to performance."

[0052] 2. Design and development of the internal power module system.

[0053] To build a high-precision, compact AI design module based on an intelligent design and manufacturing database, such as Figure 3 As shown, an AI prediction module algorithm is built based on a deep neural network. The AI ​​prediction module is repeatedly trained using data sets of power module performance parameters and design parameters of various device modules from the database, improving recognition and prediction accuracy. Various power module circuit topologies are taken into consideration, enabling the design module to achieve full coverage of design parameters for different circuit topologies. This improves the module's topology versatility and allows the production line big data of various power modules to be used for module training, increasing the utilization efficiency of the intelligent design and manufacturing database.

[0054] Statistical analysis methods are employed to analyze the impact of the amount of training data required for the design module on prediction accuracy and training / prediction processing time. This improves the computational speed of the design module while reducing its dependence on the amount of training data in the database, making the design module applicable to the automated design of power modules with various database sizes, thus expanding its application scope. Based on a manufacturing database, multiple design module circuits are established to achieve high-precision circuit simulation of the power module designs generated by the design modules. The system links and provides feedback between the design module's design parameter prediction and high-precision circuit simulation, assisting design engineers in achieving highly efficient automated design and iterative verification of packaged DC-DC power modules.

[0055] 3. Development of a collaborative optimization module for the internal power module system.

[0056] To obtain a simulation module for training an AI neural network and optimize related process parameter combinations, the following technical approach and research methods are proposed: Due to the involvement of various circuit components and the complexity of the process parameters involved, a Genetic Animation Array (GAN) is used to construct the AI ​​neural network simulation module for efficient and controllable processing of such complex datasets. GAN employs adversarial training between the generator and discriminator, a simple and controllable training method that significantly improves the training efficiency of the generative module. Furthermore, since GAN does not require complex sampling sequences, it can directly generate batches of sample data, increasing the efficiency of new sample generation. Moreover, both the generator and discriminator in a GAN utilize neural networks, allowing for optimization of the algorithm's performance by adjusting the neural network architecture. This high degree of design freedom enables GAN to flexibly adapt to different application scenarios and requirements. However, in constructing the neural network simulation module using GAN, manual parameter tuning is inefficient due to the large number of parameters involved, resulting in unsatisfactory stability of the simulation module. Therefore, a genetic algorithm is introduced to collaboratively optimize the network structure parameters of the GAN sub-module and the manufacturing process parameters it processes, thereby obtaining a stable and highly accurate AI simulation module tool. Genetic algorithms (GA) can quickly search for excellent solutions in the entire solution space by simulating operations such as selection, crossover, and mutation in the evolutionary process, thus improving optimization efficiency.

[0057] 4. Prototype verification of the application of high-efficiency and low-cost intelligent design and optimization technologies.

[0058] To verify the application prototype of high-efficiency and low-cost intelligent design and optimization technologies, data from existing mature DC / DC power supply products that were not included in the database training set were summarized. After setting the product requirement indicators to be verified for the AI ​​module, and after training, appropriate topologies and electronic components were selected and circuit diagrams were provided, specific parameters of the two were compared with those of existing mature DC / DC power supply module systems. These parameters included the specific topology of the DC / DC power supply, the selection of specific components, and whether the performance of the AI ​​module-designed power supply module system could approach or surpass the parameters of the existing product power supply module system.

[0059] Prototype testing of both in-package systems and on-chip power modules verified the superiority of DC-DC power modules in shortening design cycles and reducing R&D costs for applications in marine vessels and aerospace.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent assisted design optimization system for military power modules, characterized in that, include: The intelligent design and manufacturing database is used to structure and store the production line big data of the power module in the package. The production line big data includes successful design data that meets the design requirements, manufacturing process and its corresponding measured performance data, as well as negative sample data containing design defects or process defects. The design module, trained based on successful design data in the intelligent design and manufacturing database, is used to receive target electrical performance parameters and output preliminary circuit design parameters and corresponding initial manufacturing process parameters. The collaborative optimization module includes a generative adversarial network submodule and a genetic algorithm optimization submodule. Based on the training of the intelligent design and manufacturing database, it is used to receive the preliminary circuit design parameters and the corresponding initial manufacturing process parameters, and output the predicted electrical performance parameters. The output of the design module is connected to the input of the collaborative optimization module to form a design and verification closed loop.

2. The intelligent assisted design optimization system for military power modules according to claim 1, characterized in that, The generative adversarial network submodule is used to simulate the impact of manufacturing processes on circuit performance; the genetic algorithm optimization submodule is used to collaboratively optimize the network parameters of the generative adversarial network submodule and the input manufacturing process parameters.

3. The intelligent assisted design optimization system for military power modules according to claim 2, characterized in that, The generative adversarial network submodule includes a generator and a discriminator; The generator generates analog performance parameters based on the input circuit design parameters and process parameters; The discriminator is used to determine the authenticity of the simulated performance parameters and the measured performance parameters of the intelligent design and manufacturing database; The genetic algorithm optimization submodule iteratively optimizes the network structure parameters of the generator and discriminator, and synchronously optimizes the input process parameters to minimize the difference between the simulated performance parameters and the target performance parameters.

4. The intelligent assisted design optimization system for military power modules according to claim 1, characterized in that, The design module is constructed using a deep convolutional neural network to achieve the mapping from the performance parameter space to the circuit design parameter space; The basic prediction framework of the collaborative optimization module is constructed using a deep graph neural network to model the topological relationships between circuit elements, process parameters, and electrical performance.

5. The intelligent assisted design optimization system for military power modules according to claim 1, characterized in that, The system also includes a data preprocessing module, which is used to perform multi-level normalization preprocessing on the data in the intelligent design and manufacturing database; The preprocessing performs standardization and equalization processes on the data distribution of different component parameters, process parameters, and performance parameters.

6. The intelligent assisted design optimization system for military power modules according to claim 1, characterized in that, The system also includes an iteration control module for: Compare the predicted electrical performance parameters output by the collaborative optimization module with the target electrical performance parameters; When the difference between the two exceeds a preset threshold, a feedback signal containing performance difference information is generated and fed back to the design module and the collaborative optimization module to start a new round of iterative optimization. When the difference between the two is less than or equal to the preset threshold, the current circuit design parameters and the optimized manufacturing process parameters are output as the final solution.

7. The intelligent assisted design optimization system for military power modules according to claim 1, characterized in that, The system also includes a prototype verification module, which is used to compare the performance of the final circuit design generated by the iterative control module with a benchmark military power module product that did not participate in the database construction and training.

8. The intelligent assisted design optimization system for military power modules according to claim 1, characterized in that, Both the target electrical performance parameters and the predicted electrical performance parameters include specific reliability indicators of the power module. These specific reliability indicators include at least one of the following: the output voltage drift rate at extreme temperatures and the stability parameter under mechanical stress.

9. A method for intelligent assisted design optimization of a military power supply module, characterized in that, Includes the following steps: S1. Build and maintain an intelligent design and manufacturing database, and perform multi-level normalization preprocessing on production line big data; S2. Based on the preprocessed data, the design module and the collaborative optimization module are trained and obtained respectively; S3. Receive the target electrical performance parameters input, and generate initial circuit design parameters and corresponding initial manufacturing process parameters through the design module; S4. Input the initial circuit design parameters and initial manufacturing process parameters into the collaborative optimization module. Through the collaborative effect of its internal generative adversarial network submodule and genetic algorithm optimization submodule, the optimized manufacturing process parameters and corresponding predicted electrical performance parameters are output. S5. Compare the predicted electrical performance parameters with the target electrical performance parameters; If the difference is greater than a preset threshold, a feedback signal containing performance difference information is generated and fed back to step S3 for iterative optimization. If the difference is less than or equal to the preset threshold, the final circuit design parameters and co-optimized manufacturing process parameters that meet the requirements will be output.

10. The method according to claim 9, characterized in that, The synergistic effect between the generative adversarial network submodule and the genetic algorithm optimization submodule includes: The generator of the generative adversarial network submodule generates simulated performance based on the current circuit design parameters and process parameters; The discriminator of the generative adversarial network submodule evaluates the realism of the simulation performance; The genetic algorithm optimization submodule, based on the evaluation results and the difference between the simulation performance and the target performance, aims to improve the realism of the simulation performance and make it closer to the target performance. At the same time, it performs encoding, selection, crossover, and mutation operations on the network structure parameters of the generator and discriminator, as well as the input process parameters, to generate the next generation of optimization parameters.

Images