A method for determining the design parameters of a silicon carbide switch
By integrating multi-source data and modeling manufacturing deviations, combined with robust multi-objective solving and reinforcement learning, the influence of manufacturing deviations in silicon carbide switch design was addressed. This enabled the generation of robust parameter windows, improving the stability and reliability of design parameters and meeting the application requirements of high-frequency and high-power applications.
Patent Information
- Application Number
- CN202610213607.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-13
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-02-13
AI Technical Summary
In the current silicon carbide switch design process, it is difficult to systematically consider the impact of manufacturing deviation factors such as batch differences, device discreteness, parasitic changes in packaging, and temperature drift. This leads to unstable determination of design parameters, making it difficult to meet the requirements of high switching frequency, high power density, and high reliability.
By collecting multi-source data, a design parameter space is established, manufacturing deviation factors are identified, a manufacturing deviation description model is constructed, manufacturing deviation constraints are generated, a robust evaluation model is established, robust multi-objective solution and reinforcement learning training are performed, a robust parameter window is generated, and simulation and experimental evaluation are conducted to ensure the robustness and coverage of design parameters.
It significantly improves the robustness and engineering applicability of silicon carbide switch design parameters, reduces design uncertainty, shortens the design cycle, reduces R&D costs, and ensures the traceability and reliability of design results.
Smart Images

Figure CN122154433B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer-aided design technology, and in particular to a method for determining the design parameters of a silicon carbide switch. Background Technology
[0002] Silicon carbide (SiC) switching devices are widely used in new energy power generation, electric vehicles, power electronic converters, and high-power-density power supply systems due to their high voltage, high frequency, high temperature, and low loss characteristics. As applications evolve towards higher switching frequencies, higher power densities, and higher reliability, the device structure parameters, packaging parasitic parameters, drive link parameters, and thermal management-related parameters of SiC switches exhibit a high degree of coupling.
[0003] In the existing silicon carbide switch design process, the determination of design parameters usually relies on empirical rules, simulation analysis under a single operating condition, or small-scale prototype testing. Such methods often assume that the device parameters and operating environment have ideal consistency, making it difficult to systematically consider the influence of manufacturing deviations such as batch differences, device discreteness, packaging parasitic changes, and temperature drift.
[0004] Therefore, we propose a method for determining the design parameters of silicon carbide switches. The information disclosed above in the Background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for determining the design parameters of silicon carbide switches, thereby solving the technical problems mentioned in the background section.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for determining the design parameters of a silicon carbide switch includes the following steps: S1. Collect and process multi-source data related to silicon carbide switches, extract evaluation elements based on multi-source data, establish a design parameter space, and generate a set of candidate design parameters with unique identifiers within the design parameter space. S2. Identify manufacturing deviation factors from multi-source data and form deviation characterizations. Based on the deviation characterizations, construct a manufacturing deviation description model, generate manufacturing deviation constraints, establish a mapping relationship with the candidate design parameter set, and perform consistency verification on the manufacturing deviation constraints. S3. Establish a robust evaluation model with the candidate design parameter set, deviation characterization, manufacturing deviation constraints and evaluation elements as inputs, calculate the robust performance evaluation results corresponding one-to-one with the candidate design parameter set and form a robust performance evaluation record, and update the evaluation credibility information based on the robust performance evaluation record. S4. Based on the robust performance evaluation records, determine the objective set and constraint set for robust multi-objective solution, screen the optimal feasible candidate design parameter subset, generate an initial robust parameter window, and perform coverage verification on the initial robust parameter window to obtain a usable initial robust parameter window. S5. Construct an active sampling iterative environment and adopt reinforcement learning to train the sampling decision strategy. Select the sampling object in the initial robust parameter window or iterative robust parameter window and trigger simulation evaluation and experimental evaluation to obtain new evaluation results. Use the new evaluation results to update the manufacturing deviation description model and robust evaluation model, and shrink the initial robust parameter window to generate the iterative robust parameter window and form a closed loop iteration. S6. Perform convergence determination based on the iterative robust parameter window. When the convergence condition is met, output the final robust parameter window as the design parameter determination result, and generate verification records and traceable evaluation basis.
[0007] S1 specifically includes: collecting transient waveform data, temperature response data, operating condition data, and historical reliability statistics of the switch; performing time alignment, data cleaning, and standardization on the multi-source data to obtain standardized multi-source data; extracting efficiency-related evaluation elements, switch stress-related evaluation elements, and reliability-related evaluation elements based on the standardized multi-source data and establishing corresponding relationships; establishing and associating a design parameter space that includes the value ranges of device structure design parameters, package parasitic design parameters, drive link design parameters, and thermal management design parameters; generating a set of candidate design parameters that meet the coverage requirements within the design parameter space and generating a unique identifier.
[0008] S2 specifically includes: identifying manufacturing deviation factors such as device discreteness, packaging parasitic differences, drive link differences, and temperature drift differences from standardized multi-source data; grouping these factors by batch identifier, temperature range identifier, and operating condition identifier to quantify and characterize them to form deviation characterizations; constructing a manufacturing deviation description model based on the deviation characterizations to obtain an estimate of the fluctuation range of each indicator under the influence of manufacturing deviation factors; generating manufacturing deviation constraints based on the fluctuation range estimates and establishing a mapping relationship with the candidate design parameter set; performing boundary conflict verification and feasibility verification on the manufacturing deviation constraints, and triggering backtracking adjustments when feasibility is insufficient.
[0009] S3 specifically includes: establishing a robust evaluation model with candidate design parameter values, deviation characterization, manufacturing deviation constraints, and evaluation elements as inputs; calculating the robust performance evaluation results of efficiency-related indicators, switching stress-related indicators, and reliability-related indicators, and corresponding them one-to-one with the unique identifiers of the candidate design parameters; organizing the robust performance evaluation results into robust performance evaluation records that include indicator statistics, lower bound estimates, upper bound estimates, and constraint judgment labels; and updating the evaluation credibility information based on the deviation between observed and predicted indicator values to characterize the reliability of the robust evaluation model within different candidate design parameter ranges.
[0010] S4 specifically includes: determining the objective set and constraint set for robust multi-objective solution based on robust performance evaluation records, and using the constraint judgment label satisfaction as a feasibility criterion; screening feasible candidate design parameters from the candidate design parameter set and obtaining the optimal feasible candidate design parameter subset using non-dominated sorting; statistically obtaining the lower and upper boundaries of each design parameter based on the optimal feasible candidate design parameter subset to generate an initial robust parameter window; and verifying the coverage of the initial robust parameter window on the key operating condition set, triggering backtracking adjustment and updating the initial robust parameter window when the coverage is insufficient.
[0011] S5 specifically includes: constructing an active sampling iterative environment with evaluation credibility information, robust performance evaluation record distribution characteristics, iterative robust parameter window boundary information, and key operating condition set coverage as state information, and defining the sampling objects as a list of sampling objects represented by unique identifiers of candidate design parameters; using reinforcement learning to train the sampling decision strategy to output the sampling objects and trigger simulation evaluation and experimental evaluation to obtain new evaluation results, supplementing the new evaluation results to the standardized multi-source data and synchronously updating the robust performance evaluation records; updating the manufacturing deviation description model and robust evaluation model based on the updated standardized multi-source data, and re-executing robust multi-objective solution to shrink the initial robust parameter window to obtain the iterative robust parameter window, while backfilling the updated window boundary information, evaluation credibility information, feasibility rate, and coverage rate into the active sampling iterative environment to form a closed-loop iteration.
[0012] S6 specifically includes: performing convergence determination based on iterative robust parameter windows, the convergence determination including at least a comprehensive determination of the window boundary change magnitude, confidence improvement magnitude, feasibility rate, and coverage rate; determining and outputting the final robust parameter window as the design parameter determination result when the comprehensive convergence condition is met, and generating a verification record containing the simulation evaluation results, measured evaluation results, and cross-condition verification results corresponding to the final robust parameter window; and generating traceable evaluation basis, which includes at least the manufacturing deviation description model version, robust evaluation model version, sampling decision strategy version, final snapshot of robust performance evaluation record, convergence determination result, and key condition set coverage verification result, for use in reproducing, auditing, and deploying the design parameter determination result.
[0013] The beneficial effects of this invention are as follows: This invention introduces a multi-source data fusion and manufacturing deviation modeling mechanism to explicitly consider manufacturing deviation factors such as device discreteness, packaging parasitic differences, and temperature drift during the design parameter determination process. This ensures that the determined silicon carbide switch design parameters maintain stable performance under batch manufacturing and multi-condition operation, significantly improving the robustness and engineering applicability of the design results. By constructing a robust evaluation model and performing statistical performance evaluation and constraint determination on the design parameters, coordinated optimization can be achieved among multiple objectives such as efficiency, switching stress, and reliability. This avoids the performance imbalance problems caused by traditional single-condition or single-index optimization methods, thereby improving the overall performance of silicon carbide switches.
[0014] This invention employs a combination of robust multi-objective solution and parameter window generation to directly output a robust parameter window with engineering significance, rather than a single-point optimal solution. This allows designers to select components and integrate systems with a clear understanding of parameter value ranges, reducing design uncertainty. By introducing an active sampling iteration mechanism based on reinforcement learning, the simulation and experimental evaluation objects can be adaptively selected according to model credibility and coverage information. This ensures design accuracy while reducing the number of invalid simulations and experiments, effectively lowering R&D costs and shortening the design cycle.
[0015] This invention ensures that the obtained robust parameter window covers the critical operating conditions and converges stably during iteration by incorporating coverage verification and convergence determination mechanisms during the parameter determination process. This avoids engineering risks caused by excessively wide or narrow parameter ranges. By generating complete verification records and traceable evaluation evidence, it achieves end-to-end traceability from raw data, models, optimization processes to the final design parameter results, which is beneficial for the reproduction, auditing, and subsequent product iteration and upgrades of design results. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of a method for determining the design parameters of a silicon carbide switch according to the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] Example 1: As Figure 1 As shown in the figure, this embodiment provides a method for determining the design parameters of a silicon carbide switch, including the following steps: S1. Collect and process multi-source data related to silicon carbide switches, extract evaluation elements based on multi-source data, establish a design parameter space, and generate a set of candidate design parameters with unique identifiers within the design parameter space. S2. Identify manufacturing deviation factors from multi-source data and form deviation characterizations. Based on the deviation characterizations, construct a manufacturing deviation description model, generate manufacturing deviation constraints, establish a mapping relationship with the candidate design parameter set, and perform consistency verification on the manufacturing deviation constraints. S3. Establish a robust evaluation model with the candidate design parameter set, deviation characterization, manufacturing deviation constraints and evaluation elements as inputs, calculate the robust performance evaluation results corresponding one-to-one with the candidate design parameter set and form a robust performance evaluation record, and update the evaluation credibility information based on the robust performance evaluation record. S4. Based on the robust performance evaluation records, determine the objective set and constraint set for robust multi-objective solution, screen the optimal feasible candidate design parameter subset, generate an initial robust parameter window, and perform coverage verification on the initial robust parameter window to obtain a usable initial robust parameter window. S5. Construct an active sampling iterative environment and adopt reinforcement learning to train the sampling decision strategy. Select the sampling object in the initial robust parameter window or iterative robust parameter window and trigger simulation evaluation and experimental evaluation to obtain new evaluation results. Use the new evaluation results to update the manufacturing deviation description model and robust evaluation model, and shrink the initial robust parameter window to generate the iterative robust parameter window and form a closed loop iteration. S6. Perform convergence determination based on the iterative robust parameter window. When the convergence condition is met, output the final robust parameter window as the design parameter determination result, and generate verification records and traceable evaluation basis.
[0019] S1 specifically includes the following sub-steps: S110. Collect and aggregate multi-source data from silicon carbide switches. The multi-source data includes at least transient waveform data, temperature response data, operating condition data, and historical reliability statistics. Perform unified time alignment and data cleaning on the multi-source data. Time alignment includes at least resampling and aligning data at different sampling rates using the trigger event of the switch transient as the zero-time reference. Data cleaning includes at least missing value handling and outlier removal. After completing time alignment and data cleaning, perform standardization on the multi-source data to obtain standardized multi-source data. The standardization process must at least satisfy the following formula: ; in, The original data value; This represents the mean of the original data values in the corresponding sample set. This represents the standard deviation of the original data values in the corresponding sample set. The data values are standardized; standardized multi-source data are organized into a set of data records that can be directly read by subsequent steps, and each data record contains at least a timestamp field, a waveform sequence field, a temperature sequence field, an operating condition identifier field, and a reliability statistics field.
[0020] S120. Based on standardized multi-source data extraction, evaluation elements related to design parameters are extracted. These evaluation elements include at least efficiency-related evaluation elements, switching stress-related evaluation elements, and reliability-related evaluation elements. To ensure that the evaluation elements are calculable and reproducible, switching energy is calculated from the switching transient waveform data as one of the efficiency-related evaluation elements. The switching energy satisfies the following formula: ; in, For switching energy; The start time of the time window for calculating switching energy; This is the end time of the time window for calculating the switching energy. The waveform of drain-source voltage changing over time; This is the waveform of the drain current changing over time. It is a time variable; and, the start time and the end time The selection rules include at least the following: taking the zero moment corresponding to the triggering event as a reference and sampling within a preset time window, or taking the moment when the voltage and current reach the steady-state threshold as the boundary; At the same time, establish a correspondence between evaluation elements and standardized multi-source data. The correspondence should be recorded in the form of "evaluation element name - input field - calculation window - output unit" to ensure that the subsequent steps call the evaluation elements in a consistent manner.
[0021] S130. Establish the design parameter space for silicon carbide switches. The design parameter space shall include at least the value ranges of device structure design parameters, package parasitic design parameters, drive link design parameters, and thermal management design parameters. The determination of the value range includes at least the following: providing initial upper and lower limits based on the device datasheet or process capability; correcting the initial upper and lower limits based on the safety boundary; and discretizing the value range based on the adjustable resolution or selectable level. The design parameter space is expressed in a structured form as a parameter list, which includes at least the name, unit, lower limit, upper limit and selectable level information of each design parameter. The parameter list is then associated with the evaluation elements. This association includes at least a set of evaluation elements that specify the impact of each type of design parameter, thus forming a consistent mapping between "design parameters and evaluation elements" for direct reference in subsequent robustness evaluation and solution steps.
[0022] S140. Generate a set of candidate design parameters in the design parameter space. The set of candidate design parameters contains multiple sets of candidate design parameters. The generation of the set of candidate design parameters must at least meet the coverage requirement. The coverage requirement includes at least the following: each type of design parameter is covered to a preset number of levels within its value range, and the combination coverage of different types of design parameters reaches a preset ratio. The candidate generation method shall employ at least one of the following: hierarchical coverage sampling, Latin hypercube sampling, or grid sampling, to ensure that the candidate design parameter set is uniformly distributed within the design parameter space. To ensure consistent correspondence of the candidate design parameter set in subsequent evaluation and traceability, a unique identifier shall be generated for each group of candidate design parameters. The unique identifier shall be formed by concatenating the candidate sequence number, operating condition identifier, and generation batch number. The unique identifier shall be used as the primary key to bind each group of candidate design parameters to the subsequently generated robust performance evaluation results.
[0023] S2 specifically includes the following sub-steps: S210. Identify manufacturing deviation factors from standardized multi-source data. Manufacturing deviation factors include at least device discreteness, packaging parasitic differences, drive chain differences, and temperature drift differences. To quantify manufacturing deviation factors, standardized multi-source data are grouped according to preset grouping rules, which include grouping by batch identifier, temperature range identifier, and operating condition identifier. For each type of manufacturing deviation factor, its quantified value is extracted within the corresponding group, and a statistical measure is calculated, including at least the mean and standard deviation. Furthermore, a deviation interval is formed based on the mean and standard deviation, and the deviation interval must satisfy at least the following formula: ; in, A quantitative value corresponding to a certain manufacturing deviation factor; This is the mean of the quantified value in the corresponding grouped sample set; This represents the standard deviation of the quantified value within the corresponding grouped sample set; The coverage factor is used to set the coverage width of the deviation range; and the deviation range is output as the deviation representation of the manufacturing deviation factor for direct use by the subsequent manufacturing deviation description model.
[0024] S220. Construct a manufacturing deviation description model based on deviation characterization. The manufacturing deviation description model is used to describe the propagation relationship of the influence of manufacturing deviation factors on the performance corresponding to the candidate design parameter set. The manufacturing deviation description model takes candidate design parameters, deviation characterization and operating condition identification as inputs and the statistics of robust performance evaluation results as outputs. The robust performance evaluation results include at least efficiency-related indicators, switching stress-related indicators and reliability-related indicators. To estimate the fluctuation range of robust performance evaluation results, the mean and standard deviation of each indicator are calculated, and lower bound estimates and upper bound estimates are formed. The lower bound estimates and upper bound estimates must satisfy at least the following formula: ; in, This refers to a specific index value in the robust performance evaluation results. This is the mean value of the indicator under the influence of manufacturing deviation factors; This represents the standard deviation of the indicator value under the influence of manufacturing deviation factors. This is a lower bound estimate of the index value; This is an upper bound estimate of the index value; The coverage coefficient is consistent with the coverage coefficient in step S210; and for each set of candidate design parameters, a fluctuation range estimation record carrying its unique identifier is output, so that the fluctuation range estimation record and the set of candidate design parameters are in one-to-one correspondence.
[0025] S230. Generate manufacturing deviation constraints based on the manufacturing deviation description model. The manufacturing deviation constraints are used to limit the safety boundary and performance lower limit that must still be met when manufacturing deviation factors exist. Manufacturing deviation constraints are uniformly expressed as upper bound constraints and lower bound constraints. Upper bound constraints limit the upper bound estimate of the indicator value to no more than the corresponding upper limit constraint threshold, while lower bound constraints limit the lower bound estimate of the indicator value to no less than the corresponding lower limit constraint threshold. Each constraint must satisfy at least the following formula: ; ; in, This is an upper bound estimate of the index value; This is a lower bound estimate of the index value; The upper limit constraint threshold corresponding to the indicator value is used to limit the maximum allowable value of the indicator value; The lower limit constraint threshold corresponding to the index value is used to limit the minimum allowable value of the index value; and a mapping relationship is established between the manufacturing deviation constraint and the candidate design parameter set. The mapping relationship includes at least: recording a unique identifier for each manufacturing deviation constraint carrying the corresponding candidate design parameter, so as to use the unified calling and tracing of the manufacturing deviation constraint in subsequent steps.
[0026] S240. Perform consistency verification on manufacturing deviation constraints. Consistency verification includes at least boundary conflict verification and feasibility verification. Boundary conflict verification is used to identify whether there are contradictory upper and lower bound constraints under the same operating condition. Feasibility verification is used to statistically analyze the proportion of candidate design parameters that meet the manufacturing deviation constraints. To make feasibility verification calculable, a feasibility rate is defined and backtracking adjustments are triggered accordingly. The feasibility rate must at least satisfy the following formula: ; in, Feasibility rate; The number of candidate design parameters to meet manufacturing deviation constraints; The total number of candidate design parameters; when the feasibility rate Less than the minimum feasible threshold Time triggers backtracking adjustment, in which, The minimum feasible threshold; The backtracking adjustment includes at least adjusting the deviation representation, the value range of the design parameter space, and the generation strategy of the candidate design parameter set in order of preset priority, and re-executing the consistency check after the backtracking adjustment is completed until feasible manufacturing deviation constraints are obtained.
[0027] S3 specifically includes the following sub-steps: S310. Establish a robust evaluation model. The robust evaluation model takes the set of candidate design parameters and manufacturing deviation constraints as inputs and the robust performance evaluation results as outputs. To ensure that the robust evaluation model can be implemented and reproduced, the inputs of the robust evaluation model shall include at least: unique identifiers of candidate design parameters, values of candidate design parameters, operating condition identifiers, deviation representations output from step S210, manufacturing deviation constraints output from step S230, and evaluation elements output from step S120. The output of a robust evaluation model includes at least the mean and standard deviation of efficiency-related indicators, switching stress-related indicators, and reliability-related indicators, and can further derive lower bound estimates for the indicator values. Compared with the upper bound estimate ,in, The lower bound estimate of the index value and in step S220 Maintain consistency The upper bound estimate of the index value and in step S220 Maintain consistency; The robust evaluation model is formed in at least the following ways: by fitting training samples generated from simulation evaluation and experimental evaluation, or by combining a mechanistic calculation model with data-driven correction terms. Furthermore, the robust evaluation model updates its parameters according to a preset update rule when new evaluation results are introduced in subsequent steps.
[0028] In a preferred implementation of this embodiment, the robust evaluation model is constructed using a multi-layer feedforward neural network (e.g., a backpropagation (BP) neural network). This neural network establishes a nonlinear mapping relationship by learning from a large amount of simulation input and output data. When calculating the robust performance evaluation results, multiple sets of manufacturing deviation sampling data conforming to the deviation distribution are input in batches into the trained neural network to quickly obtain corresponding multiple sets of performance prediction values. Statistical analysis is then performed on these multiple sets of prediction values to calculate the mean and standard deviation of the index values. This approach avoids repeatedly performing time-consuming physical field simulations, significantly improving computational efficiency.
[0029] S320. For each group of candidate design parameters, calculate its robust performance evaluation result under manufacturing deviation constraints based on the robust evaluation model, and maintain a one-to-one correspondence with the unique identifier of the candidate design parameter; to make the robust performance evaluation result computable, for any index value in the robust performance evaluation result, calculate the mean and standard deviation of the index value based on the sample set, and the mean and standard deviation shall at least satisfy the following formula: ; ; in, The sample size represents the number of evaluation samples obtained under the influence of manufacturing deviation factors for the same candidate design parameter unique identifier and the same working condition identifier. For the first The indicator values corresponding to each evaluation sample; The mean of the index values and the mean of the values in step S220 Maintain consistency; The standard deviation of the indicator value, and its Chinese definition is the same as in step S220. Maintain consistency; The evaluation samples are derived from the output results of simulation evaluation and / or experimental evaluation, and each evaluation sample carries a unique identifier for the candidate design parameters to ensure a one-to-one correspondence; and the corresponding robust performance evaluation results are generated based on the mean and standard deviation for subsequent recording and judgment.
[0030] S330. Organize the robust performance evaluation results into a robust performance evaluation record. The robust performance evaluation record shall at least include: a unique identifier for the candidate design parameter, an operating condition identifier, a reference information for manufacturing deviation constraints, and the mean value of the aforementioned index. Standard deviation of indicator values Lower bound estimation of index values Upper bound estimation of index values Upper limit constraint threshold Lower limit constraint threshold and constraint determination labels; in, The upper limit constraint threshold corresponding to the indicator value and the Chinese definition is the same as in step S230. The Chinese definition remains consistent. The lower limit constraint threshold corresponding to the indicator value and its Chinese definition is the same as in step S230. To maintain consistency with the Chinese definitions, and to ensure that constraint decision labels are computable, the constraint decision labels are defined to satisfy the following rules: ; Wherein, b is a constraint judgment label, used to indicate whether the corresponding candidate design parameter satisfies the manufacturing deviation constraint condition under the corresponding working condition; and the constraint judgment label is bound to the unique identifier of the candidate design parameter to ensure that the subsequent solution steps have a consistent basis for screening feasible candidate design parameters.
[0031] S340. Perform a confidence update based on the robust performance evaluation records. The confidence update is used to mark the reliability of the robust evaluation model in different candidate design parameter regions and generate evaluation confidence information. To make the confidence update feasible, define the prediction error and calculate the confidence based on it. The prediction error must at least satisfy the following formula: ; in, This represents the prediction error; The observed index value represents the index value obtained from measured evaluation and / or simulation evaluation. Let be the predicted index value, representing the index prediction value output by the robust evaluation model; further, the confidence level is calculated based on the prediction error, and the confidence level must at least satisfy the following formula: ; in, The credibility is used to characterize the reliability of the robust evaluation model under the unique identifier of the corresponding candidate design parameter and the corresponding operating condition identifier; The scaling parameter is used to adjust the intensity of the impact of prediction error on reliability; when new evaluation samples are added to the robust performance evaluation record, the newly added observation index value is used. Calculate prediction error And update the corresponding credibility. The updated evaluation credibility information is then bound to the unique identifier of the candidate design parameters and the working condition identifier, so that they can be uniformly called in subsequent robust multi-objective solution and active sampling iteration steps.
[0032] For example, scale parameters The value can be taken as 0.5 to 1.0 times the historical standard deviation of the corresponding indicator. In a specific application scenario, if the standard deviation of the efficiency indicator is approximately 0.5%, then it can be set to... This means that when the prediction error reaches 0.5%, the model's reliability decays to [a certain value]. (Approximately 0.368), thus enabling a sensitive measure of model bias.
[0033] S4 specifically includes the following sub-steps: S410. Determine the objective set and constraint set for robust multi-objective solution based on robust performance evaluation records. The objective set shall include at least efficiency-related objectives and reliability-related objectives, and the constraint set shall include at least manufacturing deviation constraints and switching stress boundary constraints. To ensure that the objective set and constraint set are realizable and reproducible, both the objective set and constraint set are composed of computable quantities from the robust performance evaluation records, and these computable quantities include at least the mean of the index values. Standard deviation of indicator values Lower bound estimation of index value Upper bound estimate of index value ,in, The mean of the index values and the mean of the values in step S220 Maintain consistency The standard deviation of the index value and the value in step S220 Maintain consistency The lower bound estimate of the index value and in step S220 Maintain consistency The upper bound estimate of the index value and in step S220 Maintain consistency; Meanwhile, the feasibility criterion for the constraint set includes at least: the constraint decision label obtained in step S330. As the criterion for determining whether candidate design parameters meet manufacturing deviation constraints, among which, To constrain the determination label and in step S330 To maintain consistency, and to further avoid solving in unreliable regions of the model, the constraint set may also include confidence filtering conditions. ,in, For credibility and in step S340 Maintain consistency This is the minimum confidence threshold, used to filter candidate design parameters whose confidence level is lower than the minimum confidence threshold.
[0034] S420. Perform robust multi-objective solution on the candidate design parameter set. The robust multi-objective solution includes at least: screening feasible candidate design parameters from the candidate design parameters that satisfy the constraint set, and performing multi-objective optimal screening on the feasible candidate design parameters to obtain the optimal subset of feasible candidate design parameters. Among these, the selection of feasible candidate design parameters should include at least: retaining only the constraint-satisfying labels. The candidate design parameters are selected, and when using confidence filtering conditions, only those that meet the criteria are retained. Candidate design parameters; multi-objective optimal screening is achieved at least by using non-dominated ranking rules, which are used to make trade-offs between efficiency-related objectives and reliability-related objectives, and the first-level solution set of the non-dominated ranking is determined as the subset of optimal feasible candidate design parameters; Each set of candidate design parameters in the subset of optimal feasible candidate design parameters carries a unique identifier to maintain a one-to-one correspondence with robust performance evaluation records and ensure traceability.
[0035] S430. Generate an initial robust parameter window based on a subset of the optimal feasible candidate design parameters. The initial robust parameter window is used to give the boundary of the range of available design parameter values in the design parameter space. To make the initial robust parameter window computable, for the first... For each design parameter, the lower and upper boundaries are statistically derived from a subset of the optimal feasible candidate design parameters. The lower and upper boundaries must satisfy at least the following formula: , ; in, For the first The first group of candidate design parameters The values of each design parameter; For the first The lower boundary of each design parameter in the initial robust parameter window; For the first The design parameters are at the upper boundary of the initial robust parameter window; It is an index set of the subset of optimal feasible candidate design parameters, used to refer to the range of candidate design parameters participating in the initial robust parameter window boundary statistics; and the initial robust parameter window retains a traceable mapping relationship between the subset of optimal feasible candidate design parameters and the robust performance evaluation record, so as to use the window shrinking update in subsequent active sampling iterations.
[0036] S440. Perform a coverage check on the initial robust parameter window. The coverage check is used to verify the applicability of the initial robust parameter window within the critical operating condition range. The critical operating condition range is represented by a set of critical operating conditions, which consists of a preset number of operating condition identifiers. To make the coverage check quantifiable, a coverage rate is defined, and the coverage is used to determine whether the coverage meets the requirements. The coverage rate must at least satisfy the following formula: ; in, Coverage is used to characterize the degree to which the initial robust parameter window covers the set of critical operating conditions; The total number of operating conditions in the set of critical operating conditions; This represents the number of load cases in the set of critical load cases that can be covered by the initial robust parameter window, and "covered" must at least satisfy the following criterion: Under the corresponding load case identifier, there is at least one set of candidate design parameters in the initial robust parameter window that satisfies the constraint criterion label. ; When coverage When the coverage rate is less than the preset threshold, the initial robust parameter window is deemed to have insufficient coverage and backtracking adjustment is triggered. Backtracking adjustment includes at least: backtracking to the candidate design parameter set to expand the parameter region related to insufficient coverage and re-executing robust multi-objective solution to update the optimal feasible candidate design parameter subset, or backtracking to the design parameter space to expand the value range related to insufficient coverage and regenerating the initial robust parameter window, until the coverage verification passes.
[0037] S5 specifically includes the following sub-steps: S510. Construct an active sampling iterative environment. The active sampling iterative environment is used to generate sampling decisions under the constraints of the initial robust parameter window and the iterative robust parameter window. The state information of the active sampling iterative environment includes at least the evaluation confidence information, the distribution characteristics of the robust performance evaluation records, the boundary information of the iterative robust parameter window, and the coverage of the key operating condition set. Among them, the evaluation confidence information is the information generated in step S340 and bound to the unique identifier of the candidate design parameter. The iterative robust parameter window is the information generated in step S540 and used to replace the initial robust parameter window for continued iteration. The key operating condition set is the information used for coverage verification in step S440. The action of the active sampling iteration environment is defined as selecting the next batch of candidate design parameters as sampling objects within the current iteration robust parameter window. The sampling objects are represented by a list of unique identifiers for candidate design parameters, so that the action can directly trigger subsequent simulation evaluation and experimental evaluation and establish a one-to-one correspondence with the newly added evaluation results.
[0038] S520. Train a sampling decision policy based on an active sampling iterative environment. The sampling decision policy is implemented using reinforcement learning. Reinforcement learning enables the sampling decision policy to output the sampling object with the maximum information gain by selecting actions, obtaining rewards, and updating the policy in a given state. To make reinforcement learning reproducible, a reward value is defined and the action value function is updated accordingly. The reward value must at least satisfy the following formula: ; in, The reward value is used to evaluate the overall benefit of a single sampling decision. This is the credibility weight, used to adjust the proportion of average credibility in the reward value; This is the feasibility rate weight, used to adjust the proportion of feasibility rate in the reward value; This is the coverage weight, used to adjust the proportion of coverage in the reward value; The average confidence level represents the confidence level calculated for the newly added evaluation samples corresponding to this sampling decision. The average value of, where For credibility and in step S340 Maintain consistency; For feasibility and in step S240 Maintain consistency; For coverage and in step S440 To maintain consistency; furthermore, define the action value function and update it accordingly. The update of the action value function must at least satisfy the following formula: ; in, The action value function is used to characterize the state. Take action below Value; The state represents the state information of the actively sampled iterative environment; The action represents the sampling decision for selecting a sampling object within the iterative robust parameter window; The learning rate is used to adjust the update magnitude of the action value function; This is a discount factor used to adjust the proportion of future rewards in the current update; The successor state indicates the execution of an action. And complete the new state obtained after updating in steps S530 and S540; A candidate action represents the action in the successor state. The sampling decision is optional; and the sampling decision strategy is based on the current state. Output Action And actions A unique list of candidate design parameters for the corresponding sampling object, for consistent use in subsequent steps.
[0039] As a specific example of parameter settings, to balance the impact of credibility, feasibility, and coverage, the following settings can be configured: Furthermore, a learning rate can be set for the hyperparameters in the action-value function update formula. Discount factor The parameter values mentioned above are for illustrative purposes only, and technicians can adjust them according to the actual data volume and convergence speed requirements.
[0040] S530. Use the sampling decision strategy to select sampling objects from the iterative robust parameter window, and trigger simulation evaluation and experimental evaluation for the sampling objects to obtain new evaluation results. In order to improve feasibility and reduce unnecessary experimental costs, the simulation evaluation and experimental evaluation shall at least meet the following triggering rules: first perform simulation evaluation on the sampling objects to obtain initial evaluation results, and then perform experimental evaluation on the sampling objects whose evaluation confidence information indicates low confidence or which contributes significantly to coverage verification to obtain calibration evaluation results. The newly added evaluation results should include at least: unique identifiers of candidate design parameters, operating condition identifiers, and observed index values. and constraint determination labels ,in The observed index value is the same as that in step S340. Maintain consistency To constrain the determination label and in step S330 To maintain consistency, new evaluation results will be added to the standardized multi-source data using the unique identifier of the candidate design parameters as the primary key, and robust performance evaluation records will be updated simultaneously to ensure that subsequent model updates and window shrinkage maintain a traceable one-to-one correspondence.
[0041] S540. Based on the updated standardized multi-source data, manufacturing deviation description model and robust evaluation model, window shrinking is performed to generate an iterative robust parameter window; The window shrinking process includes at least the following steps: updating the manufacturing deviation description model and robust evaluation model using the newly added evaluation results supplemented in step S530, re-executing robust multi-objective solution based on the updated robust performance evaluation records to update the subset of optimal feasible candidate design parameters, and then recalculating and updating the window boundaries of each design parameter according to the boundary calculation rules of the initial robust parameter window in step S430, thereby obtaining the iterative robust parameter window. And the updated boundary information of the iterative robust parameter window, the updated evaluation confidence information, and the updated feasibility rate will be added. and the updated coverage The system is backfilled into the active sampling iterative environment to form a closed-loop iterative process of "reinforcement learning training - sampling object selection - simulation evaluation and experimental evaluation - model update - window shrinking - state backfilling", thereby providing a consistent basis for subsequent convergence determination and final robust parameter window output.
[0042] S6 specifically includes the following sub-steps: S610. Convergence determination is performed based on iterative robust parameter windows. The convergence determination includes at least the determination of the window boundary change magnitude, the determination of the confidence improvement magnitude, and the determination of feasibility and coverage. To make the convergence determination computable, the window boundary change magnitude is defined and used to characterize the degree of window boundary change between two adjacent iterations. The window boundary change magnitude must at least satisfy the following formula: ; in, The magnitude of the change in the window boundary; This is an iteration round index, used to refer to the 1st iteration round. The next iteration; This is a design parameter index, used to refer to the first parameter in the design parameter space. One design parameter; For the first The result of the iteration is the first The lower boundary of each design parameter in the iterative robust parameter window; For the first The result of the iteration is the first The upper boundary of each design parameter in the iterative robust parameter window; For the first The result of the iteration is the first The lower boundary of each design parameter in the iterative robust parameter window; For the first The result of the iteration is the first The design parameters are defined within the upper bound of the iterative robust parameter window; further, the confidence improvement magnitude is defined to characterize the average confidence change between two adjacent iterations, and the confidence improvement magnitude must satisfy at least the following formula: ; in, To increase credibility; For the first The average confidence level corresponding to each iteration represents the confidence level of the newly added evaluation samples in that iteration. The average value of, where For credibility and in step S340 Maintain consistency; For the first The average reliability corresponding to each iteration; and the magnitude of the window boundary change. Credibility improvement Feasibility and coverage Both serve as criteria for convergence determination, among which For feasibility and in step S240 Maintain consistency For coverage and in step S440 Maintain consistency; convergence is determined when the following comprehensive convergence condition is met: ; in, This is the window convergence threshold, used to limit the magnitude of changes in the window boundaries. Maximum allowed value; This is the credibility convergence threshold, used to limit the extent to which credibility can be increased. Maximum allowed value; The minimum feasibility threshold and in step S240 Maintain consistency; This is the minimum coverage threshold, used to limit coverage. The minimum allowed value.
[0043] S620. When the comprehensive convergence condition is met, determine the final robust parameter window and output the final robust parameter window as the design parameter determination result of the silicon carbide switch. The final robust parameter window should be output as at least the set of lower and upper boundaries of each design parameter, and the output should at least carry reference information of the corresponding key operating condition set, reference information of manufacturing deviation constraints, and the magnitude of the window boundary change when convergence is passed. Credibility improvement Feasibility With coverage The final value is determined to ensure that the design parameter determination results are reproducible and deployable.
[0044] S630. Generate a verification record. The verification record is used to verify the validity of the final robust parameter window. The verification record should include at least the final robust parameter window, the unique identifiers of the candidate design parameters of the verification samples extracted from the final robust parameter window, the corresponding load condition identifiers, the corresponding simulation evaluation results, the corresponding measured evaluation results, and the corresponding constraint judgment labels. ,in To constrain the determination label and in step S330 To maintain consistency, verification records must also include at least cross-condition retest results, which must cover at least the set of critical operating conditions, and be labeled with constraint decision tags. The statistical results are used as the basis for re-verification to ensure that the verification process and the manufacturing deviation constraints have the same judgment criteria.
[0045] S640. Generate traceable evaluation criteria, which are used to reproduce, audit, and deploy the design parameter determination results. The evaluation criteria include at least: a version of the manufacturing deviation description model and its input deviation representation version, a version of the robust evaluation model and the evaluation confidence information version, a version of the sampling decision strategy, and a learning rate used to train the sampling decision strategy. With discount factor ,in The learning rate and in step S520 Maintain consistency It is a discount factor and is consistent with step S520 Maintain consistency; The evaluation criteria should also include at least the final snapshot of the robust performance evaluation record, the final robust parameter window, and the convergence determination results. The results, along with the key operating condition set and coverage verification results, form a traceable closed loop covering the entire chain of "standardized multi-source data - model - strategy - window - verification".
[0046] The robust design parameter window determined by the above method can meet the needs of high-end integrated circuit applications that are sensitive to parameter fluctuations and noise, and is especially suitable for power supply and interface systems related to quantum integrated circuits that have high requirements for power supply stability and reliability.
[0047] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.
[0048] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.
[0049] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0050] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0051] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.
[0052] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0053] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.
[0054] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0055] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0056] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method of determining design parameters for a silicon carbide switch, comprising: Includes the following steps: S1. Collect multi-source data related to silicon carbide switches, process the multi-source data to obtain standardized multi-source data, extract evaluation elements based on the standardized multi-source data, establish a design parameter space, and generate a set of candidate design parameters with unique identifiers within the design parameter space. S2. Identify manufacturing deviation factors from standardized multi-source data and form deviation characterizations. Based on the deviation characterizations, construct a manufacturing deviation description model, generate manufacturing deviation constraints, establish a mapping relationship with the candidate design parameter set, and perform consistency verification on the manufacturing deviation constraints. S3. Establish a robust evaluation model with the candidate design parameter set, deviation characterization, manufacturing deviation constraints and evaluation elements as inputs, calculate the robust performance evaluation results corresponding one-to-one with the candidate design parameter set and form a robust performance evaluation record, and update the evaluation credibility information based on the robust performance evaluation record. S4. Based on robust performance evaluation records, determine the objective set and constraint set for robust multi-objective solution, screen the optimal feasible candidate design parameter subset, generate an initial robust parameter window, and perform coverage verification on the initial robust parameter window to obtain a usable initial robust parameter window, specifically including: Based on robust performance evaluation records, the objective set and constraint set for robust multi-objective solution are determined, and the constraint judgment label is used as the feasibility criterion. From the set of candidate design parameters, feasible candidate design parameters are selected and non-dominated sorting is used to obtain the optimal subset of feasible candidate design parameters; The lower and upper boundaries of each design parameter are obtained statistically based on the subset of the optimal feasible candidate design parameters to generate an initial robust parameter window; The coverage of the initial robust parameter window on the set of critical operating conditions is verified, and if the coverage is insufficient, a backtracking adjustment is triggered and the initial robust parameter window is updated. S5. Constructing an active sampling iterative environment employs a reinforcement learning-based sampling decision strategy. Within the initial robust parameter window or iterative robust parameter window, sampling objects are selected, and simulation and experimental evaluations are triggered to obtain new evaluation results. The manufacturing deviation description model and robust evaluation model are updated, and the initial robust parameter window is shrunk to generate an iterative robust parameter window, forming a closed-loop iteration. Specifically, this includes: Construct an active sampling iterative environment with evaluation credibility information, robust performance evaluation record distribution characteristics, iterative robust parameter window boundary information, and key operating condition set coverage as state information, and define the sampling objects as a list of sampling objects represented by unique identifiers of candidate design parameters; A reinforcement learning training sampling decision strategy is adopted to output sampling objects and trigger simulation evaluation and experimental evaluation to obtain new evaluation results. The new evaluation results are added to the standardized multi-source data and the robust performance evaluation records are updated synchronously. The manufacturing deviation description model and robust evaluation model are updated based on the updated standardized multi-source data, and the robust multi-objective solution is re-executed to shrink the initial robust parameter window to obtain the iterative robust parameter window; Simultaneously, the updated window boundary information, evaluation credibility information, feasibility rate, and coverage rate are backfilled into the active sampling iteration environment to form a closed-loop iteration.
2. A method of determining design parameters for a silicon carbide switch according to claim 1, wherein Also includes: S6. Perform convergence determination based on the iterative robust parameter window. When the convergence condition is met, output the final robust parameter window as the design parameter determination result, and generate verification records and traceable evaluation basis.
3. The method of claim 1, wherein: S1 specifically includes: Collect transient waveform data, temperature response data, operating condition data, and historical reliability statistics of the switch, and perform time alignment, data cleaning, and standardization on the multi-source data to obtain standardized multi-source data; Based on standardized multi-source data extraction efficiency-related evaluation elements, switch stress-related evaluation elements, and reliability-related evaluation elements, and establish corresponding relationships; Establish and associate a design parameter space that includes the value ranges of device structure design parameters, package parasitic design parameters, drive link design parameters, and thermal management design parameters; Generate a set of candidate design parameters that meet the coverage requirements within the design parameter space and generate a unique identifier.
4. The method of claim 1, wherein S2 specifically includes: Manufacturing deviation factors such as device discreteness, packaging parasitic differences, drive link differences, and temperature drift differences are identified from standardized multi-source data. These factors are then grouped by batch identifier, temperature range identifier, and operating condition identifier to form a deviation characterization. Based on the deviation characterization, a manufacturing deviation description model is constructed to obtain an estimate of the fluctuation range of each indicator under the influence of manufacturing deviation factors. Manufacturing deviation constraints are generated based on fluctuation range estimation and a mapping relationship is established with the candidate design parameter set. Perform boundary conflict checks and feasibility checks on manufacturing deviation constraints, and trigger backtracking adjustments when feasibility is insufficient.
5. The method of claim 1, wherein: S3 specifically includes: A robust evaluation model is established with candidate design parameter values, deviation characterization, manufacturing deviation constraints and evaluation elements as inputs. The robust performance evaluation results of efficiency-related indicators, switching stress-related indicators and reliability-related indicators are calculated and correspond one-to-one with the unique identifiers of candidate design parameters. Organize the robust performance evaluation results into robust performance evaluation records that include index statistics, lower bound estimates, upper bound estimates, and constraint decision labels; The evaluation credibility information is updated based on the deviation between the observed index values and the predicted index values to characterize the reliability of the robust evaluation model in different candidate design parameter ranges.
6. The method of claim 2, wherein: S6 specifically includes: Convergence determination is based on iterative robust parameter window execution. The convergence determination includes at least a comprehensive determination of the window boundary change magnitude, confidence improvement magnitude, feasibility rate, and coverage rate. When the comprehensive convergence condition is met, the final robust parameter window is determined and output as the design parameter determination result, and a verification record containing the simulation evaluation results, measured evaluation results and cross-condition verification results corresponding to the final robust parameter window is generated, and traceable evaluation basis is generated at the same time.
7. A method of determining design parameters for a silicon carbide switch as recited in claim 6, wherein, The evaluation criteria should include at least the manufacturing deviation description model version, the robust evaluation model version, the sampling decision strategy version, the final snapshot of the robust performance evaluation record, the convergence judgment results, and the coverage verification results of the key operating condition set, so as to reproduce, audit, and deploy the design parameter determination results.
Citation Information
Patent Citations
Design and simulation method of two-dimensional semiconductor nanometer transistor
CN120124502A
General logic device optimization method based on machine learning
CN120409421A
Steel production process window optimization method and system based on full-process fluctuation hedging
CN121481463A
