SPICE automation device modeling method and system based on model parameter analysis
By using the SPICE automated device modeling method and constructing an automated process using model parameter analysis, the problem of time-consuming and error-prone device model extraction in existing technologies is solved, and efficient and accurate model extraction is achieved, which is suitable for complex integrated circuit design.
Patent Information
- Application Number
- CN202510857144.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-25
- Publication Date
- 2025-09-26
AI Technical Summary
The device model extraction method in the existing technology relies on manual operation, which is time-consuming and error-prone, making it difficult to meet the needs of complex integrated circuit design, resulting in low model accuracy and efficiency.
The SPICE automated device modeling method based on model parameter analysis is adopted. Through fitting target analysis, parameter screening and optimization, an automated modeling process is constructed to reduce manual intervention and improve model extraction efficiency and accuracy.
It achieves efficient and accurate device model extraction, reduces human errors, improves the adaptability and generalization ability of the model, and is suitable for complex integrated circuit design.
Smart Images
Figure CN120706346A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of semiconductor device modeling, and in particular to a SPICE automated device modeling method and system based on model parameter analysis. Background Art
[0002] Device models play a crucial role in integrated circuit design. They are primarily used to characterize the electrical characteristics of process devices within an IC, such as voltage-current relationships and capacitance-voltage relationships. These characteristics are crucial for circuit design and simulation. As the underlying unit in an IC, the performance of the device directly determines the functionality, performance, and reliability of the overall circuit. If the device model differs significantly from the actual device characteristics, the circuit yield will be severely affected during the final tape-out process, potentially leading to substandard product performance, excessive power consumption, or poor reliability, further increasing R&D costs and market risks.
[0003] Device models are traditionally extracted manually by engineers within an electronic design automation (EDA) tool environment, following the main model extraction workflow. SPICE (Simulation Program with Integrated Circuit Emphasis) is a circuit simulation tool widely used in the electronics design field, designed to verify integrated circuit designs and predict circuit performance. By simulating circuit behavior, SPICE helps engineers predict circuit performance before production begins, and is a core component of electronic design automation (EDA) software suites. This approach requires engineers to have a deep understanding of device physical properties and model parameters, and involves extensive iteration to ensure model accuracy and reliability. In some cases, semi-automatic methods such as selecting targets are incorporated to improve extraction efficiency. However, this manual model extraction approach has several limitations. First, it requires extensive manual parameter iteration, which is labor-intensive and time-consuming, and prone to errors due to human error. Second, even with the incorporation of some semi-automatic optimization methods, it cannot implement complex process strategies, making it difficult to meet the increasingly complex demands of integrated circuit design.
[0004] Taking MOS device BIN model extraction as an example, engineers must draw on years of accumulated experience to perform repeated model extraction for devices of varying sizes. This process involves analyzing large amounts of data, adjusting parameters, and validating the model. Each step requires precise operation and rigorous verification. Ultimately, the synthesized BIN model can be used for circuit design and simulation. Typically, a complete model extraction process takes weeks to months to complete, which is undoubtedly a huge challenge in the rapidly iterating integrated circuit market.
[0005] Therefore, efficiently extracting models while ensuring the reliability and validity of the extracted parameters is crucial. With the continuous advancement of integrated circuit technology, higher requirements are being placed on the accuracy and efficiency of device models. This necessitates the development of more intelligent and automated model extraction tools and methods to reduce manual intervention and improve extraction speed and accuracy. Summary of the Invention
[0006] The purpose of the present invention is to address the shortcomings of the above-mentioned prior art and to provide a SPICE automated device modeling method and system based on model parameter analysis. Model parameter analysis is applied to the model extraction automation process to improve the efficiency of model extraction while reducing model quality problems caused by over-optimization, under-optimization, etc.
[0007] On the one hand, a SPICE automated device modeling method based on model parameter analysis is provided, comprising the following steps: S1: Select fitting targets, including single targets and composite targets, based on the specific modeling task and expected results, and set fitting criteria to evaluate and adjust the modeling process; S2: Analyze the complexity of the fitting target and select a fitting method based on the complexity analysis result, construct a process extraction strategy for the fitting target and specific implementation steps of the strategy based on the fitting method, extract the fitting target in each step, and construct a physical model parameter set based on the parameter set corresponding to the fitting target; S3: Analyze the behavior characteristics of each parameter in the model parameter set, select physical model parameters suitable for the optimization target, and call the corresponding optimizer to perform fitting optimization on the selected physical model parameters; S4: Detecting the optimized model fitting results based on the fitting criteria, and controlling and adjusting the entire modeling process according to the detection results.
[0008] Furthermore, in step S2, analyzing the complexity of the fitting target and selecting a fitting method according to the complexity analysis result further includes: A quantitative evaluation system for fitting target complexity is constructed based on data characteristics and physical constraints. The core indicators of the evaluation system include data distribution characteristics. , relationship nonlinearity and signal-to-noise ratio , comprehensively evaluate the complexity of fitting targets through multi-dimensional indicators , which is expressed as follows: in, is the empirical threshold set based on prior knowledge, is the dimension of the input variable, Indicates that variables are independent, non-redundant, or strongly coupled. is the preset high noise threshold; The fitting method is flexibly selected according to the complexity analysis result, and the fitting method includes a single-step fitting method and a multi-step fitting method, wherein, If the fitting target is a single target and the fitting target complexity is , the single-step fitting method is preferred; If the fitting target is a composite target and the fitting target complexity is , a multi-step fitting method is preferred.
[0009] The steps of implementing the single-step fitting method include: S201: Collect fitting data from multiple data sources, including experimental measurement data, simulation data, and existing databases, clean the collected data, remove outliers and erroneous data, and standardize and normalize the cleaned data to improve the stability and convergence speed of subsequent fitting algorithms; S202: Select an appropriate fitting model based on the characteristics of the data and the physical and mathematical nature of the fitting target, analyze the data, and determine the key fitting areas that need to be optimized. At the same time, for fitting situations where physical quantities have dimensional concepts, clarify the relevant dimensional parameters and their ranges.
[0010] Preferably, the implementation steps of the multi-step fitting method include: S211: Collect fitting data from multiple data sources, organize and classify the collected data so as to allocate them to different fitting sub-steps, and perform the same data cleaning, standardization and normalization operations as step S201; S212: Divide the entire fitting process into multiple sub-steps based on the complexity of the fitting target and the characteristics of the data. Each sub-step has a clear physical and mathematical meaning, and the transitions between them are clear. S213: Selecting a suitable fitting model for each sub-step based on the data characteristics and physical mathematical principles, determining an optimization region for each sub-step based on the data distribution and physical characteristics, and determining corresponding size parameters and their ranges for each sub-step; S214: In multi-step fitting, different sub-steps are interconnected through parameter transfer and coordination. After completing the fitting of all sub-steps, the entire fitting result is evaluated as a whole.
[0011] Furthermore, in step S2, extracting the fitting targets in each step and constructing a physical model parameter set based on the parameter set corresponding to the fitting targets includes: If the single-step fitting method is used, the physical parameter is a single parameter. If the multi-step fitting method is used, the physical parameter is multiple parameters. Based on the fitting data target, the corresponding physical model and its parameter set are determined, and the parameters corresponding to the fitting target of each sub-step in the implementation step are refined into specific parameters of the physical model. The interaction between these parameters jointly determines the shape and characteristics of the fitting target.
[0012] Furthermore, in step S3, the screening of physical model parameters suitable for the optimization target includes: The physical model parameters used for fitting the target usually include multiple parameters. For the set physical model parameter set, the contribution of each parameter to the fitting data target is analyzed and ranked in descending order according to the contribution. A parameter screening threshold is set, and parameters with a contribution ranking above the parameter screening threshold are screened out for subsequent optimization targets.
[0013] Furthermore, in step S1, selecting the fitting data target according to the specific modeling task and expected results includes: Based on the modeling task and expected results, a fitting target is selected according to the device type of the fitting data. The objects selected for the fitting target include the scanning curve that characterizes the device characteristics and the key indicators that characterize the device characteristics. The fitting target can be a single target or a composite target, wherein: Single target is to select a single scanning curve or a single key indicator that characterizes the characteristics of the device as the fitting target; The composite target is to simultaneously select multiple scanning curves characterizing device characteristics or multiple key indicators characterizing device characteristics as fitting targets, or to combine scanning curves characterizing device characteristics and key indicators characterizing device characteristics as fitting targets.
[0014] The device types of the fitting data include MOS devices, bipolar junction transistors (BJTs), diodes, resistors, capacitors, and inductors.
[0015] Furthermore, in step S1, the fitting standard is set according to the selected fitting target, wherein: For the scanning curve that characterizes the device characteristics, the fitting criteria adopted include: Set the mean square error (MSE) range of the model fitting the scanning curve to ensure the model fits the scanning curve that characterizes the device characteristics; Set the maximum error range of the model fitting to the scanning curve to ensure the fitting stability of the model at different data points; For the key indicators that characterize device characteristics, the fitting standards adopted include: Set the error range of the linear threshold voltage Vtlin to ensure the accuracy of the performance analysis of the characterization device and the related circuit design; The error range of the linear region drain current Idlin is set to ensure the accuracy when describing and predicting the linear region characteristics of the device.
[0016] Furthermore, in step S3, calling a corresponding optimizer to perform fitting optimization on the selected physical model parameters includes: Arrange the data into a format supported by the optimizer, store the independent variable and dependent variable data in matrices respectively, substitute the data required for optimization into the optimizer, and set initial values for the parameters to be optimized based on the physical model and prior knowledge of the problem; The objective function is constructed based on the fitting target error, and the fitting optimization process is performed. After the optimization is completed, the optimized parameter values are extracted from the optimization results.
[0017] In another aspect, a SPICE automated device modeling system based on model parameter analysis is provided, comprising: The target and standard setting module is used to select fitting targets, including single targets and composite targets, based on specific modeling tasks and expected results, and to set fitting standards to evaluate and adjust the modeling process; A strategy and parameter construction module is used to analyze the complexity of the fitting target and select a fitting method based on the complexity analysis results, construct a process extraction strategy for the fitting target based on the fitting method and the specific implementation steps of the strategy, extract the fitting target in each step, and construct a physical model parameter set based on the parameter set corresponding to the fitting target; The parameter analysis and optimization module is used to analyze the behavior characteristics of each parameter in the model parameter set, select physical model parameters suitable for the optimization target, and call the corresponding optimizer to perform fitting optimization on the selected physical model parameters; The result detection and control module is used to detect the optimized model fitting results based on the fitting standard, and control and adjust the entire modeling process according to the detection results.
[0018] Compared with the prior art, the present invention has the following beneficial effects: From selecting fitting data targets and setting fitting standards to constructing process extraction strategies and implementation steps, the present invention reduces manual participation through automated operation processes, thus avoiding duplication of work and errors caused by human factors. This paper applies a parameter screening method based on model parameter analysis to process modeling, conducts in-depth analysis of the behavior of each parameter, accurately screens out parameters that have a significant impact on the optimization target, and performs fitting optimization on them. At the same time, by scientifically and rationally selecting fitting data targets and setting clear fitting standards, it ensures that the data and targets used in the modeling process are highly representative and accurate. The present invention not only supports the fitting of a single target, but also can select multiple scanning curves or multiple key indicators at the same time, and comprehensively consider multiple fitting targets. This flexibility enables the model to comprehensively consider multiple factors during the construction process, avoiding model deviation caused by focusing only on a single target, so that it can maintain good performance in different scenarios and conditions. At the same time, for complex modeling tasks, traditional manual modeling methods are often difficult to cope with. The automated modeling method of the present invention, by constructing detailed process extraction strategies and implementation steps, can systematically handle complex modeling processes, ensuring that when faced with multi-step, multi-parameter modeling tasks, the model construction can still be completed efficiently and accurately, further improving the model's adaptability and generalization capabilities to complex systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 A flow chart of a SPICE automated device modeling method based on model parameter analysis according to the present invention; Figure 2 This is an example diagram of a process extraction strategy for constructing a fitting target and its implementation steps of the present invention; Figure 3 The figure is a flow chart of a screening mechanism for a physical model parameter set according to the present invention. DETAILED DESCRIPTION
[0020] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0021] The present invention is an automated modeling method based on model parameter analysis, which applies model parameter analysis to the model extraction automation process to improve the efficiency of model extraction while reducing model quality problems caused by over-optimization, under-optimization, etc.
[0022] The specific implementation of the present invention is described below with reference to the accompanying drawings and embodiments.
[0023] Example 1 See also Figure 1 , a technical solution of a SPICE automated device modeling method based on model parameter analysis provided in this embodiment includes the following steps: S1: Select fitting targets, including single targets and composite targets, based on the specific modeling task and expected results, and set fitting criteria to evaluate and adjust the modeling process; S2: Analyze the complexity of the fitting target and select a fitting method based on the complexity analysis result, construct a process extraction strategy for the fitting target and specific implementation steps of the strategy based on the fitting method, extract the fitting target in each step, and construct a physical model parameter set based on the parameter set corresponding to the fitting target; S3: Analyze the behavior characteristics of each parameter in the model parameter set, select physical model parameters suitable for the optimization target, and call the corresponding optimizer to perform fitting optimization on the selected physical model parameters; S4: Detecting the optimized model fitting results based on the fitting criteria, and controlling and adjusting the entire modeling process according to the detection results.
[0024] First, we proceed to step S1 to select the fitting data target based on the specific modeling task and expected results, including: The fitting target is selected according to the device type of the fitting data. The selection objects of the fitting data target include the scanning curve characterizing the device characteristics and the key indicators characterizing the device characteristics. The fitting target selects a single target or a composite target, wherein, Single target is to select a single scanning curve or a single key indicator that characterizes the characteristics of the device as the fitting target; The composite target is to simultaneously select multiple scanning curves characterizing device characteristics or multiple key indicators characterizing device characteristics as fitting targets, or to combine scanning curves characterizing device characteristics and key indicators characterizing device characteristics as fitting targets.
[0025] The device types of the fitting data include MOS devices, bipolar junction transistors (BJTs), diodes, resistors, capacitors, and inductors.
[0026] In this embodiment, scanning curves representing device characteristics include the MOS device's drain current vs. gate-source voltage curve Ids_vgs and drain current vs. drain-source voltage curve ids_vds, and key indicators representing device characteristics include the MOS device's linear region threshold voltage Vtlin and linear region drain current Idlin. A single target refers to fitting a single scanning curve (e.g., selecting only the Ids_vgs curve) or a key indicator (e.g., selecting only Vtlin) as the target; a composite target refers to simultaneously selecting multiple scanning curves or multiple key indicators, or a combination of scanning curves and key indicators as the fitting target.
[0027] Then, we set the fitting standard according to the selected fitting data target, where: For the scanning curve that characterizes the device characteristics, the fitting criteria adopted include: Set the mean square error (MSE) range of the model fitting the scanning curve to ensure the model fits the scanning curve that characterizes the device characteristics; Set the maximum error range of the model fitting to the scanning curve to ensure the fitting stability of the model at different data points; For the key indicators that characterize device characteristics, the fitting standards adopted include: Set the error range of the linear threshold voltage Vtlin to ensure the accuracy of the performance analysis of the characterization device and the related circuit design; The error range of the linear region drain current Idlin is set to ensure the accuracy when describing and predicting the linear region characteristics of the device.
[0028] Specifically, in this embodiment, we select the Ids_vgs curve of the MOS device as the fitting target, and set the mean square error to be less than 2%, and the maximum error to be less than 3%; select the key indicator seat of the MOS device as the fitting target, and set the error of Vtlin to be less than 5mv, the error of Idlin to be less than 3%, etc.
[0029] Next, the complexity of the fitting target is analyzed in step S2 and a fitting method is selected according to the complexity analysis result, which specifically includes: A quantitative evaluation system for fitting target complexity is constructed based on data characteristics and physical constraints. The core indicators of the evaluation system include data distribution characteristics. , relationship nonlinearity and signal-to-noise ratio , comprehensively evaluate the complexity of fitting targets through multi-dimensional indicators , which is expressed as follows: in, is the empirical threshold set based on prior knowledge, is the dimension of the input variable, Indicates that variables are independent, non-redundant, or strongly coupled. is the preset high noise threshold; The fitting method is flexibly selected according to the complexity analysis result, and the fitting method includes a single-step fitting method and a multi-step fitting method, wherein, If the fitting target is a single target and the fitting target complexity is , the single-step fitting method is preferred; If the fitting target is a composite target and the fitting target complexity is , a multi-step fitting method is preferred.
[0030] Specifically, if the fitting target is a single target and the corresponding physical parameter is a single parameter, the process strategy can be a single step; if the fitting target is multiple targets and the corresponding physical parameter is multiple parameters, the strategy can be split and implemented. The above two scenarios are only examples to illustrate the process extraction strategy and implementation steps, but are not limited to the above two scenarios. Then, based on the fitting data target, the corresponding physical model and its parameter set are determined, and the parameters corresponding to the fitting target of each sub-step in the implementation step are refined into specific parameters of the physical model. The interaction between these parameters jointly determines the shape and characteristics of the fitting target.
[0031] The steps of implementing the single-step fitting method include: S201: Collect fitting data from multiple data sources, including experimental measurement data, simulation data, and existing databases, clean the collected data, remove outliers and erroneous data, and standardize and normalize the cleaned data to improve the stability and convergence speed of subsequent fitting algorithms; S202: Select an appropriate fitting model based on the characteristics of the data and the physical and mathematical nature of the fitting target, analyze the data, and determine the key fitting areas that need to be optimized. At the same time, for fitting situations where physical quantities have dimensional concepts, clarify the relevant dimensional parameters and their ranges.
[0032] The implementation steps of the multi-step fitting method include: S211: Collect fitting data from multiple data sources, organize and classify the collected data so as to allocate them to different fitting sub-steps, and perform the same data cleaning, standardization and normalization operations as step S201; S212: Divide the entire fitting process into multiple sub-steps based on the complexity of the fitting target and the characteristics of the data. Each sub-step has a clear physical and mathematical meaning, and the transitions between them are clear. S213: Selecting a suitable fitting model for each sub-step based on the data characteristics and physical mathematical principles, determining an optimization region for each sub-step based on the data distribution and physical characteristics, and determining corresponding size parameters and their ranges for each sub-step; S214: In multi-step fitting, different sub-steps are interconnected through parameter transfer and coordination. After completing the fitting of all sub-steps, the entire fitting result is evaluated as a whole.
[0033] In this embodiment, the Ids_vgs@vb0 characteristic curve for a MOS with a low Vds voltage is used as an example. To better fit this characteristic curve, the curve can be divided into two characteristic regions, each separated by the threshold voltage vth of the key output indicator. The first region is the subthreshold region, which extends from the subthreshold starting point to the key output indicator vth; the second region is the linear region, which extends from the key output indicator vth to the maximum vg.
[0034] For the implementation steps of the strategy, please refer to Figure 2 As shown: Step S210 first fits the first subthreshold region. Next, step S220 fits the second linear region. To improve the fitting, in addition to outputting Ids, gm is added as an optimization target. gm is the differential curve of Ids with respect to Vgs, which further refines the trend of Ids versus Vgs. Next, step S230 uses the subthreshold and linear regions as fitting targets to improve the overall fitting. The key output metric vth here can be vtlin, vtgm, or a combination thereof to adapt to different scenarios.
[0035] In the specific process extraction strategy and implementation method, the sub-step configuration can be specifically divided into: optimization target, parameter set, optimizer call, etc. In this embodiment, the optimization target is the fitting target in each strategy sub-step in step S2, which can be a characteristic curve or a key output indicator. If it is a characteristic curve, regularized target selection can be performed based on the characteristic points of the graph. For example, when selecting the MOS subthreshold area, a characteristic selection can be performed from the starting point of the subthreshold area to vth to achieve a universal setting. If the optimization target is a key output indicator, the key output indicator can be selected as the optimization target during optimization by pre-defining relevant key indicators such as vtlin, idlin, and idsat.
[0036] In step S3, the physical model parameters used to fit the target typically include multiple parameters. For example, the Bsim4.5.0 model contains hundreds of parameters for fitting the characteristic curve. During the process strategy implementation step, the parameter set selection criteria are based on the parameters that affect the optimization target. For example, when fitting the subthreshold region of the ids_vgs@vb0 curve at low Vds voltage in the Bsim4.5.0 MOS circuit, vth0, nfactor, and cit can be selected as the parameter set. For example, when fitting the global model of the linear region of the ids_vgs@vb0 curve at low Vds voltage, the selected parameter set may include u0, ua, ub, ud, up, lp, and their bin items, such as lu0, lua, lub, lud, wu0, wua, wub, wud, pu0, pua, pub, and pud.
[0037] Then analyze the behavior of each parameter in the model parameter set and select the physical model parameters that are suitable for the optimization goal, including: For the set physical model parameter set, the contribution of each parameter to the fitting data target is analyzed and ranked in descending order according to the contribution. The parameter screening threshold is set, and the parameters with contribution ranking above the parameter screening threshold are screened out for subsequent optimization targets.
[0038] Specifically, the goal of model parameter extraction is to maximize fitting accuracy while using as few parameters as possible to ensure model physicality and avoid potential simulation risks. However, to maximize the versatility of the modeling process and its applicability to the widest possible range of data, the model parameter set used for fitting typically needs to cover most of the necessary relevant parameters. For a specific data fit, this parameter set may contain redundant parameters for that data fit. These redundant parameters pose the following risks: 1. Increased model complexity, making subsequent model maintenance difficult; 2. Increased model parameters, which in turn increases the optimizer's optimization workload and slows down optimization; 3. Increased model complexity can lead to the risk of overfitting, resulting in a mismatch between the model and the actual silicon data. Based on these considerations, it is essential to analyze the behavior of each parameter in the model parameter set to select the physical model parameters that are appropriate for the optimization objective for model extraction.
[0039] Methods for analyzing the behavior of each parameter in the model parameter set and selecting physical model parameters suitable for the optimization objective can include the following: S310: Matching parameters to the fitting objective; S320: Optimizing primary parameters. When primary parameters do not meet the fitting requirements, additional parameters are added to further improve optimization quality; and S330: Parameter fitting contribution analysis. In S310, matching parameters to the fitting objective is performed. For ease of understanding, the Bsim4.5.0 global model is used as an example. The fitting objective is the linear region of the ids_vgs@vb0 curve at low Vds voltage. The initial physical parameter set is: u0, ua, ub, ud, up, lp, and their bin items, such as lu0, lua, lub, lud, wu0, wua, wub, wud, pu0, pua, pub, pud. If the device being fitted is a single, large device, the parameter set can be reduced to four parameters: u0, ua, ub, and ud. If the target device being fitted includes multiple devices of varying lengths within a wide width, the parameter set can be reduced to u0, ua, ub, ud, up, lp, and their L terms, such as lu0, lua, lub, and lud. In S320, when the main parameters do not meet the fitting requirements, additional parameters are added to further improve the optimization quality. For example, when fitting the linear region of the ids_vgs@vb0 curve for a single, large device at low Vds voltage using the Bsim4.5.0 global model, the initial physical parameter set is: u0, ua, ub, and ud. u0, ua, and ub are the main parameters, and ud is an additional parameter. During optimization, the primary parameters u0, ua, and ub are prioritized. If the optimization result meets the accuracy requirements, the next strategy sub-step is performed without further optimization and extraction of ud. If the primary parameter optimization result does not meet the accuracy requirements, the additional parameter ud is added for further optimization and improvement. Parameter fitting contribution analysis in S330: For example, using the Bsim4.5.0 global model to fit the linear region of the ids_vgs@vb0 curve for a single large device at low Vds voltage, the initial physical parameter set is u0, ua, and ub. The contribution of each parameter to the fitting target is examined from the initial value to the optimized value. For example, if the parameter change before and after optimization is 200%, but the improvement in the fitting target error is less than the assumed threshold of 0.01%, it can be assumed that ua has limited improvement in the fitting target optimization, and that significant use of ua can improve fitting accuracy.
[0040] Figure 3 The exemplary methods for parameter screening are only for illustrating the embodiments of step S3, and the actual usage is not limited to these.
[0041] Then call the corresponding optimizer to perform fitting optimization on the selected physical model parameters, including: Arrange the data into a format supported by the optimizer, store the independent variable and dependent variable data in matrices respectively, substitute the data required for optimization into the optimizer, and set initial values for the parameters to be optimized based on the physical model and prior knowledge of the problem; The objective function is constructed based on the fitting target error, and the fitting optimization process is performed. After the optimization is completed, the optimized parameter values are extracted from the optimization results.
[0042] Finally, the model fitting result of step S4 is detected and the process is controlled.
[0043] The optimized fitting results in S3 are tested. The test results can be logically jumped through process control to achieve flexible strategy implementation. For example, if the target object is fitted with u0, ua, and ub and the fitting result meets the set requirements, the next strategy sub-step can be entered through process control; if the fitting result cannot meet the set requirements, the set accuracy repair and improvement sub-step can be entered through process control to improve the fitting result.
[0044] This embodiment also provides a SPICE automated device modeling system based on model parameter analysis, including: The target and standard setting module is used to select fitting targets, including single targets and composite targets, based on specific modeling tasks and expected results, and to set fitting standards to evaluate and adjust the modeling process; A strategy and parameter construction module is used to analyze the complexity of the fitting target and select a fitting method based on the complexity analysis results, construct a process extraction strategy for the fitting target based on the fitting method and the specific implementation steps of the strategy, extract the fitting target in each step, and construct a physical model parameter set based on the parameter set corresponding to the fitting target; The parameter analysis and optimization module is used to analyze the behavior characteristics of each parameter in the model parameter set, select physical model parameters suitable for the optimization target, and call the corresponding optimizer to perform fitting optimization on the selected physical model parameters; The result detection and control module is used to detect the optimized model fitting results based on the fitting standard, and control and adjust the entire modeling process according to the detection results.
[0045] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that improvements and modifications that do not depart from the principles of the present invention, which are apparent to those skilled in the art, should also be considered within the scope of protection of the present invention.
[0046] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A SPICE automated device modeling method based on model parameter analysis, characterized in that: The steps include: S1: Select fitting targets, including single targets and composite targets, based on the specific modeling task and expected results, and set fitting criteria to evaluate and adjust the modeling process; S2: Analyze the complexity of the fitting target and select a fitting method based on the complexity analysis result, construct a process extraction strategy for the fitting target and specific implementation steps of the strategy based on the fitting method, extract the fitting target in each step, and construct a physical model parameter set based on the parameter set corresponding to the fitting target; S3: Analyze the behavior characteristics of each parameter in the model parameter set, select physical model parameters suitable for the optimization target, and call the corresponding optimizer to perform fitting optimization on the selected physical model parameters; S4: Detecting the optimized model fitting results based on the fitting criteria, and controlling and adjusting the entire modeling process according to the detection results.
2. The SPICE automated device modeling method based on model parameter analysis according to claim 1, characterized in that: In step S2, analyzing the complexity of the fitting target and selecting a fitting method according to the complexity analysis result further includes: A quantitative evaluation system for fitting target complexity is constructed based on data characteristics and physical constraints. The core indicators of the evaluation system include data distribution characteristics. , relationship nonlinearity and signal-to-noise ratio , comprehensively evaluate the complexity of fitting targets through multi-dimensional indicators , which is expressed as follows: in, is the empirical threshold set based on prior knowledge, is the dimension of the input variable, Indicates that variables are independent, non-redundant, or strongly coupled. is the preset high noise threshold; The fitting method is flexibly selected according to the complexity analysis result, and the fitting method includes a single-step fitting method and a multi-step fitting method, wherein, If the fitting target is a single target and the fitting target complexity is , the single-step fitting method is preferred; If the fitting target is a composite target and the fitting target complexity is , a multi-step fitting method is preferred.
3. The SPICE automated device modeling method based on model parameter analysis according to claim 2, characterized in that: The implementation steps of the single-step fitting method include: S201: Collect fitting data from multiple data sources, including experimental measurement data, simulation data, and existing databases, clean the collected data, remove outliers and erroneous data, and standardize and normalize the cleaned data to improve the stability and convergence speed of subsequent fitting algorithms; S202: Select an appropriate fitting model based on the characteristics of the data and the physical and mathematical nature of the fitting target, analyze the data, and determine the key fitting areas that need to be optimized. At the same time, for fitting situations where physical quantities have dimensional concepts, clarify the relevant dimensional parameters and their ranges.
4. The SPICE automated device modeling method based on model parameter analysis according to claim 3, characterized in that: The implementation steps of the multi-step fitting method include: S211: Collect fitting data from multiple data sources, organize and classify the collected data so as to allocate them to different fitting sub-steps, and perform the same data cleaning, standardization and normalization operations as step S201; S212: Divide the entire fitting process into multiple sub-steps based on the complexity of the fitting target and the characteristics of the data. Each sub-step has a clear physical and mathematical meaning, and the transitions between them are clear. S213: Selecting a suitable fitting model for each sub-step based on the data characteristics and physical mathematical principles, determining an optimization region for each sub-step based on the data distribution and physical characteristics, and determining corresponding size parameters and their ranges for each sub-step; S214: In multi-step fitting, different sub-steps are interconnected through parameter transfer and coordination. After completing the fitting of all sub-steps, the entire fitting result is evaluated as a whole.
5. The SPICE automated device modeling method based on model parameter analysis according to claim 4, characterized in that: In step S2, extracting the fitting targets in each step and constructing the physical model parameter set based on the parameter set corresponding to the fitting targets further includes: If a single-step fitting method is used, the physical parameter is a single parameter. If a multi-step fitting method is used, the physical parameter is multiple parameters. Based on the fitting data target, the corresponding physical model and its parameter set are determined, and the parameters corresponding to the fitting target of each sub-step in the implementation step are refined into specific parameters of the physical model. The interaction between these parameters jointly determines the shape and characteristics of the fitting target.
6. The SPICE automated device modeling method based on model parameter analysis according to claim 5, characterized in that: In step S3, the screening of physical model parameters suitable for the optimization target further includes: The physical model parameters used for fitting the target usually include multiple parameters. For the set physical model parameter set, the contribution of each parameter to the fitting data target is analyzed and ranked in descending order according to the contribution. A parameter screening threshold is set, and parameters with a contribution ranking above the parameter screening threshold are screened out for subsequent optimization targets.
7. The SPICE automated device modeling method based on model parameter analysis according to claim 1, characterized in that: In step S1, selecting a fitting data target according to a specific modeling task and expected results further includes: Based on the modeling task and expected results, a fitting target is selected according to the device type of the fitting data. The objects selected for the fitting target include the scanning curve that characterizes the device characteristics and the key indicators that characterize the device characteristics. The fitting target can be a single target or a composite target, wherein: Single target is to select a single scanning curve or a single key indicator that characterizes the characteristics of the device as the fitting target; The composite target is to simultaneously select multiple scanning curves characterizing device characteristics or multiple key indicators characterizing device characteristics as fitting targets, or to combine scanning curves characterizing device characteristics and key indicators characterizing device characteristics as fitting targets.
8. The SPICE automated device modeling method based on model parameter analysis according to claim 7, characterized in that: The device types of the fitting data include MOS devices, bipolar junction transistors (BJTs), diodes, resistors, capacitors, and inductors.
9. The SPICE automated device modeling method based on model parameter analysis according to claim 2, characterized in that: In step S1, the fitting standard is set according to the selected fitting target, wherein: For the scanning curve that characterizes the device characteristics, the fitting criteria adopted include: Set the mean square error (MSE) range of the model fitting the scanning curve to ensure the model fits the scanning curve that characterizes the device characteristics; Set the maximum error range of the model fitting to the scanning curve to ensure the fitting stability of the model at different data points; For the key indicators that characterize device characteristics, the fitting standards adopted include: Set the error range of the linear threshold voltage Vtlin to ensure the accuracy of the performance analysis of the characterization device and the related circuit design; The error range of the linear region drain current Idlin is set to ensure the accuracy when describing and predicting the linear region characteristics of the device.
10. The SPICE automated device modeling method based on model parameter analysis according to claim 1, characterized in that: In step S3, calling the corresponding optimizer to perform fitting optimization on the selected physical model parameters further includes: Arrange the data into a format supported by the optimizer, store the independent variable and dependent variable data in matrices respectively, substitute the data required for optimization into the optimizer, and set initial values for the parameters to be optimized based on the physical model and prior knowledge of the problem; The objective function is constructed based on the fitting target error, and the fitting optimization process is performed. After the optimization is completed, the optimized parameter values are extracted from the optimization results.
11. A SPICE automated device modeling system based on model parameter analysis, characterized in that: include: The target and standard setting module is used to select fitting targets, including single targets and composite targets, based on specific modeling tasks and expected results, and to set fitting standards to evaluate and adjust the modeling process; A strategy and parameter construction module is used to analyze the complexity of the fitting target and select a fitting method based on the complexity analysis results, construct a process extraction strategy for the fitting target based on the fitting method and the specific implementation steps of the strategy, extract the fitting target in each step, and construct a physical model parameter set based on the parameter set corresponding to the fitting target; The parameter analysis and optimization module is used to analyze the behavior characteristics of each parameter in the model parameter set, select physical model parameters suitable for the optimization target, and call the corresponding optimizer to perform fitting optimization on the selected physical model parameters; The result detection and control module is used to detect the optimized model fitting results based on the fitting standard, and control and adjust the entire modeling process according to the detection results.