Calibration methods and related devices for computer-aided design technology models

By constructing a TCAD device model and optimizing parameters using random forest and Gaussian process regression models, the accuracy and efficiency issues in TCAD model calibration were resolved, achieving rapid and high-precision model calibration and supporting radiation effect research.

CN122310971APending Publication Date: 2026-06-30NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610422157.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-01
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Traditional TCAD model calibration methods suffer from low accuracy, low efficiency, slow convergence, and poor multi-objective optimization performance, making it difficult to achieve fast and high-precision model calibration.

Method used

By constructing a TCAD device model, determining calibration parameters, defining the effective parameter domain using a random forest model, optimizing the optimal parameter set using a Gaussian process regression model, and employing adaptive grid partitioning and dynamic weight adjustment methods, efficient parameter space utilization is achieved.

Benefits of technology

It enables rapid and high-precision calibration of TCAD models, improving computational efficiency and calibration accuracy, and supporting subsequent research on radiation effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122310971A_ABST
    Figure CN122310971A_ABST
Patent Text Reader

Abstract

This invention relates to the field of semiconductor device simulation and process development technology, and particularly to a calibration method and related apparatus for computer-aided design technology models (TCAD). The method includes constructing a TCAD device model and determining calibration parameters for the TCAD device model; training a random forest model based on an initial parameter set and labels of the initial parameter set to determine the effective parameter domain of the calibration parameters; and training a Gaussian process regression model within the effective parameter domain to determine the optimal parameter set of the calibration parameters. It can be seen that the proposed embodiment helps to solve the problems of low accuracy, low efficiency, slow convergence, and poor multi-objective optimization effects of traditional calibration methods, achieving rapid and high-precision calibration of TCAD models for subsequent research in areas such as radiation effects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of semiconductor device simulation and process development technology, and in particular to a calibration method and related apparatus for computer-aided design technology models. Background Technology

[0002] With the rapid development of aerospace technology, the reliability of electronic devices in the space environment has become increasingly prominent. Under space radiation, they face severe radiation effects. The effects of space radiation on electronic devices can be mainly classified into two categories: Single Event Effect (SEE) and Total Ionizing Dose (TID). Metal-oxide-semiconductor field-effect transistors (MOSFETs) are the core components of modern integrated circuits. Complementary metal-oxide-semiconductor (CMOS) circuits composed of MOSFETs are highly sensitive to SEE and prone to soft errors. Research on radiation hardening technology for semiconductor electronic devices has significant engineering practical value for improving their reliability in space applications. Establishing accurate and calibrated device models is crucial for studying radiation effect mechanisms and hardening design.

[0003] TCAD (Technology Computer Aided Design) is a core tool for semiconductor device design, process development, and performance prediction. Its core function is to predict the electrical characteristics of devices by solving physical equations such as the Poisson equation and the continuity equation. The accuracy of its simulation results directly determines the R&D efficiency and performance of semiconductor products. TCAD model calibration is a crucial step in semiconductor device simulation and process development. Traditional TCAD calibration methods suffer from several technical drawbacks: First, the calibration process often uses derivative operations to extract device performance indicators, which easily introduces interpolation and random errors, reducing the training effect of machine learning models and affecting calibration accuracy. Second, it employs a "cramming" approach with massive amounts of data for simulation, resulting in blind parameter space sampling, requiring tens of thousands of parameter simulations, and taking two to three months, leading to extremely low computational efficiency. Third, it fails to effectively partition the parameter space, directly performing global optimization, resulting in slow model convergence and difficulty in quickly finding the optimal parameter combination that meets engineering tolerances. Fourth, in multi-objective calibration, the weights of each performance indicator are fixed and cannot be dynamically adjusted according to the calibration progress, making it difficult to achieve collaborative optimization of multiple indicators. Summary of the Invention

[0004] This application provides a calibration method and related apparatus for computer-aided design technology models, which helps to solve the problems of low accuracy, low efficiency, slow convergence, and poor multi-objective optimization effect of traditional calibration methods, and realizes fast and high-precision calibration of TCAD models for subsequent research on radiation effects.

[0005] The first aspect of this application provides a calibration method for a TCAD model, the method comprising:

[0006] Construct a TCAD device model and determine the calibration parameters of the TCAD device model; A random forest model is trained based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters; A Gaussian process regression model is trained within the effective parameter domain to determine the optimal set of parameters for the calibration parameters.

[0007] Optionally, constructing the TCAD device model includes: The TCAD geometric model of a semiconductor device is constructed based on semiconductor device technology. The TCAD geometric model integrates a back-entry well structure, a lightly doped drain structure, and a halo injection structure. A stress effect physical model is embedded in the TCAD geometric model to obtain an intrinsic physical model. The intrinsic physical model describes the physical processes inside the device through fundamental physical equations and the model itself. Based on the intrinsic physical model, the TCAD geometric structure model is adaptively meshed to obtain a mesh model; Configure voltage scan parameters for the mesh model, the voltage scan parameters including scan step size and fixed bias conditions; The original parameters provided by the wafer fab are injected into the mesh model after configuring the voltage scan parameters to obtain the TCAD device model.

[0008] Optionally, determining the calibration parameters of the TCAD device model includes: The variable parameters in the physical parameters of the TCAD device model are determined as the calibration parameters of the TCAD device model.

[0009] Optionally, before training a random forest model based on the initial parameter set and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters, the method further includes: Determine the calibration target, and sample the calibration parameters within the allowable range of values ​​of the calibration parameters to obtain an initial parameter set; The initial parameter set is subjected to TCAD simulation, and the simulation values ​​of the calibration target are labeled according to the preset physical validity criteria to obtain the label of the initial parameter set.

[0010] Optionally, determining the calibration target includes: The electrical performance indicators of the device that can be directly read from the simulation curves without the need for derivation are selected as the calibration targets of the TCAD device model.

[0011] Optionally, training a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters includes: A random forest classification model is trained using the initial parameter set and the labels of the initial parameter set to obtain the trained random forest classification model and the effective probability of each initial parameter set. Calculate whether the ROC AUC value and F1 score of the random forest classification model are greater than a preset threshold; If it is greater than the specified value, then the effective region predicted by the random forest classification model is determined as the effective parameter domain of the calibration parameters. If it is less than the effective probability, a hybrid sampling strategy is used to generate the set of iterative parameters to be simulated in the next round. TCAD simulation is performed on the iterative parameter set, and the simulation values ​​of the calibration target are marked according to the preset physical validity criteria to obtain the label of the iterative parameter set; The iterative parameter set and its labels are added to the initial parameter set, the initial parameter set and its labels are updated, and the step of training a random forest classification model using the initial parameter set and its labels is performed to obtain the trained random forest classification model and the effective probability of each initial parameter set.

[0012] Optionally, training the Gaussian process regression model within the effective parameter domain to determine the optimal set of calibration parameters includes: An initial set of valid parameters is generated based on the valid parameter domain; A Gaussian process regression model is trained based on the initial set of effective parameters to obtain the predicted mean and standard deviation of the calibration target, and the weights of each calibration target are dynamically adjusted according to the calibration progress. The Bayesian optimized acquisition function value of the calibration target is calculated based on the predicted mean, the standard deviation, the weight of the calibration target, and the Kappa factor. The initial set of effective parameters that minimizes the value of the Bayesian optimization acquisition function is determined as the candidate optimal parameter set; TCAD simulation is performed on the candidate optimal parameter set to obtain the effective simulation value of the calibration target; Calculate the loss function value based on the effective simulation value and the target value; When the loss function value is less than or equal to a preset loss threshold, the candidate optimal parameter set is determined as the optimal parameter set; When the loss function value is greater than a preset loss threshold, the initial effective parameter set is updated, and the step of training a Gaussian process regression model based on the initial effective parameter set is executed to obtain the predicted mean and standard deviation of the calibration target.

[0013] A second aspect of this application provides a calibration device for a TCAD model, the device comprising: The model building unit is used to build a TCAD device model and determine the calibration parameters of the TCAD device model. The parameter domain determination unit is used to train a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters. An optimal parameter determination unit is used to train a Gaussian process regression model within the effective parameter domain to determine the optimal set of parameters for the calibration parameters.

[0014] A third aspect of this application provides an electronic device, including: a processor and a memory; The processor is connected to a memory, wherein the memory is used to store computer programs and the processor is used to invoke the computer programs to execute the methods as described in the first aspect of the embodiments of this application.

[0015] A fourth aspect of this application provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, perform the method as described in the first aspect of this application.

[0016] This application embodiment constructs a TCAD device model and determines the calibration parameters of the TCAD device model; trains a random forest model based on the initial parameter set and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters; and trains a Gaussian process regression model within the effective parameter domain to determine the optimal parameter set of the calibration parameters. It can be seen that using this application embodiment helps to solve the problems of low accuracy, low efficiency, slow convergence, and poor multi-objective optimization effects of traditional calibration methods, achieving rapid and high-precision calibration of the TCAD model for subsequent research on radiation effects. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of the operating environment of a calibration application for a TCAD model provided in one embodiment of this application is shown; Figure 2 A schematic flowchart of a calibration method for a TCAD model provided in one embodiment of this application is shown; Figure 3 This invention provides a schematic diagram of the structure of a calibration device for a TCAD model according to an embodiment of the present application. Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of this application is shown. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] Please refer to Figure 1 This diagram illustrates the runtime environment of a calibration application for a TCAD model provided in one embodiment of this application. The runtime environment may include a terminal 10 and a server 20.

[0021] Terminal 10 includes, but is not limited to, electronic devices such as mobile phones, computers, smart voice interaction devices, smart home appliances, in-vehicle terminals, game consoles, e-book readers, multimedia playback devices, and wearable devices. Application clients can be installed on terminal 10.

[0022] In this embodiment, the application described above can be any application capable of providing calibration services for TCAD models. Typically, this application is an industrial application. Of course, besides industrial applications, other types of applications can also provide calibration services dependent on TCAD models. For example, research applications, browser applications, virtual reality (VR) applications, augmented reality (AR) applications, etc., are not limited in this embodiment. Optionally, a client of the aforementioned application runs on terminal 10.

[0023] Server 20 provides background services to clients of applications in terminal 10. For example, server 20 can be a background server for the aforementioned applications. Server 20 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, server 20 can simultaneously provide background services to applications in multiple terminals 10.

[0024] Optionally, terminal 10 and server 20 can communicate with each other via network 30. Terminal 10 and server 20 can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0025] Please refer to Figure 2 This document illustrates a flowchart of a TCAD model calibration method according to an embodiment of this application. This method can be applied to computer equipment, which refers to electronic devices capable of data computation and processing. For example, the executing entity for each step could be... Figure 1 The application runtime environment shown is either terminal 10 or server 20. The method may include the following steps: Step 201: Construct a TCAD device model and determine the calibration parameters of the TCAD device model.

[0026] Specifically, the construction of the TCAD device model includes: The TCAD geometric model of a semiconductor device is constructed based on semiconductor device technology. The TCAD geometric model integrates a back-entry well structure, a lightly doped drain structure, and a halo injection structure. A stress effect physical model is embedded in the TCAD geometric model to obtain an intrinsic physical model. The intrinsic physical model describes the physical processes inside the device through fundamental physical equations and the model itself. Based on the intrinsic physical model, the TCAD geometric structure model is adaptively meshed to obtain a mesh model; Configure voltage scan parameters for the mesh model, the voltage scan parameters including scan step size and fixed bias conditions; The original parameters provided by the wafer fab are injected into the mesh model after configuring the voltage scan parameters to obtain the TCAD device model.

[0027] The TCAD geometric model serves as the physical framework for the TCAD device model. Within TCAD tools (such as Synopsys Sentaurus or Silvaco Athena), by defining material layers, etching windows, and ion implantation energy / dosage / angle, a 2D or 3D structure file containing the aforementioned complex doping distribution can be generated. The backwell structure allows for low surface concentration, improving carrier mobility, while high concentration at deeper depths prevents latch-up and suppresses the short-channel effect (SCE). The lightly doped drain structure alleviates the high electric field near the drain, suppresses hot carrier injection, prevents device degradation over time, and improves reliability. The halo-ring injection structure is specifically designed to suppress the short-channel effect, prevent source-drain punch-through when the channel is too short, and precisely control the threshold voltage.

[0028] Among these, the stress-effect physical model gives the geometric skeleton a physical soul. Stress-related models (such as StressDependentMobility) can be enabled in the solver settings to ensure that the simulator dynamically adjusts the mobility parameters based on the local stress tensor when calculating current. In modern advanced processes, stress is often artificially introduced to improve performance (such as in SiGe source / drain systems and stress pads), which significantly alters carrier mobility. Without a stress model, simulation results will deviate severely from reality. An intrinsic physical model refers to a set of equations coupling the Poisson equation (potential), the carrier continuity equation (current), drift-diffusion or hydrodynamic equations (transport), and a stress-mobility model (such as the Piezoresistive model).

[0029] The mesh model discretizes the continuous physical structure into a computer-computable mesh. Adaptive mesh generation allows for denser meshes in critical regions and sparser meshes in non-critical regions. For example, extremely dense meshes are needed in areas with drastic physical changes, such as the depletion region of a PN junction, the inversion layer channel, the oxide layer interface, and areas with large stress gradients, to capture steep changes in electric field and concentration. Conversely, coarse meshes are used in regions with gradual physical changes, such as deep within the substrate, to save computational memory and time. The solver automatically generates a non-uniform mesh based on these rules. This step directly determines the convergence and accuracy of the simulation.

[0030] The voltage scan parameters are used to define the electrical testing environment and simulate the probe station testing process in the laboratory. The voltage scan parameters include the scan step size and fixed bias conditions. For example, the scan step size can be set to 0.05V to improve the calculation speed while meeting the accuracy requirements. In a single simulation, two fixed bias conditions, 0.1V and 1.1V, can be set simultaneously, and the results at two voltages can be obtained at the same time, further improving the simulation efficiency.

[0031] The original parameters provided by the wafer fab transform the general simulation template into a dedicated model for a specific process node, which can maintain the originality and stability of the parameter file and can be fine-tuned according to the fitting requirements in the subsequent optimization stage.

[0032] The TCAD model contains a vast number of physical parameters, sometimes dozens or even hundreds, which can be divided into two categories: fixed parameters and variable parameters. Fixed parameters are core process parameters that have been validated extensively by the wafer fab and are highly reliable. Variable parameters are those subject to process fluctuations, model uncertainties, or require fine-tuning for specific device performance. Through sensitivity analysis (such as the Sobol index method), the top N parameters with the greatest impact on the calibration target are selected as the calibration parameters for the TCAD device model, while parameters with negligible impact are eliminated. This reduces the risk of the dimensionality curse in machine learning and accelerates convergence. N can be 9, 13, 17, or other values, and is not limited here.

[0033] In this embodiment, high-dimensional parameter space optimization is employed to accelerate model convergence: the physical parameters of the TCAD model are simplified in dimension, identifying fewer fixed and variable parameters verified by wafer fabs, and setting a reasonable range of values ​​for the variable parameters to reduce the computational complexity of machine learning; physically infeasible parameters are filtered out in advance through a classification model to avoid invalid samples from participating in regression model training, thereby improving the efficiency of parameter space utilization, solving the problem of slow convergence in traditional global optimization, and achieving rapid optimization in a high-dimensional parameter space.

[0034] Step 202: Train a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters.

[0035] Furthermore, before training a random forest model based on the initial parameter set and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters, the method further includes: Determine the calibration target, and sample the calibration parameters within the allowable range of values ​​of the calibration parameters to obtain an initial parameter set; The initial parameter set is subjected to TCAD simulation, and the simulation values ​​of the calibration target are labeled according to the preset physical validity criteria to obtain the label of the initial parameter set.

[0036] Specifically, determining the calibration target includes: The electrical performance indicators of the device that can be directly read from the simulation curves without the need for derivation are selected as the calibration targets of the TCAD device model.

[0037] For example, the calibration targets could be the off-state current, threshold voltage, and on-state current under high and low operating conditions, where high and low operating conditions refer to different drain-source voltages. ,For example and Off-state current When the device is in the off state, i.e., the gate voltage is... Or, when the voltage is below the threshold voltage, the leakage current flowing from the drain to the source. Threshold voltage The minimum gate voltage required for the device channel to begin forming and for current to begin conducting significantly is the critical point at which the device transitions from off to on. On-state current. The drive current flowing through the device when the device is fully turned on, i.e., when the gate voltage is equal to the rated operating voltage.

[0038] In this embodiment, a derivative-free index design is adopted, which can improve calibration accuracy from the root: abandoning performance indicators that require differentiation, such as subthreshold swing, and selecting high and low pressure that can be directly read from the IV curve as the calibration target, eliminating interpolation errors and random noise introduced by differentiation calculations, avoiding the problem of parameter and performance mapping instability caused by simulation curve distortion, thereby reducing the mean linearity error of calibration results and significantly improving the fitting degree between TCAD model simulation and measured data.

[0039] The allowable range of values ​​for the calibration parameters cannot be infinitely large, otherwise the optimization algorithm will get lost in the vast space and struggle to find the optimal solution. Nor can it be too narrow; if the true optimal value falls outside the set range, the optimization will fail indefinitely. Therefore, the order of magnitude of the calibration parameters can be determined through experimentally obtained empirical constants, and then allowed to fluctuate within one order of magnitude to obtain the value range. A fluctuation of one order of magnitude is usually sufficient to cover process variations and model errors.

[0040] The calibration parameters include N parameters, therefore each initial parameter set includes these N calibration parameters. Similarly, each iterative parameter set, initial effective parameter set, candidate optimal parameter set, and optimal parameter set described below also include these N calibration parameters. The dataset can include multiple of the above parameter sets.

[0041] For example, TCAD simulation is performed on the initial parameter set, and the simulated values ​​of the calibration target are labeled according to a preset physical validity criterion to obtain the labels of the initial parameter set, including: The initial simulation values ​​of the calibration target are obtained by performing simulation using the TCAD device model based on the initial parameter set. The initial simulation values ​​are evaluated for validity based on preset physical validity criteria to obtain the initial validity evaluation result; The initial parameter set is labeled according to the initial validity judgment result, and the labels of the initial parameter set are obtained.

[0042] Among them, the physical validity criterion is a set of rules used in the process of semiconductor device simulation (TCAD) and model calibration to automatically determine whether a set of process parameters or simulation results conform to basic physical laws and engineering common sense. For example, it may include whether there is obvious leakage current, whether the threshold voltage is near the design target, whether the on-state current is large enough (greater than the preset on-state current value), and whether the off-state current is small enough (less than the preset off-state current value).

[0043] The validity judgment result can be either valid or invalid. A valid label can be identified by 1, and an invalid label can be identified by 0. Of course, other numbers, letters or symbols can also be used to identify it, and there are no restrictions here.

[0044] Specifically, the step of training a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters includes: A random forest classification model is trained using the initial parameter set and the labels of the initial parameter set to obtain the trained random forest classification model and the effective probability of each initial parameter set. Calculate whether the ROC AUC value and F1 score of the random forest classification model are greater than a preset threshold; If it is greater than the specified value, then the effective region predicted by the random forest classification model is determined as the effective parameter domain of the calibration parameters. If it is less than the effective probability, a hybrid sampling strategy is used to generate the set of iterative parameters to be simulated in the next round. TCAD simulation is performed on the iterative parameter set, and the simulation values ​​of the calibration target are marked according to the preset physical validity criteria to obtain the label of the iterative parameter set; The iterative parameter set and its labels are added to the initial parameter set, the initial parameter set and its labels are updated, and the step of training a random forest classification model using the initial parameter set and its labels is performed to obtain the trained random forest classification model and the effective probability of each initial parameter set.

[0045] In the random forest model, the effective probability This equals the number of trees predicted as valid (i.e., labeled 1). The ratio of the total number of trees ,Right now ROC AUC, or Area Under the Receiver Operating Characteristic Curve, depicts the trade-off between true positives (TP) and false positives (FPR) at different classification thresholds. AUC is the area under the ROC curve. The F1 score is the harmonic mean of precision and recall, used to comprehensively evaluate the model's classification performance at a specific threshold (usually 0.5). The preset thresholds for ROC AUC and F1 score can be the same or different; for example, both can be set to 85%, or other values ​​are not limited here.

[0046] The hybrid sampling strategy can be to select a preset number of samples respectively. Utilization points and preset quantities The exploration points form the set of iterative parameters to be simulated in the next round. Among them, the points used are the parameter combinations with the highest effective probability predicted by the classifier. Group parameters; the exploration point is the parameter combination predicted by the classifier with an effective probability closest to 50%. Group parameters. Details and The value of can be adjusted according to computing resources. The two can be equal or unequal, and no restriction is made here.

[0047] In this embodiment, an active learning domain division strategy is adopted to achieve an order-of-magnitude improvement in computational efficiency. Subsequent regression optimization is carried out only within the effective domain, eliminating a large number of invalid parameter combinations. Compared with the months-long time required by traditional massive sampling, the efficiency is improved by more than 100 times, and the consumption of computing resources is greatly reduced.

[0048] Step 203: Train a Gaussian process regression model within the effective parameter domain to determine the optimal set of calibration parameters.

[0049] Specifically, training the Gaussian process regression model within the effective parameter domain to determine the optimal set of calibration parameters includes: An initial set of valid parameters is generated based on the valid parameter domain; A Gaussian process regression model is trained based on the initial set of effective parameters to obtain the predicted mean and standard deviation of the calibration target, and the weights of each calibration target are dynamically adjusted according to the calibration progress. The Bayesian optimized acquisition function value of the calibration target is calculated based on the predicted mean, the standard deviation, the weight of the calibration target, and the Kappa factor. The initial set of effective parameters that minimizes the value of the Bayesian optimization acquisition function is determined as the candidate optimal parameter set; TCAD simulation is performed on the candidate optimal parameter set to obtain the effective simulation value of the calibration target; Calculate the loss function value based on the effective simulation value and the target value; When the loss function value is less than or equal to a preset loss threshold, the candidate optimal parameter set is determined as the optimal parameter set; When the loss function value is greater than a preset loss threshold, the initial effective parameter set is updated, and the step of training a Gaussian process regression model based on the initial effective parameter set is executed to obtain the predicted mean and standard deviation of the calibration target.

[0050] For example, generating an initial set of valid parameters based on the valid parameter domain can be achieved by performing dense sampling or grid sampling within the valid domain to generate a set of initial valid parameters; it can also be the set of valid parameters in the incremental parameter set finally generated during the iteration of the random forest classification model, that is, the set of initial valid parameters with valid labels in the updated initial set of initial valid parameters in the last iteration; or it can be determined in other ways, which are not limited here.

[0051] The Bayesian optimized acquisition function is as follows: , Where n is the number of calibration targets. Let i be the weight of the i-th calibration target. To predict the mean, For the carpa factor, Standard deviation, This is the set of parameters. The Kappa factor is gradually reduced with each iteration.

[0052] The preset loss threshold can be calculated based on the relative error between the effective simulation value and the target value. For example, it can be set to 15%, 10%, 5% of the relative error, or other values, which are not limited here. If the value is greater than the preset loss threshold, the currently calculated set of candidate optimal parameters, its target value, and the relative error are added to the initial set of effective parameters to update the initial set of effective parameters.

[0053] In this embodiment, the invention employs dynamic adjustment of the Kappa factor to avoid local optima and achieve global optimization: the Kappa factor is embedded into the Bayesian optimization upper confidence bound acquisition function, and the exploration and utilization relationship of the optimization process is dynamically adjusted through the factor value; in the early stage of optimization, the factor is increased to expand the exploration range of the unknown parameter region and avoid getting trapped in local optima; in the later stage of optimization, the factor is adjusted to 0, and fine sampling is performed in the known optimal region to ensure that a globally optimal parameter combination that meets the engineering tolerance is found.

[0054] Furthermore, after obtaining the optimal parameter set, other core parameters in the TCAD device model can be fine-tuned based on this optimal parameter set. These other core parameters could be, for example, well region doping concentration, LDD doping dose, etc., with fine-tuning increments of ±5%, ±3%, ±1%, or other values, which are not limited here. After fine-tuning, TCAD simulation is performed again to obtain the final calibration target simulation value. Then, quantitative and qualitative verification is performed on this simulation value. If both verifications pass, the calibration parameters at this point are determined as the final calibration parameters. Quantitative verification involves comparing the simulation value with the target value and calculating whether the relative error is less than the preset error. Qualitative verification involves comparing the obtained IV simulation curve with the PDK target curve to see if they match.

[0055] As can be seen, the embodiments of this application construct a TCAD device model and determine the calibration parameters of the TCAD device model; train a random forest model based on the initial parameter set and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters; and train a Gaussian process regression model within the effective parameter domain to determine the optimal parameter set of the calibration parameters. It can be seen that adopting the embodiments of this application helps to solve the problems of low accuracy, low efficiency, slow convergence, and poor multi-objective optimization effects of traditional calibration methods, achieving rapid and high-precision calibration of the TCAD model for subsequent research on radiation effects.

[0056] Figure 3 A schematic diagram of a calibration device for a TCAD model according to an embodiment of this application is shown. The device includes: The model building unit 301 is used to build a TCAD device model and determine the calibration parameters of the TCAD device model; The parameter domain determination unit 302 is used to train a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters. The optimal parameter determination unit 303 is used to train a Gaussian process regression model within the effective parameter domain to determine the optimal parameter set of the calibration parameters.

[0057] Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of this application is shown, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the functions of the computer system of the calibration method for the TCAD model in any of the above embodiments.

[0058] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a computer, causes the computer to perform the functions of a computer system for the calibration method of the TCAD model in any of the above embodiments.

[0059] This application also provides a computer program product containing instructions that, when executed by a computer, cause the computer to perform the functions of the computer system of the TCAD model calibration method in any of the above embodiments.

[0060] It is understood that the specific examples in this application are only intended to help those skilled in the art better understand the implementation methods of this application, and are not intended to limit the scope of the invention.

[0061] It is understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not limit the implementation process of the embodiments of this application in any way.

[0062] It is understood that the various implementation methods described in this application can be implemented individually or in combination, and the implementation methods in this application are not limited in this respect.

[0063] Unless otherwise stated, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0064] It is understood that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0065] It is understood that the memory in the embodiments of this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Specifically, non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0066] Those skilled in the art will recognize that the units 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.

[0067] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the aforementioned method implementations, and will not be repeated here.

[0068] 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 units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0069] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0070] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0071] If a function is implemented as a software functional unit 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 part of the technical solution, can be embodied in the form of a software product. The 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 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.

[0072] The above are merely specific embodiments of this application, but the scope of protection of this invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A calibration method for a computer-aided design technology model, characterized in that, The method includes: Construct a computer-aided design technology device model and determine the calibration parameters of the computer-aided design technology device model; A random forest model is trained based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters; A Gaussian process regression model is trained within the effective parameter domain to determine the optimal set of parameters for the calibration parameters.

2. The method according to claim 1, characterized in that, The construction of the computer-aided design device technology model includes: The computer-aided design technology geometric model of semiconductor devices is constructed based on semiconductor device technology. The computer-aided design technology geometric model integrates the back well structure, the lightly doped drain structure and the halo ring injection structure. By embedding a stress effect physical model into the geometric structure model of the computer-aided design technology, an intrinsic physical model is obtained. The intrinsic physical model describes the physical processes inside the device through fundamental physical equations and the model itself. Based on the intrinsic physical model, the geometric structure model of the computer-aided design technology is adaptively meshed to obtain a mesh model; Configure voltage scan parameters for the mesh model, the voltage scan parameters including scan step size and fixed bias conditions; The original parameters provided by the wafer fab are injected into the mesh model after configuring the voltage scanning parameters to obtain a computer-aided design technology device model.

3. The method according to claim 1, characterized in that, Determining the calibration parameters of the computer-aided design technology device model includes: The variable parameters in the physical parameters of the computer-aided design technology device model are determined as the calibration parameters of the computer-aided design technology device model.

4. The method according to claim 1, characterized in that, Before training a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters, the method further includes: Determine the calibration target, and sample the calibration parameters within the allowable range of values ​​of the calibration parameters to obtain an initial parameter set; The initial parameter set is simulated using computer-aided design technology, and the simulated values ​​of the calibration target are labeled according to a preset physical validity criterion to obtain the label of the initial parameter set.

5. The method according to claim 4, characterized in that, The determination of calibration targets includes: Electrical performance indicators of the device that can be directly read from the simulation curves without the need for derivation are selected as the calibration targets for the device model of the computer-aided design technology.

6. The method according to claim 1, characterized in that, The step of training a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters includes: A random forest classification model is trained using the initial parameter set and the labels of the initial parameter set to obtain the trained random forest classification model and the effective probability of each initial parameter set. Calculate whether the ROC AUC value and F1 score of the random forest classification model are greater than a preset threshold; If it is greater than the specified value, then the effective region predicted by the random forest classification model is determined as the effective parameter domain of the calibration parameters. If it is less than the effective probability, a hybrid sampling strategy is used to generate the set of iterative parameters to be simulated in the next round. The iterative parameter set is simulated using computer-aided design technology, and the simulated values ​​of the calibration target are labeled according to a preset physical validity criterion to obtain the label of the iterative parameter set; The iterative parameter set and its labels are added to the initial parameter set, the initial parameter set and its labels are updated, and the step of training a random forest classification model using the initial parameter set and its labels is performed to obtain the trained random forest classification model and the effective probability of each initial parameter set.

7. The method according to claim 1, characterized in that, The step of training a Gaussian process regression model within the effective parameter domain to determine the optimal set of calibration parameters includes: An initial set of valid parameters is generated based on the valid parameter domain; A Gaussian process regression model is trained based on the initial set of effective parameters to obtain the predicted mean and standard deviation of the calibration target, and the weights of each calibration target are dynamically adjusted according to the calibration progress. The Bayesian optimized acquisition function value of the calibration target is calculated based on the predicted mean, the standard deviation, the weight of the calibration target, and the Kappa factor. The initial set of effective parameters that minimizes the value of the Bayesian optimization acquisition function is determined as the candidate optimal parameter set; The candidate optimal parameter set is simulated using computer-aided design technology to obtain the effective simulated value of the calibration target; Calculate the loss function value based on the effective simulation value and the target value; When the loss function value is less than or equal to a preset loss threshold, the candidate optimal parameter set is determined as the optimal parameter set; When the loss function value is greater than a preset loss threshold, the initial effective parameter set is updated, and the step of training a Gaussian process regression model based on the initial effective parameter set is executed to obtain the predicted mean and standard deviation of the calibration target.

8. A calibration device for a computer-aided design technology model, characterized in that, The device includes: The model building unit is used to build a computer-aided design technology device model and determine the calibration parameters of the computer-aided design technology device model. The parameter domain determination unit is used to train a random forest model based on the initial parameter set of the calibration parameters and the labels of the initial parameter set to determine the effective parameter domain of the calibration parameters. An optimal parameter determination unit is used to train a Gaussian process regression model within the effective parameter domain to determine the optimal set of parameters for the calibration parameters.

9. An electronic device, characterized in that, include: Processor and memory; The processor is connected to a memory, wherein the memory is used to store a computer program, and the processor is used to invoke the computer program to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, perform the method as described in any one of claims 1-7.