Effective value checking method and system in model parameter optimization process

Abstract

The invention relates to the technical field of model parameter optimization, and provides a method for checking an effective value in a model parameter optimization process, which comprises the following steps of: S1, determining a parameter to be optimized; s2, searching whether the to-be-optimized parameter is a model parameter influencing the effective value of the length L or the width W of the device or not; s3, a detection parameter list is constructed according to the search result, and if the to-be-optimized parameters are model parameters influencing the effective value of the length L or the width W, all bin parameters related to the effective value of the length L or the width W in the model are added to the detection parameter list; s4, in the process of optimizing the to-be-optimized parameters, synchronously detecting whether the effective value of each parameter in the detection parameter list exceeds a preset boundary or not; s5, if the effective value of at least one parameter in the detection parameter list exceeds the preset boundary, adopting a dichotomy fallback algorithm to fallback the optimized parameter value; otherwise, updating the numerical value of the to-be-optimized parameter, and continuing the subsequent optimization process. And multi-correlation parameter border crossing caused by single parameter optimization can be effectively avoided.

Description

Technical Field

[0001] This invention relates to the technical field of model parameter optimization, and in particular to a method and system for checking effective values ​​during the model parameter optimization process. Background Technology

[0002] In the field of semiconductor device modeling and integrated circuit design, automated modeling processes are a core component for achieving accurate device performance simulation and shortening chip development cycles. Parameter optimization based on standardized models (such as the BSIM series models) is a crucial step determining modeling accuracy and efficiency. Currently, BSIM models (such as BSIM-4 and BSIM-6) have become the mainstream device behavior description models in the industry, widely used in modeling core semiconductor devices such as MOSFETs. The rationality of their model parameters directly affects the simulation accuracy of key performance indicators such as chip current, voltage, and power loss. Therefore, optimizing model parameters (especially core parameters related to device geometry) is an indispensable part of automated modeling processes.

[0003] In the parameter optimization practice of the BSIM series models, the length (L) and width (W, also often represented by NFIN) of a device are fundamental geometric parameters that determine its electrical characteristics. The effective values ​​of many key parameters in the model depend on the effective values ​​of L or W. For example, in the BSIM-4 model, parameters such as lint (length injection parameter), ll (length bias parameter), and lw (length-width correlation parameter) directly participate in the calculation of the effective value of L, while parameters such as wint (width injection parameter), wl (width bias parameter), and ww (width-width correlation parameter) directly affect the accuracy of the effective value of W. Once these L and W-related parameters enter the optimization process, their numerical adjustments are passed through the effective values ​​of L or W, further affecting the numerical stability of "binable parameters" in the model. Here, "binable parameters" specifically refers to a class of parameters in the model that need to be calculated based on the effective values ​​of L or W. These parameters cover those related to key performance characteristics such as device on-current, threshold voltage, gate capacitance, and breakdown voltage, and their numerical rationality directly determines whether the model can accurately reflect the actual operating state of the device.

[0004] However, existing parameter optimization schemes in automated modeling processes have significant technical flaws: traditional optimization methods often focus on adjusting the numerical values ​​of single target parameters, failing to establish a collaborative checking mechanism between "core parameters (parameters affecting the effective value of L / W) and related binatable parameters." Specifically, when optimizing parameters such as lint, ll, and lw in the BSIM-4 model, constraints are only applied to the numerical range of these target parameters themselves, neglecting the impact of changes in the effective value of L after adjustment on all binatable parameters that depend on L calculations. The effective values ​​of such binatable parameters may exceed preset physical boundaries or engineering allowable ranges due to abnormal fluctuations in the effective value of L, leading to distorted model simulation results (such as non-physical current surges, threshold voltage drift, etc.). Similarly, when optimizing parameters such as wint, wl, and ww in the BSIM-4 model, traditional schemes also fail to simultaneously monitor the impact of changes in the effective value of W on related binatable parameters, which can easily lead to binatable parameter out-of-bounds issues.

[0005] More notably, while different BSIM models (such as BSIM-6 and BSIM-4) share the same core logic, the specific parameter definitions affecting the effective values ​​of L and W, as well as the types and calculation methods of binarable parameters, differ. Existing optimization schemes lack a universal "correlated parameter effective value checking mechanism": localized checking logic tailored to a specific model cannot be migrated to the optimization process of other models, resulting in poor compatibility and weak scalability of the automated modeling process. These issues directly lead to two key consequences: firstly, after parameter optimization, the validity of each associated binarable parameter needs to be manually checked, breaking the closed loop of the automated modeling process and significantly reducing modeling efficiency; secondly, if binarable parameters are not detected out of bounds in time, chip design simulations based on that model will deviate, potentially leading to subsequent tape-out failures and significantly increasing R&D costs and risks.

[0006] In summary, existing parameter optimization techniques in automated modeling processes have gaps in the collaborative control of "optimization of core parameters affecting the effective value of L / W" and "ensuring the effective value of associated binable parameters." There is an urgent need for a technical solution that can automatically identify associated parameters, synchronously check for out-of-bounds effective values, and implement anomaly rollback to solve the problem of multiple associated parameters going out of bounds due to single parameter optimization, and ensure the rationality of model parameter optimization and the stability of automated modeling processes. Summary of the Invention

[0007] To address the aforementioned problems, the present invention aims to provide a method and system for valid value checking during model parameter optimization. In the automated modeling process of model parameter optimization (such as optimizing parameters affecting the valid value of L / W, such as lint, ll, lw or wint, wl, ww in BSIM series models), by automatically identifying and including all binarable parameters affected by L / W in the detection, and combining the bisection backoff algorithm to handle out-of-bounds errors, it can effectively avoid multiple related parameters from going out of bounds due to the optimization of a single parameter, ensuring the rationality of model parameter optimization, and improving the efficiency of automated modeling and the accuracy of subsequent simulations.

[0008] The above-mentioned objective of this invention is achieved through the following technical solutions: A method for checking valid values ​​during model parameter optimization includes the following steps: S1: Obtain the list of model parameters to be optimized and determine the parameters to be optimized; S2: Based on the official manual corresponding to the current model, check whether the parameter to be optimized is a model parameter that affects the effective value of the device length L or width W; S3: Construct a detection parameter list based on the search results. If the parameter to be optimized is a model parameter that affects the effective value of length L or width W, then add all binable parameters in the model that are related to the effective value of length L or width W to the detection parameter list. If the parameter to be optimized is not a model parameter that affects the effective value of length L or width W, then retain the parameter to be optimized as a unique element in the detection parameter list. S4: During the optimization of the parameters to be optimized, simultaneously detect whether the effective values ​​of each parameter in the detection parameter list exceed the preset boundary; S5: If at least one parameter in the detection parameter list has a valid value that exceeds the preset boundary, the optimized parameter value is rolled back using the binary backoff algorithm, and step S4 is executed again; if the valid values ​​of all parameters in the detection parameter list do not exceed the preset boundary, the value of the parameter to be optimized is updated, and the subsequent optimization process continues.

[0009] Furthermore, in step S2, the current model is a standardized model used for semiconductor device modeling, including models such as BSIM-4 and BSIM-6; When the current model is the BSIM-4 model, the model parameters affecting the effective value of the device length L specifically include the length injection parameter lint, the length offset parameter ll, and the length-width correlation parameter lw. These parameters directly participate in the determination of the effective value of the length L through the built-in formula of the BSIM-4 model. The model parameters affecting the effective value of the device width W specifically include the width injection parameter wint, the width offset parameter wl, and the width-width correlation parameter ww. These parameters directly participate in the determination of the effective value of W through the built-in formula of the BSIM-4 model. When the current model is the BSIM-6 model, the model parameters affecting the effective value of the device length L specifically include the length process error parameter lpe0, the length nonlinearity correction parameter lpe1, and the length temperature coefficient parameter ltemp. These parameters directly participate in the determination of the effective value of the length L through the built-in effective value calculation model of the BSIM-6 model. The model parameters affecting the effective value of the device width W specifically include the width process error parameter wpe0, the width nonlinearity correction parameter wpe1, and the width temperature coefficient parameter wtemp. These parameters directly participate in the determination of the effective value of W through the built-in effective value calculation model of the BSIM-6 model.

[0010] Furthermore, in step S3, the binable parameters related to the effective value of length L or width W refer to the parameters in the model that need to be calculated based on the effective value of length L or width W and can be used for performance classification of semiconductor devices. The binarable parameters specifically include parameters related to device on-current, gate capacitance, threshold voltage, breakdown voltage, and power loss. The accuracy of these parameters directly determines the accuracy of the model's simulation of the semiconductor device's operating state.

[0011] Furthermore, in step S4, the preset boundary is the allowable range of valid values ​​of each parameter pre-set based on the manufacturing process requirements of semiconductor devices, the constraints of theoretical physical models, and the range of measured data obtained from testing actual devices. If the effective value of the parameter exceeds the preset boundary, the simulation results of the model will deviate from the electrical characteristics of the actual device, thereby affecting the accuracy of integrated circuit design based on the model.

[0012] Furthermore, in step S5, the binary search backoff algorithm combines the efficiency of binary search with the robustness of step-by-step backoff, including a binary search phase and a backoff mechanism. The specific execution process is as follows: Binary search phase: First, initialize the search range [low, high] of the parameter to be optimized, where low is the minimum allowed value of the parameter and high is the maximum allowed value of the parameter; simultaneously, set the target value, which is the preset boundary of the valid value of the corresponding parameter in the parameter list; then proceed with iterative calculation, calculate the midpoint mid=(low+high) / 2 of the current search range, substitute the parameter value corresponding to mid into the model, and calculate the valid value data[mid] corresponding to the parameter; compare the valid value data[mid] with the target value target. If the valid value is equal to the target value, then mid is directly determined as the target parameter value that meets the requirements; if the valid value does not match the target value, then adjust low or high according to the comparison result to narrow the search range, and repeat the above iterative calculation until the target parameter value meets the requirements or low>high; Backoff mechanism: When the binary search phase ends due to low > high and no parameter value is found that makes the effective value equal to the target value, the backoff mechanism is triggered. Starting from the mean mi obtained from the last calculation in the binary search phase, a linear search is first performed in one direction. If the first parameter value that makes the effective value of the corresponding parameter meet the preset boundary is found, it is taken as a feasible solution. If no parameter value that meets the requirements is found in that direction, the linear search continues in the opposite direction until the first parameter value that meets the preset boundary is found as a feasible solution.

[0013] Furthermore, the time complexity of the binary search stage is O(log n), where n is the number of discrete values ​​that the parameter to be optimized can take within the search range [low, high]. This time complexity reflects that, in the ideal scenario where the parameter values ​​and effective values ​​change monotonically and the target value exists, the binary search can quickly narrow the search range with logarithmic efficiency, significantly improving the iterative efficiency of parameter optimization.

[0014] Furthermore, the rollback mechanism is used to handle abnormal scenarios, including those where the effective value of the parameter and the value of the parameter to be optimized are not monotonic, and where the target value is missing within the search range. In these scenarios, the backoff mechanism, through linear expansion search, can start from the termination point of the binary search and find the parameter value that is closest to the target value and meets the preset boundary as a feasible solution. This ensures that the parameter optimization process will not be interrupted due to abnormal scenarios, and realizes closed-loop control of detecting out-of-bounds → backoff to find a feasible solution → re-optimization.

[0015] A system for checking effective values ​​during model parameter optimization, used to perform the effective value checking method described above, includes: The parameter determination module is used to obtain a list of model parameters to be optimized and determine the parameters to be optimized. The parameter lookup module is used to search, based on the official manual corresponding to the current model, whether the parameter to be optimized is a model parameter that affects the effective value of the device length L or width W. The list building module is used to build a detection parameter list based on the search results. If the parameter to be optimized is a model parameter that affects the effective value of length L or width W, then all binable parameters in the model that are related to the effective value of length L or width W are added to the detection parameter list. If the parameter to be optimized is not a model parameter that affects the effective value of length L or width W, then the parameter to be optimized is retained as a unique element in the detection parameter list. The boundary detection module is used to simultaneously detect whether the valid values ​​of each parameter in the detection parameter list exceed the preset boundary during the optimization process of the parameter to be optimized. The rollback iteration module is used to roll back the optimized parameter value using a binary rollback algorithm and re-execute the function of the boundary detection module if at least one parameter in the detection parameter list has a valid value that exceeds the preset boundary; if none of the valid values ​​of all parameters in the detection parameter list exceed the preset boundary, the value of the parameter to be optimized is updated and the subsequent optimization process continues.

[0016] A computer device includes a memory and one or more processors, the memory storing computer code that, when executed by the one or more processors, causes the one or more processors to perform the method described above.

[0017] A computer-readable storage medium storing computer code that, when executed, performs the method described above.

[0018] Compared with the prior art, the present invention has at least one of the following beneficial effects: (1) Effectively avoid multiple related parameters going out of bounds due to single parameter optimization, and ensure the rationality of optimization. This invention accurately identifies "parameters that affect the effective value of L / W" through steps S2-S3, and automatically includes all binarable parameters affected by L / W into the detection list (instead of focusing only on a single optimization parameter). Then, through steps S4-S5, it simultaneously detects out-of-bounds errors and performs binary backoff processing. This fundamentally solves the defect of traditional optimization that "only considers a single parameter and ignores the influence of related factors", ensuring that all related binarable parameters meet the boundary requirements after parameter optimization, and guaranteeing the physical rationality and engineering effectiveness of model parameter optimization.

[0019] (2) Improve the efficiency of automated modeling and reduce the cost of manual intervention. The entire process eliminates the need for manual verification of each associated parameter: Step S2 automatically determines parameter correlation based on the official model manual, Step S3 automatically constructs a detection list, and Step S5 automatically handles out-of-bounds errors using a binary backoff algorithm (without requiring manual adjustment of backoff values). This completely breaks the process breakpoint of "manually checking for out-of-bounds errors after optimization" in traditional optimization, significantly shortens the iteration cycle of automated modeling, and reduces manual operation costs and the risk of human error.

[0020] (3) It is compatible with multiple types of standardized models and has strong universality. This invention explicitly supports standardized semiconductor device models such as BSIM-4 and BSIM-6, and precisely defines the "parameter types affecting the effective value of L / W" for different models (such as lint / ll / lw in BSIM-4 and lpe0 / lpe1 in BSIM-6), adapting to the parameter calculation logic of different models. It does not require customizing a dedicated solution for a single model and can be widely applied to various automated modeling scenarios based on standardized models, with strong universality and scalability.

[0021] (4) Improve the accuracy of model simulation and ensure the reliability of subsequent design. On the one hand, the bin parameters in step S3 cover core performance indicators such as device on-current, gate capacitance, and threshold voltage, ensuring a comprehensive detection range. On the other hand, the "preset boundary" in step S4 is set based on semiconductor device manufacturing process requirements, theoretical physical model constraints, and measured data, rather than subjective values. This can effectively avoid the problem of "simulation results deviating from the actual electrical characteristics of the device" caused by parameters exceeding the boundary, providing accurate model support for subsequent integrated circuit design (such as chip current and power loss simulation) and reducing the risk of tape-out failure.

[0022] (5) The algorithm balances efficiency and robustness to ensure the stability of the optimization process. The binary backoff algorithm used has two advantages: Efficiency: The time complexity of the binary search phase is O(logn), which can logarithmically narrow down the search range and quickly locate the optimal parameter value in ideal scenarios where the parameters and effective values ​​have a monotonic relationship. Robustness: The rollback mechanism can handle abnormal scenarios such as "parameters and effective values ​​are not monotonic" and "target values ​​are missing". It linearly searches for feasible solutions from the binary search termination point, ensuring that the optimization process will not be interrupted even under non-ideal conditions, and achieving closed-loop stable control of "detection → rollback → re-optimization". Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the effective value checking method during the model parameter optimization process of this invention. Figure 2 This is a diagram of the overall system for checking effective values ​​during the optimization of model parameters in this invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0026] This invention provides a processing scheme to ensure the rationality of model parameter optimization in the parameter optimization process. Specifically, in the pre-optimization preparation stage of the parameter optimization process, the parameters to be optimized are first searched. The core of the search is to screen out the model parameters that affect the device length L or width W (i.e., NFIN), because such parameters directly affect the effective value calculation of other binatable parameters—and the effective value calculation of binatable parameters requires the effective values ​​of L and W as necessary components. The search process must be based on the official technical manual of the currently used model (such as BSIM-4, BSIM-6, etc.). When the parameter to be optimized is detected to be a model parameter that affects L or W, all binatable parameters in that model are added to the detection parameter list. In the subsequent optimization process of the parameter to be optimized, the effective values ​​of each parameter in the detection parameter list are simultaneously calculated to see if they exceed the preset boundary. If any parameter in the list exceeds the preset boundary, the optimized parameter value is rolled back through the algorithm, and the optimization and effective value detection process is re-executed until the optimization result no longer causes any parameter in the detection parameter list to exceed the boundary. At this point, the parameter optimization is considered successful. This approach effectively addresses the unreasonable problem of optimizing only a single parameter leading to multiple related parameters going out of bounds, thus helping to ensure the rationality of model parameter optimization.

[0027] The following is an illustration through specific examples: First Embodiment like Figure 1 As shown in the figure, this embodiment provides a method for checking effective values ​​during model parameter optimization, characterized by including the following steps: S1: Obtain the list of model parameters to be optimized and determine the parameters to be optimized.

[0028] S2: Based on the official manual corresponding to the current model, check whether the parameter to be optimized is a model parameter that affects the effective value of the device length L or width W.

[0029] In step S2, the current model is a standardized model used for semiconductor device modeling, including models such as BSIM-4 and BSIM-6; When the current model is the BSIM-4 model, the model parameters affecting the effective value of the device length L specifically include the length injection parameter lint, the length offset parameter ll, and the length-width correlation parameter lw. These parameters directly participate in the determination of the effective value of the length L through the built-in formula of the BSIM-4 model. The model parameters affecting the effective value of the device width W specifically include the width injection parameter wint, the width offset parameter wl, and the width-width correlation parameter ww. These parameters directly participate in the determination of the effective value of W through the built-in formula of the BSIM-4 model. When the current model is the BSIM-6 model, the model parameters affecting the effective value of the device length L specifically include the length process error parameter lpe0, the length nonlinearity correction parameter lpe1, and the length temperature coefficient parameter ltemp. These parameters directly participate in the determination of the effective value of the length L through the built-in effective value calculation model of the BSIM-6 model. The model parameters affecting the effective value of the device width W specifically include the width process error parameter wpe0, the width nonlinearity correction parameter wpe1, and the width temperature coefficient parameter wtemp. These parameters directly participate in the determination of the effective value of W through the built-in effective value calculation model of the BSIM-6 model.

[0030] The core function of step S2 is to accurately identify the correlation between the parameters to be optimized and the effective values ​​of the core geometric parameters (length L, width W) of the device, providing a basis for subsequently constructing a comprehensive list of detection parameters. The key to this step is to use the official model manual as the authoritative reference, because different standardized models (such as BSIM-4 and BSIM-6) differ in the definition, physical meaning, and logic of the parameters involved in the calculations. The manual is the only standard for clearly defining these correlations.

[0031] (1) Regarding the positioning of the "current model" The term "current model" specifically refers to standardized models (such as the BSIM series) in the field of semiconductor device modeling. These models are developed by industry organizations, unifying the description logic of device electrical characteristics and serving as universal tools for chip design and simulation. Taking BSIM-4 and BSIM-6 as examples, although both are used for MOSFET device modeling, with technological advancements, BSIM-6 offers more refined parameter definitions (such as introducing temperature coefficients and process error corrections), thus resulting in differences in the types of parameters affecting the effective value of L / W and their calculation methods.

[0032] (2) Parameters affecting the effective value of L / W in the BSIM-4 model and their effects The BSIM-4 model calculates the effective values ​​of L and W using explicit mathematical formulas, with relevant parameters directly participating in the formula calculations: Parameters affecting the effective value of L: Length injection parameter (lint): Used to correct the deviation between the actual channel length and the design length during device manufacturing. For example, when there is a lithographic deviation in chip manufacturing, lint will directly adjust the effective value of L (L_eff) through the formula L_eff=L_design-2×lint (simplified formula). Length offset parameter (ll): compensates for the nonlinear deviation in length caused by the short channel effect. When L is small (e.g., less than 0.1 μm), ll will affect L_eff through the quadratic correction formula. Length-width correlation parameter (lw): Considering the coupling effect between L and W (such as the edge effect when the width-to-length ratio is too small), it participates in the calculation of L_eff through cross terms. For example, if the optimization lint is adjusted from 0.01μm to 0.015μm, L_eff will decrease accordingly after being substituted into the formula, which directly causes the values ​​of all binatable parameters that depend on L_eff (such as drain current Ids, whose formula contains the reciprocal of L_eff) to change.

[0033] Parameters affecting the effective value of W: The logic of the width injection parameter (wint), width bias parameter (wl), and width-width correlation parameter (ww) is similar to that of the L-related parameters, except that they are calculated for the effective value (W_eff) of the width W. For example, wint corrects the actual width deviation by using W_eff = W_design - 2 × wint. Optimizing wint will directly change W_eff, which in turn affects the width-related binarable parameters (such as the gate capacitance Cgg, which is proportional to W_eff).

[0034] (3) Parameters affecting the effective value of L / W in the BSIM-6 model and their effects The BSIM-6 model employs a more complex "computational model" (rather than a single formula), introducing more physical effect corrections (such as process errors and temperature effects). Relevant parameters participate in the L / W RMS value calculation through multi-dimensional corrections. Parameters affecting the effective value of L: Length process error parameter (lpe0): compensates for the systematic length error caused by batch manufacturing differences. For example, if the overall length of a certain batch of chips is 0.02μm smaller, lpe0 will correct L_eff through the basic item. Length nonlinearity correction parameter (lpe1): handles the quantum effect deviation of L at its minimum value (such as sub-10 nanometer scale) by correcting the nonlinear change of L_eff through an exponential term; Length temperature coefficient parameter (ltemp): Considering the effect of temperature on material expansion, such as a slight elongation of the device length at high temperatures, ltemp dynamically adjusts its effective value using L_eff = L_base × (1 + ltemp × ΔT). For example, if ltemp is optimized from 0.0001 / ℃ to 0.00015 / ℃, the correction amount for L_eff will increase from 0.5% to 0.75% when the temperature changes by 50℃, directly affecting binarable parameters dependent on L_eff (such as breakdown voltage, which is positively correlated with L_eff).

[0035] Parameters affecting the effective value of W: The width process error parameter (wpe0), width nonlinearity correction parameter (wpe1), and width temperature coefficient parameter (wtemp) correspond to the process error, nonlinear effect, and temperature influence in the width direction, respectively, and participate in the calculation model of W_eff through similar logic. For example, wpe0 corrects the width deviation at different wafer positions. Optimizing wpe0 will change the basic value of W_eff, thereby affecting the width-related binarable parameters (such as source-drain resistance, which is inversely proportional to W_eff).

[0036] In summary, step S2, by referring to the model manual, accurately locates the "parameters that affect the effective value of L / W". Essentially, it identifies the "scope of influence" of parameter optimization. This step is a prerequisite for building the detection list, ensuring that all binable parameters that may be affected by changes in L / W are included in the monitoring, and avoiding the "butterfly effect" (adjusting a single parameter causing multiple parameters to go out of bounds) in the optimization process.

[0037] S3: Construct a detection parameter list based on the search results. If the parameter to be optimized is a model parameter that affects the effective value of length L or width W, then add all binable parameters in the model that are related to the effective value of length L or width W to the detection parameter list. If the parameter to be optimized is not a model parameter that affects the effective value of length L or width W, then retain the parameter to be optimized as a unique element in the detection parameter list.

[0038] In step S3, the binatable parameters related to the effective value of length L or width W refer to the parameters in the model that need to be calculated based on the effective value of length L or width W and can be used for performance classification of semiconductor devices. The binarable parameters specifically include parameters related to device on-current, gate capacitance, threshold voltage, breakdown voltage, and power loss. The accuracy of these parameters directly determines the accuracy of the model's simulation of the semiconductor device's operating state.

[0039] The core function of step S3 is to accurately define the range of parameters to be monitored based on the correlation between the parameters to be optimized and the effective value of L / W, providing a clear target for subsequent detection of effective value out-of-bounds errors. This step, through the logic of "differentiated construction of the detection list," ensures both the comprehensiveness of monitoring (when the parameter affects L / W) and avoids meaningless redundant detection (when the parameter does not affect L / W), ultimately achieving the goal of "targeted monitoring and efficient optimization."

[0040] (1) Logic for constructing the detection parameter list: distinction between two scenarios The construction of the detection parameter list follows the principle that "the scope of influence determines the scope of monitoring": Scenario 1: The parameter to be optimized affects the effective value of L or W. At this point, since the effective values ​​of L or W are passed to all binarable parameters that depend on them through the calculation formula, the values ​​of these binarable parameters may fluctuate due to changes in L / W. Therefore, all binarable parameters in the model that are related to the effective values ​​of L or W need to be included in the detection list to ensure that any potential out-of-bounds risks can be captured.

[0041] Scenario 2: The parameter to be optimized does not affect the effective value of L or W. At this point, the impact of parameter optimization is limited to the parameter itself (it will not affect other parameters through L / W), so the detection list only needs to retain the parameter to be optimized itself to avoid unnecessary detection overhead.

[0042] (2) Core features and examples of binarable parameters However, bin parameters are key parameters in the model that combine "computational dependency" and "engineering practicality": Calculation dependency: It must be based on the effective value of L or W as input, and calculated through the built-in formula of the model (i.e. its value is directly affected by the effective value of L / W). Engineering applicability: It can be used for performance binning of semiconductor devices – during chip mass production, devices can be classified into different levels (such as high-performance binning, low-power binning) based on the numerical range of these parameters, guiding subsequent circuit design and selection.

[0043] Specific examples (using the BSIM-4 model as an example): On-state current (e.g., drain current Ids): Its calculation formula is Ids=(μCox / 2)×(W_eff / L_eff)×(Vgs-Vth) 2(Simplified form), where W_eff (RMS value of W) and L_eff (RMS value of L) are the core variables. Optimizing lint (which affects L_eff) will directly increase Ids if L_eff decreases. If this exceeds preset boundaries (such as the maximum allowable current), it may cause the device to overheat. Therefore, Ids needs to be included in the detection list.

[0044] Gate capacitance (e.g., gate-source capacitance Cgs): The calculation formula includes the term W_eff × L_eff (related to device area). If wint is optimized (affecting W_eff), changes in W_eff will cause Cgs to change synchronously. Cgs directly affects the switching speed of the circuit. If it exceeds the boundary, it will cause timing simulation deviations, so it needs to be included in the detection.

[0045] Threshold voltage (Vth): In short-channel devices, Vth decreases as L_eff decreases (short-channel effect), and its calculation requires the inclusion of an L_eff correction term. If ll (the bias parameter affecting L_eff) is optimized, Vth may deviate from the design value, leading to abnormal device on / off states. Therefore, Vth needs to be included in the detection.

[0046] Breakdown voltage (BVdss): It is positively correlated with L_eff (the larger L_eff is, the higher the breakdown voltage). If L_eff is too small due to parameter optimization, BVdss may be lower than the safe value, resulting in insufficient device withstand voltage. Therefore, it needs to be included in the test.

[0047] Power loss (P): Calculated by P = Ids × Vds. Ids is affected by L / W, therefore the value of P also indirectly depends on the effective value of L / W. If P exceeds the limit, it may cause the chip power consumption to exceed the standard, which needs to be included in the testing.

[0048] (3) Detection list examples in two scenarios Scenario 1: The lint (affecting L_eff) detection list for parameters to be optimized using BSIM-4 must include all binable parameters that depend on L_eff, such as: drain current Ids, gate-source capacitance Cgs, threshold voltage Vth, breakdown voltage BVdss, power loss P, etc. During the optimization process, it is necessary to simultaneously monitor whether the effective values ​​of these parameters exceed the limits.

[0049] Scenario 2: The parameter to be optimized is an independent temperature compensation parameter (not affecting L / W). Assuming a parameter is only used to fine-tune the temperature coefficient of the device, and its calculation does not involve the effective value of L / W, the detection list only includes the temperature compensation parameter itself. During optimization, it is only necessary to monitor whether its effective value is within the allowable range.

[0050] In summary, step S3, through the precise mapping of "association range → detection range", ensures comprehensive monitoring during the optimization of L / W related parameters (avoiding out-of-bounds correlation parameters) and simplifies the detection process during the optimization of non-association parameters (improving efficiency), laying a precise and efficient foundation for subsequent handling of out-of-bounds effective values.

[0051] S4: During the optimization of the parameters to be optimized, simultaneously detect whether the effective values ​​of each parameter in the detection parameter list exceed the preset boundary.

[0052] In step S4, the preset boundary is the allowable range of effective values ​​of each parameter pre-set based on the manufacturing process requirements of semiconductor devices, the constraints of theoretical physical models, and the range of measured data obtained from testing actual devices. If the effective value of the parameter exceeds the preset boundary, the simulation results of the model will deviate from the electrical characteristics of the actual device, thereby affecting the accuracy of integrated circuit design based on the model.

[0053] Step S4 is a "real-time monitoring checkpoint" in the parameter optimization process. Its core function is to track the effective values ​​and verify the boundaries of all parameters in the parameter list while the parameters to be optimized are iteratively adjusted, ensuring that the optimization process does not deviate from physical laws and engineering requirements. This step, through a "synchronous detection" mechanism, controls potential risks of exceeding limits during the optimization process (rather than after optimization is completed), providing timely basis for subsequent rollback adjustments.

[0054] (1) The core logic of "synchronous detection" "Simultaneous detection" emphasizes the parallelism between "optimization actions" and "detection actions": while the algorithm is numerically adjusting the target parameters (such as lint in BSIM-4) (e.g., iterating from 0.01μm to 0.012μm), it immediately calls the model's built-in formulas to calculate the effective values ​​of all parameters in the detection list (such as Ids, Cgs, Vth, etc.) and compares them with preset boundaries. This parallelism avoids the invalid iterations that may be caused by "completing optimization first and then detecting" (e.g., discovering that the associated parameters are out of bounds only at the end of the optimization process, requiring rework), significantly improving optimization efficiency.

[0055] (2) Three main criteria for setting the preset boundary (taking the conduction current Ids as an example) The preset boundaries are not subjectively set, but are determined comprehensively based on the "physical nature" and "engineering practice" of semiconductor devices, specifically including: Manufacturing process requirements: These are explicitly defined in the process specification (PDK) provided by the chip foundry. For example, a certain 180nm process specifies that the maximum on-state current Ids of an NMOS transistor must not exceed 10mA / μm (to avoid overheating and melting of the metal interconnects). This value is the upper limit of Ids.

[0056] Theoretical physical model constraints: Limit values ​​derived from semiconductor physics principles. For example, according to the Shockley equation, Ids is related to physical quantities such as carrier mobility and gate oxide capacitance. When Ids exceeds (μCoxW / (2L))×(Vdd)... 2 When Vdd is the power supply voltage, it will violate the square law and enter a non-physical state. Therefore, this value constitutes the theoretical upper limit of Ids.

[0057] Actual device test data: Statistical range obtained through actual measurements on test chips before mass production. For example, after testing 1000 samples, it was found that the normal fluctuation range of Ids is 2mA / μm~8mA / μm. Therefore, the boundary was set to 1.5mA / μm (lower limit, leaving a 10% margin)~8.5mA / μm (upper limit, leaving a 10% margin), covering 99% of normal devices.

[0058] Taking all three factors into account, the final preset boundary set for Ids may be "1.5mA / μm≤Ids≤8.5mA / μm", which satisfies both process limitations and physical laws, while also being compatible with individual differences in actual devices.

[0059] (3) Hazards of parameter effective values ​​exceeding preset boundaries (example illustration) If the detection finds that the effective value of the parameter is out of bounds, it will directly destroy the "simulation realism" of the model and have a chain reaction on subsequent integrated circuit design: Example 1: On-state current Ids exceeds the upper limit. Assuming that after lint optimization, L_eff decreases, causing the effective value of Ids to reach 9mA / μm (exceeding the preset upper limit of 8.5mA / μm), the simulation model will incorrectly assume that the device can withstand a larger current. Power management circuits designed based on this may set the output current too high, while the actual device will overheat and burn out due to the Ids exceeding the limit, leading to fabrication failure.

[0060] Example 2: Threshold voltage Vth is lower than the lower limit. If the preset lower limit of Vth is 0.4V (based on process requirements), but L_eff decreases abnormally after optimization ll, the effective value of Vth drops to 0.3V. In simulation, it is assumed that the device can conduct at 0.3V, but in the actual circuit, this will cause the device to turn on falsely at low voltage (such as abnormal power consumption in standby mode), violating the low power consumption design requirements.

[0061] Example 3: Gate capacitance Cgs exceeds the upper limit. The preset upper limit of Cgs is determined by the circuit timing requirements (e.g., high-frequency circuits require Cgs ≤ 10fF). If W_eff increases after optimizing wint, causing the effective value of Cgs to reach 12fF, the simulation will underestimate the switching delay of the circuit, and the actual chip may not be able to work at the target frequency due to timing inconsistencies.

[0062] In summary, step S4, through a dual mechanism of "synchronous detection + scientific boundary," provides a "safety valve" for the parameter optimization process: on the one hand, it ensures that all parameters to be monitored are always within a physically reasonable and engineering feasible range; on the other hand, by promptly detecting out-of-bounds errors, it provides a precise trigger signal for subsequent bisection backoff, avoiding the optimization process from "progressing with defects," and ultimately ensuring the consistency between the model simulation results and the actual device characteristics.

[0063] S5: If at least one parameter in the detection parameter list has a valid value that exceeds the preset boundary, the optimized parameter value is rolled back using the binary backoff algorithm, and step S4 is executed again; if the valid values ​​of all parameters in the detection parameter list do not exceed the preset boundary, the value of the parameter to be optimized is updated, and the subsequent optimization process continues.

[0064] In step S5, the binary search backoff algorithm combines the efficiency of binary search with the robustness of step-by-step backoff, including a binary search phase and a backoff mechanism. The specific execution process is as follows: Binary search phase: First, initialize the search range [low, high] of the parameter to be optimized, where low is the minimum allowed value of the parameter and high is the maximum allowed value of the parameter; simultaneously, set the target value, which is the preset boundary of the valid value of the corresponding parameter in the parameter list; then proceed with iterative calculation, calculate the midpoint mid=(low+high) / 2 of the current search range, substitute the parameter value corresponding to mid into the model, and calculate the valid value data[mid] corresponding to the parameter; compare the valid value data[mid] with the target value target. If the valid value is equal to the target value, then mid is directly determined as the target parameter value that meets the requirements; if the valid value does not match the target value, then adjust low or high according to the comparison result to narrow the search range, and repeat the above iterative calculation until the target parameter value meets the requirements or low>high; Backoff mechanism: When the binary search phase ends due to low > high and no parameter value is found that makes the effective value equal to the target value, the backoff mechanism is triggered. Starting from the mean mi obtained from the last calculation in the binary search phase, a linear search is first performed in one direction. If the first parameter value that makes the effective value of the corresponding parameter meet the preset boundary is found, it is taken as a feasible solution. If no parameter value that meets the requirements is found in that direction, the linear search continues in the opposite direction until the first parameter value that meets the preset boundary is found as a feasible solution.

[0065] Furthermore, the time complexity of the binary search stage is O(log n), where n is the number of discrete values ​​that the parameter to be optimized can take within the search range [low, high]. This time complexity reflects that, in the ideal scenario where the parameter values ​​and effective values ​​change monotonically and the target value exists, the binary search can quickly narrow the search range with logarithmic efficiency, significantly improving the iterative efficiency of parameter optimization.

[0066] Furthermore, the rollback mechanism is used to handle abnormal scenarios, including those where the effective value of the parameter and the value of the parameter to be optimized are not monotonic, and where the target value is missing within the search range. In these scenarios, the backoff mechanism, through linear expansion search, can start from the termination point of the binary search and find the parameter value that is closest to the target value and meets the preset boundary as a feasible solution. This ensures that the parameter optimization process will not be interrupted due to abnormal scenarios, and realizes closed-loop control of detecting out-of-bounds → backoff to find a feasible solution → re-optimization.

[0067] The binary search backoff algorithm in step S5 is the core execution logic for solving the "valid value out of bounds" problem in parameter optimization. Its design essence is a combination of "efficient positioning" and "fault tolerance fallback": it quickly locks the ideal solution through binary search, and deals with complex abnormal scenarios through the backoff mechanism, ultimately ensuring that a feasible solution that meets the boundary requirements can be found regardless of whether the parameter relationship is ideal, thus avoiding the interruption of the optimization process.

[0068] (1) Core logic of the algorithm: the synergistic effect of "binary search + backoff mechanism" The two phases of the binary search backoff algorithm do not run independently, but rather form a progressive relationship of "first efficient search, then fault-tolerant fallback": Binary search phase: Utilizing the characteristic of "shrinking the range by half each time", in the ideal scenario where the parameter and the effective value have a monotonic relationship (such as lint increasing → L_eff decreasing → Ids increasing), the parameter value that makes the effective value exactly equal to the preset boundary (target) is quickly located, achieving "precise hit"; Backoff mechanism: When the binary search fails due to complex parameter relationships (such as non-monotonicity or missing target value), the search is linearly extended from the binary search termination point to ensure that the "nearest solution that meets the boundary" is found, thus achieving "backup guarantee".

[0069] (2) Binary search stage: Efficiently locate the ideal solution (example illustration) Taking the optimization of lint parameters (which affect L_eff) in the BSIM-4 model as an example: The search range for the parameter lint to be optimized is: [low=0.01μm, high=0.05μm] (low is the minimum allowable value, and high is the maximum allowable value). In the list of detection parameters, the preset upper limit of the conduction current Ids is 8.5mA / μm (i.e., target=8.5mA / μm, and it must be ensured that Ids≤8.5). Given the following relationship: lint increases → L_eff decreases → Ids increases (this is a monotonic relationship, which meets the ideal conditions for binary search).

[0070] Binary search execution process 1. Initialization: low=0.01μm, high=0.05μm, target=8.5mA / μm; 2. First iteration: Calculate mid = (0.01 + 0.05) / 2 = 0.03 μm; Substituting lint=0.03μm into the model, we calculate Ids=9mA / μm (>target, out of bounds). Since lint is positively correlated with Ids (the larger the lint, the larger the Ids), lint needs to be reduced to lower Ids, so high=0.03μm is adjusted (to reduce the upper bound). 3. Second iteration: New mid = (0.01 + 0.03) / 2 = 0.02 μm; Substituting the values, we get Ids = 8mA / μm (≤ target, which meets the requirements). To find a better solution (as close to the target as possible to retain performance margin), adjust low=0.02μm (narrow the small bound); 4. Third iteration: New mid = (0.02 + 0.03) / 2 = 0.025 μm; Substituting the values, we get Ids = 8.5 mA / μm (= target, which is a perfect match). At this point, the ideal solution is found, and lint=0.025μm is determined as the target parameter value, thus ending the binary search phase.

[0071] The above process finds the target value in just 3 iterations, while a linear search (detecting values ​​incrementally from 0.01μm to 0.001μm) might require 25 iterations. This highlights the advantage of binary search's O(logn) time complexity—when the search range contains n discrete values, only log2n iterations are needed to narrow it down to the vicinity of the target. This is particularly significant in large-scale parameter optimization (e.g., when n=1000, linear search requires 1000 iterations, while binary search only requires 10), resulting in a substantial efficiency improvement.

[0072] (3) Rollback mechanism: fault tolerance and fallback handling of abnormal scenarios (example illustration) Taking lint optimization as an example, the relationship between lint and Ids becomes non-monotonic due to the "short-channel quantum effect" in the device: when lint < 0.02 μm, increasing lint increases Ids; when lint > 0.02 μm, increasing lint decreases Ids (quantum confinement effect dominates). In this case, binary search may fail. Search range [0.01μm, 0.05μm], target=8.5mA / μm; After the binary search iteration, because the relationship is not monotonic, the final values ​​are low = 0.022μm and high = 0.021μm (low > high). No solution was found that makes Ids = 8.5, triggering the backoff mechanism.

[0073] rollback mechanism execution process 1. Determine the starting point: Use mid = 0.0215μm, calculated in the last binary search, as the starting point; 2. Unidirectional linear search: First, detect in the direction of decreasing lint (0.021μm→0.020μm→...): When lint=0.021μm, Ids=8.6mA / μm (out of bounds); When lint=0.020μm, Ids=8.4mA / μm (≤target, meets the requirements); 3. Determine a feasible solution: The first solution that meets the boundary conditions with lint=0.020μm is a feasible solution, and the backoff mechanism ends.

[0074] If the search is not found in that direction (e.g., all smaller lint values ​​cause Ids to go out of bounds), then search in the opposite direction (increasing lint) until a value that meets the requirements is found.

[0075] (4) The core value of the algorithm Balancing efficiency and robustness: In ideal scenarios, binary search is used for fast location (O(log n) efficiency), while in complex scenarios, a fallback mechanism is used as a backup to avoid optimization interruption due to abnormal parameter relationships; Guarantee optimization closed loop: Regardless of whether the exact target value is found, a feasible solution that meets the preset boundary can be returned in the end, supporting the closed loop process of "detecting out of bounds → rollback and adjustment → re-optimization", ensuring that the model parameters are always within the range of physical reasonableness and engineering feasibility; Adapting to semiconductor device characteristics: For semiconductor parameters that often exhibit characteristics such as "nonlinearity and temperature / process sensitivity" (such as the complex relationship between parameters and effective values ​​in the BSIM-6 model), reliable numerical adjustment schemes are provided to avoid cascading out-of-bounds problems caused by optimizing a single parameter.

[0076] In short, the algorithm is both "fast" and "stable," and is a key technical support for ensuring the rationality of model parameter optimization.

[0077] Second Embodiment like Figure 2 As shown, this embodiment provides a valid value checking system for model parameter optimization, used to perform the valid value checking method in the model parameter optimization process as described in the first embodiment, comprising: The parameter determination module is used to obtain a list of model parameters to be optimized and determine the parameters to be optimized. The parameter lookup module is used to search, based on the official manual corresponding to the current model, whether the parameter to be optimized is a model parameter that affects the effective value of the device length L or width W. The list building module is used to build a detection parameter list based on the search results. If the parameter to be optimized is a model parameter that affects the effective value of length L or width W, then all binable parameters in the model that are related to the effective value of length L or width W are added to the detection parameter list. If the parameter to be optimized is not a model parameter that affects the effective value of length L or width W, then the parameter to be optimized is retained as a unique element in the detection parameter list. The boundary detection module is used to simultaneously detect whether the valid values ​​of each parameter in the detection parameter list exceed the preset boundary during the optimization process of the parameter to be optimized. The rollback iteration module is used to roll back the optimized parameter value using a binary rollback algorithm and re-execute the function of the boundary detection module if at least one parameter in the detection parameter list has a valid value that exceeds the preset boundary; if none of the valid values ​​of all parameters in the detection parameter list exceed the preset boundary, the value of the parameter to be optimized is updated and the subsequent optimization process continues.

[0078] A computer-readable storage medium stores computer code that, when executed, performs the methods described above. Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0079] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0080] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above 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.

[0081] It should be noted that the above embodiments can be freely combined as needed. The above description is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for checking effective values ​​during model parameter optimization, characterized in that, Includes the following steps: S1: Obtain the list of model parameters to be optimized and determine the parameters to be optimized; S2: Based on the official manual corresponding to the current model, check whether the parameter to be optimized is a model parameter that affects the effective value of the device length L or width W; S3: Construct a detection parameter list based on the search results. If the parameter to be optimized is a model parameter that affects the effective value of length L or width W, then add all binable parameters in the model that are related to the effective value of length L or width W to the detection parameter list. If the parameter to be optimized is not a model parameter that affects the effective value of length L or width W, then retain the parameter to be optimized as a unique element in the detection parameter list. S4: During the optimization of the parameters to be optimized, simultaneously detect whether the effective values ​​of each parameter in the detection parameter list exceed the preset boundary; S5: If at least one parameter in the detection parameter list has a valid value that exceeds the preset boundary, the optimized parameter value is rolled back using the binary backoff algorithm, and step S4 is executed again; if the valid values ​​of all parameters in the detection parameter list do not exceed the preset boundary, the value of the parameter to be optimized is updated, and the subsequent optimization process continues.

2. The method for checking effective values ​​during model parameter optimization according to claim 1, characterized in that, In step S2, the current model is a standardized model used for semiconductor device modeling, including models such as BSIM-4 and BSIM-6; When the current model is the BSIM-4 model, the model parameters affecting the effective value of the device length L specifically include the length injection parameter lint, the length offset parameter ll, and the length-width correlation parameter lw. These parameters directly participate in the determination of the effective value of the length L through the built-in formula of the BSIM-4 model. The model parameters affecting the effective value of the device width W specifically include the width injection parameter wint, the width offset parameter wl, and the width-width correlation parameter ww. These parameters directly participate in the determination of the effective value of W through the built-in formula of the BSIM-4 model. When the current model is the BSIM-6 model, the model parameters affecting the effective value of the device length L specifically include the length process error parameter lpe0, the length nonlinearity correction parameter lpe1, and the length temperature coefficient parameter ltemp. These parameters directly participate in the determination of the effective value of the length L through the built-in effective value calculation model of the BSIM-6 model. The model parameters affecting the effective value of the device width W specifically include the width process error parameter wpe0, the width nonlinearity correction parameter wpe1, and the width temperature coefficient parameter wtemp. These parameters directly participate in the determination of the effective value of W through the built-in effective value calculation model of the BSIM-6 model.

3. The method for checking effective values ​​during model parameter optimization according to claim 1, characterized in that, In step S3, the binatable parameters related to the effective value of length L or width W refer to the parameters in the model that need to be calculated based on the effective value of length L or width W and can be used for performance classification of semiconductor devices. The binarable parameters specifically include parameters related to device on-current, gate capacitance, threshold voltage, breakdown voltage, and power loss. The accuracy of these parameters directly determines the accuracy of the model's simulation of the semiconductor device's operating state.

4. The method for checking effective values ​​during model parameter optimization according to claim 1, characterized in that, In step S4, the preset boundary is the allowable range of effective values ​​of each parameter pre-set based on the manufacturing process requirements of semiconductor devices, the constraints of theoretical physical models, and the range of measured data obtained from testing actual devices. If the effective value of the parameter exceeds the preset boundary, the simulation results of the model will deviate from the electrical characteristics of the actual device, thereby affecting the accuracy of integrated circuit design based on the model.

5. The method for checking effective values ​​during model parameter optimization according to claim 1, characterized in that, In step S5, the binary search backoff algorithm combines the efficiency of binary search with the robustness of step-by-step backoff, including a binary search phase and a backoff mechanism. The specific execution process is as follows: Binary search phase: First, initialize the search range [low, high] of the parameter to be optimized, where low is the minimum allowed value of the parameter and high is the maximum allowed value of the parameter; simultaneously, set the target value, which is the preset boundary of the valid value of the corresponding parameter in the parameter list; then proceed with iterative calculation, calculate the midpoint mid=(low+high) / 2 of the current search range, substitute the parameter value corresponding to mid into the model, and calculate the valid value data[mid] corresponding to the parameter; compare the valid value data[mid] with the target value target. If the valid value is equal to the target value, then mid is directly determined as the target parameter value that meets the requirements; if the valid value does not match the target value, then adjust low or high according to the comparison result to narrow the search range, and repeat the above iterative calculation until the target parameter value meets the requirements or low>high; Backoff mechanism: When the binary search phase ends due to low > high and no parameter value is found that makes the effective value equal to the target value, the backoff mechanism is triggered. Starting from the mean mi obtained from the last calculation in the binary search phase, a linear search is first performed in one direction. If the first parameter value that makes the effective value of the corresponding parameter meet the preset boundary is found, it is taken as a feasible solution. If no parameter value that meets the requirements is found in that direction, the linear search continues in the opposite direction until the first parameter value that meets the preset boundary is found as a feasible solution.

6. The method for checking effective values ​​during model parameter optimization according to claim 5, characterized in that, The time complexity of the binary search stage is O(log n), where n is the number of discrete values ​​that the parameter to be optimized can take within the search range [low, high]. This time complexity reflects that, in the ideal scenario where the parameter values ​​and effective values ​​change monotonically and the target value exists, the binary search can quickly narrow the search range with logarithmic efficiency, significantly improving the iterative efficiency of parameter optimization.

7. The method for checking effective values ​​during model parameter optimization according to claim 5, characterized in that, The rollback mechanism is used to handle abnormal scenarios, including those where the effective value of the parameter and the value of the parameter to be optimized are not monotonic, and the target value is missing within the search range. In these scenarios, the backoff mechanism, through linear expansion search, can start from the termination point of the binary search and find the parameter value that is closest to the target value and meets the preset boundary as a feasible solution. This ensures that the parameter optimization process will not be interrupted due to abnormal scenarios, and realizes closed-loop control of detecting out-of-bounds → backoff to find a feasible solution → re-optimization.

8. A system for checking effective values ​​during model parameter optimization, used to perform the effective value checking method as described in any one of claims 1-7, characterized in that, include: The parameter determination module is used to obtain a list of model parameters to be optimized and determine the parameters to be optimized. The parameter lookup module is used to search, based on the official manual corresponding to the current model, whether the parameter to be optimized is a model parameter that affects the effective value of the device length L or width W. The list building module is used to build a detection parameter list based on the search results. If the parameter to be optimized is a model parameter that affects the effective value of length L or width W, then all binable parameters in the model that are related to the effective value of length L or width W are added to the detection parameter list. If the parameter to be optimized is not a model parameter that affects the effective value of length L or width W, then the parameter to be optimized is retained as a unique element in the detection parameter list. The boundary detection module is used to simultaneously detect whether the valid values ​​of each parameter in the detection parameter list exceed the preset boundary during the optimization process of the parameter to be optimized. The rollback iteration module is used to roll back the optimized parameter value using a binary rollback algorithm and re-execute the function of the boundary detection module if at least one parameter in the detection parameter list has a valid value that exceeds the preset boundary; if none of the valid values ​​of all parameters in the detection parameter list exceed the preset boundary, the value of the parameter to be optimized is updated and the subsequent optimization process continues.

9. A computer device comprising a memory and one or more processors, the memory storing computer code that, when executed by the one or more processors, causes the one or more processors to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer code, wherein when the computer code is executed, the method of any one of claims 1 to 7 is performed.

Images