A circuit design parameter solving method and device based on statistical angle drift control

Abstract

The application provides a circuit design parameter solving method and device based on statistical corner drift control, and belongs to the field of integrated circuit design. The method comprises the following steps: determining a circuit design parameter set and a process random parameter set corresponding to a target yield; fixing the design parameter set, obtaining a circuit performance sample through simulation and calculating a statistical corner; during iterative optimization, determining an optimization starting point according to a difference between a performance value and a threshold value, updating a size of a trust region according to a statistical corner drift of adjacent periods to obtain a real-time trust region; and optimizing parameters in the real-time trust region until a stop condition is met to obtain optimal design parameters, so that the circuit reaches the target yield under process random disturbance. Through statistical corner drift constraint and adaptive adjustment of the trust region, the statistical corner drift is effectively inhibited, the number of simulations is reduced, and the stability and efficiency of yield optimization are improved.

Description

Technical Field

[0001] This application relates to the field of integrated circuit design, and in particular to a method and apparatus for solving circuit design parameters based on statistical angle drift control. Background Technology

[0002] As integrated circuit manufacturing processes advance to the nanoscale, the impact of device mismatch and process randomness on circuit performance becomes increasingly significant, making yield optimization in high-end chip design ever more urgent. In practical design scenarios, especially during the development of memory and high-precision analog circuits, designers face challenges such as high simulation sampling costs, the difficulty of optimizing high-dimensional parameter spaces, and the inability of traditional methods to balance accuracy and efficiency. For circuit design itself, while pursuing performance, it is difficult to ensure stable mass production yields under the influence of large-scale fluctuations in process random parameters. By introducing Bayesian optimization and statistical learning methods to optimize statistical angles, designers can explore the design space with fewer simulations, reducing design iteration costs. Among these, yield estimation based on statistical angles and trust region constraint optimization, as core technologies of EDA intelligence, are crucial for improving chip design success rates and yields through algorithmic innovation. Currently, the increasing market demand for high-performance, high-reliability chips is driving the rapid evolution of yield optimization technologies based on machine learning.

[0003] Existing circuit yield optimization methods mainly employ two approaches: The first is direct yield estimation and optimization based on Monte Carlo simulation, which involves large-scale sampling and circuit simulation of the random process parameter set in each iteration to directly evaluate the yield at the current design point and use this as a guide for parameter adjustment; the second is approximate optimization based on statistical angles, which involves extracting a few "worst process angles" that characterize the yield boundary offline or online, simplifying the high-dimensional yield optimization problem into optimizing the performance of these key points, thereby reducing simulation overhead.

[0004] However, the above two methods have the following problems: 1. High simulation overhead for yield estimation: Both traditional Monte Carlo and improved Monte Carlo methods require a large number of samples to estimate the current yield, and new design yields need to be continuously estimated during the optimization process; 2. Unstable optimization process: Non-MC yield estimation methods may lead to misjudgment of yield, affecting the actual optimization effect. Using yield as the model target masks the specific characterization of circuit performance, increasing the difficulty of design parameter optimization; 3. Statistical angle drift: In optimization based on statistical angles, the performance under the new design parameters is actually estimated using the statistical angles extracted beforehand, which will deviate to a certain extent from the actual quantile performance of the current design parameters. This deviation affects the reliability of the optimization results. Summary of the Invention

[0005] In view of this, this application provides a method and apparatus for solving circuit design parameters based on statistical angle drift control, in order to solve the problems of unreliable yield optimization results and huge simulation overhead caused by statistical angle drift in the prior art.

[0006] Specifically, this application is implemented through the following technical solution:

[0007] The first aspect of this application provides a method for solving circuit design parameters based on statistical angle drift control, the method comprising:

[0008] Determine the set of circuit design parameters and the set of random process parameters corresponding to the target yield of the finished circuit product;

[0009] Using the circuit design parameter set and the process random parameter set as input, the circuit performance value corresponding to the input is obtained based on the circuit simulator, a circuit design sample set is generated, and a statistical angle is calculated based on the circuit design sample set. The statistical angle is used to describe the relationship between the process random parameters and the circuit performance value.

[0010] The circuit design parameter set is iteratively optimized based on statistical angles. In each cycle, the starting point for optimization of the circuit design parameter set is determined according to the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. The circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met to obtain the optimal circuit design parameter set. The circuit generated by the optimal circuit design parameter set meets the target yield requirement under the perturbation of the process random parameter set.

[0011] In a preferred embodiment of this application, the iterative optimization of the circuit design parameter set based on statistical angles includes:

[0012] Based on the statistical angle, determine whether the circuit design parameter set meets the yield requirements;

[0013] If not satisfied, calculate the target violation distance of the circuit design parameter set in the (k+1)th round;

[0014] Calculate the real-time statistical angle in the (k+1)th round; calculate the circuit performance parameters corresponding to the historical statistical angle and the real-time statistical angle under the circuit design parameter set in the (k+1)th round respectively, and obtain the statistical angle drift;

[0015] Based on the target violation distance, determine the starting point for optimization in the (k+2)th round;

[0016] Based on the statistical angle drift, calculate the trust region reliability index, and update the trust region radius based on the trust region reliability index.

[0017] Starting from the optimization point, the real-time trust region for the (k+2)th round is constructed using the updated trust region radius;

[0018] The circuit design parameter set is optimized in the real-time trust domain, and the step of calculating the statistical angle based on the circuit design sample set is returned to output the optimal circuit design parameter set.

[0019] In another preferred embodiment of this application, determining the optimization starting point of the circuit design parameter set based on the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold includes:

[0020] Based on the calculation of the statistical angle, the set of simulation data with the worst circuit performance is selected, and the performance gap corresponding to each subset of simulation data is calculated.

[0021] The violation distance of the current circuit design parameter set is obtained by summing the performance gaps.

[0022] The weighted target performance threshold is obtained by multiplying the weighting parameter by the number of violations of the target performance threshold in the simulation data set with the worst circuit performance.

[0023] The target violation distance is obtained by summing the weighted violation target performance threshold and the violation distance.

[0024] In another preferred embodiment of this application, updating the size of the trust region of the circuit design parameter set based on the statistical angle drift in adjacent periods includes:

[0025] By fixing the random parameter set of the process, the real-time circuit design parameter set and the predicted statistical angle performance parameters of the (k+1)th round of optimization calculation are obtained. The predicted performance parameters are calculated based on the real-time circuit design parameter set and the historical statistical angles corresponding to the kth round of optimization.

[0026] The true statistical angle is recalculated based on the real-time circuit design parameter set.

[0027] Calculate the performance parameters of the real statistical angle based on the real statistical angle and the real-time circuit design parameters;

[0028] The statistical angle drift is obtained based on the actual circuit performance parameters and the predicted statistical angle performance parameters.

[0029] In another preferred embodiment of this application, determining the optimization starting point of the circuit design parameter set based on the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold includes:

[0030] The target violation distance in round (k+1) is compared with that in round (k) to determine the optimization starting point. When the target violation distance in round (k+1) is less than the target violation distance under optimization in round (k), the circuit design parameter set in round (k+1) is determined as the optimization starting point. When the target violation distance in round (k+1) is greater than the target violation distance under optimization in round (k), the circuit design parameter set in round (k) is determined as the optimization starting point.

[0031] In another preferred embodiment of this application, updating the size of the trust region of the circuit design parameter set based on the statistical angle drift in adjacent periods includes:

[0032] Based on the historical statistical angle and historical circuit design parameter set, the historical statistical angle performance parameters are obtained;

[0033] Subtracting the historical statistical angle performance parameters from the actual statistical angle performance parameters yields the difference in actual performance parameters.

[0034] Subtracting the historical statistical angle performance parameters from the predicted statistical angle performance parameters yields the difference in predicted performance parameters.

[0035] The reliability metric index of the trust region is obtained based on the ratio of the difference in actual performance parameters to the difference in predicted performance parameters; the reliability metric index of the trust region is used to describe the degree of statistical angle drift.

[0036] The radius of the trust region is updated based on the trust region reliability metric index and the optimization starting point.

[0037] In another preferred embodiment of this application, updating the radius of the trust region based on the trust region reliability metric index and the target violation distance includes:

[0038] When the target violation distance of the circuit design parameter set in the (k+1)th round is less than the target violation distance under the optimization in the kth round, and the reliability index of the trust region meets the preset requirements, the optimization starting point of the (k+2)th round is determined to be valid, and the radius of the trust region is expanded by the first ratio.

[0039] If the target violation distance of the circuit design parameter set in the (k+1)th round is less than the target violation distance under the optimization in the kth round, and the reliability index of the trust region does not meet the preset requirements, the optimization starting point of the (k+2)th round is determined to be valid, and the radius of the trust region is maintained or reduced according to the second method.

[0040] When the degree of deviation of the target from the circuit design parameter set in the (k+1)th round is greater than the degree of deviation of the target from the optimization in the kth round, the optimization direction in the (k+2)th round is determined to be invalid, and the radius of the trust region is reduced by a third ratio; wherein the third ratio is greater than the second ratio.

[0041] In another preferred embodiment of this application, the step of constructing the real-time trust region for the (k+2)th round using the updated trust region radius includes:

[0042] The circuit design parameter set of the (k+1)th round of optimization is determined as the optimization starting point. Using the optimization starting point as the sampling starting point, sphere sampling is performed with the updated trust region radius to obtain a set of candidate circuit design parameter sets. A surrogate model is constructed based on the simulation information of the kth round of optimization to evaluate the score of each candidate circuit design parameter set.

[0043] Select the top K% of the sampling points in the score to construct the smallest hyperrectangular region surrounding the sampling points, and use the smallest hyperrectangular region as the trust region for parameter optimization in the (k+2)th round.

[0044] In another preferred embodiment of this application, before the optimal circuit design parameter set for output, the following is included:

[0045] Should the yield requirement be adjusted? If yes, use the optimal circuit design parameter set as the initial design parameter set and return to the step of generating the circuit design sample set.

[0046] A second aspect of this application provides an optimal circuit design parameter solving device based on statistical angle drift control, the device comprising:

[0047] The determination module is used to determine the set of circuit design parameters and the set of random process parameters corresponding to the target yield of the finished circuit product.

[0048] An extraction module is used to fix a set of circuit design parameters, take the set of circuit design parameters and the set of random process parameters as inputs, obtain the circuit performance values ​​corresponding to the inputs based on the circuit simulator, generate a set of circuit design samples, and calculate statistical angles based on the set of circuit design samples.

[0049] The optimization module is used to iteratively optimize the circuit design parameter set based on statistical angles. In each cycle, the starting point for optimization of the circuit design parameter set is determined according to the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. The circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met to obtain the optimal circuit design parameter set. The circuit generated by the optimal circuit design parameter set meets the target yield requirement under the perturbation of the process random parameter set.

[0050] The circuit design parameter solving method based on statistical angle drift control provided in this application has the following advantages:

[0051] This application employs a combination of statistical angle drift constraints and trust region guidance to optimize circuit yield design. In existing methods combining Bayesian optimization and Monte Carlo simulation, traditional techniques use fixed statistical angles to solve for the optimal design parameter set during multiple iterations of the circuit design parameter set optimization based on statistical angles. However, the relationships represented by statistical angles differ from the actual correlations in the actual simulation cycle, leading to errors in the optimal design parameter set solution. In the design method and apparatus provided in this application, the statistical angle corresponds to the trust region size of the optimized design in each iteration. Before optimizing the circuit design parameter set in the current cycle, the adaptive customized trust region corresponding to its optimization needs to be determined first. Using the value of the statistical angle calculated in the previous cycle, the trust region radius and other size information of the optimal design parameter set solution in each cycle are corrected by solving the statistical angle drift information, and the optimization starting point is adjusted. The calculation speed of each optimization is improved directly through the constraint of the solution information. Furthermore, by adjusting the solution range, it is ensured that the optimization solution of each iteration is consistent with the actual circuit performance of the current cycle, thereby improving the accuracy of parameter design and realizing the fast and accurate solution of the optimal design parameter set that meets the target yield requirements. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a circuit design parameter solving method based on statistical angle drift control according to an exemplary embodiment;

[0053] Figure 2 This is a schematic diagram illustrating a circuit design parameter solving method based on statistical angle drift control according to an exemplary embodiment.

[0054] Figure 3 This is a block diagram illustrating a circuit design parameter solving device based on statistical angle drift control, according to an exemplary embodiment.

[0055] Figure 4 This is a block diagram of an electronic device illustrating a method for solving circuit design parameters based on statistical angle drift control, according to one embodiment. Detailed Implementation

[0056] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0057] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0058] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0059] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0060] Example 1:

[0061] Figure 1 This is a flowchart of an embodiment of a circuit design parameter solving method based on statistical angle drift control provided in this application. Please refer to... Figure 1 The method provided in this embodiment includes steps S1 to S3:

[0062] Step S1: Determine the circuit design parameter set and process random parameter set corresponding to the target yield of the finished circuit.

[0063] In step S1, the circuit consists of multiple components. Each component has parameters such as size and model that need to be determined, as well as parameters such as the relative positional relationships between components. The combination of these parameters can affect the circuit's performance, ultimately determining whether the circuit can function properly or not. Simultaneously, due to random variations in the manufacturing process, these variations also affect circuit performance, ultimately impacting its functionality. In the mass production of circuits, yield is defined as the probability of meeting design specifications. Target yield refers to the minimum performance qualification rate that the circuit must meet after considering the fluctuations in random process parameters, pre-set during the integrated circuit design phase. The circuit design parameter set refers to the deterministic set of parameters actively set and adjustable by the designer during the integrated circuit design phase based on the circuit's functional and performance requirements. This set includes multiple circuit design parameters, each with a different type. The process random parameter set refers to the set of uncontrollable random variables generated during chip manufacturing processes such as photolithography, etching, and doping. Optionally, the deterministic parameter set may be the aspect ratio of the transistor, the resistance value of the resistor, the capacitance value of the capacitor, etc., and the random variable set may be the transistor threshold voltage fluctuation, the deviation of the effective channel length, the change of oxide layer thickness, etc.

[0064] Optionally, step S1 specifically includes:

[0065] Step S11: Determine the design parameter set and target yield, and initialize the trust region.

[0066] The design parameter set refers to multiple design parameters that need to be optimized and solved simultaneously in each optimization process. The design parameter set includes multiple design parameters that collectively affect the circuit configuration and performance values. Determining the design parameter set at least includes determining the types of design parameters. As an optional implementation, a parameter set whose influence exceeds a preset threshold is selected from all parameters affecting circuit performance as the design parameter set. This further determines the parameter set whose impact on circuit yield exceeds the preset threshold. Determining the design parameter set includes at least determining the values ​​of the initial design parameter set as the starting point for optimization. The method provided by this invention reduces the number of yield calculations under different random design parameter sets by using statistical angles. This ensures that each optimization of the design parameter set only needs to be solved within the range corresponding to the statistical angle, eliminating the need for yield calculations for various random design parameter sets. This naturally guarantees that the circuit performance parameters corresponding to this circuit design parameter set reach the target yield, thereby reducing the process of judging whether each optimized design parameter meets the target yield. The design parameters do not have excessively high setting requirements. Optionally, the design parameters can be transistor aspect ratios, resistance values, etc., as long as they are within the parameter value range for normal circuit simulation. This is used to adjust the value of each design parameter so that the circuit can operate under random process parameters. Even under fluctuations, performance requirements can still be met; an initial target yield is determined. Optionally, the target yield can be set to 3 sigma, i.e., 99.73%. The target yield can be gradually increased in subsequent rounds of higher yield optimization. Optionally, the increase can be to increase the target yield from 99.73% to 4 sigma, i.e., 99.9937%, during the optimization process, which is used as the termination criterion for the yield optimization process; preferably, the trust region is initialized using the entire optimization space; wherein, the optimization space refers to the Cartesian product region formed by the upper and lower limits of the physical values ​​of all design parameters in the circuit design parameter set. The trust region allows the range of design parameter optimization solutions. It is a local parameter search region centered on the current design point and bounded by an adjustable radius, which is used to limit the range of design parameter optimization solutions in the next round, ensuring that the variation of design parameters is within a controllable range.

[0067] Step S12: Determine the set of random process parameters.

[0068] Using sampling methods such as importance sampling, low-discrepancy sampling, and hypersphere sampling, a set of random samples was obtained. The matrix is ​​used as a random parameter for the process. For the dimension of random parameters in the process, The initial number of samples is used to determine the set of process random parameters as the process random parameter set.

[0069] Step S2: Fix the circuit design parameter set, use the circuit design parameter set and the process random parameter set as input, obtain the circuit performance value corresponding to the input based on the circuit simulator, generate the circuit design sample set, and calculate the statistical angle based on the circuit design sample set.

[0070] In step S2, the circuit simulator refers to a computer-aided design tool or software module used to simulate the electrical behavior of integrated circuits under given design parameters and process conditions. Based on the input circuit design parameter set and process random parameter set, it simulates and outputs circuit performance values. The circuit design sample set refers to a set of triplets consisting of the circuit design parameter set, the process random parameter set, and the corresponding circuit performance values. A statistical angle refers to a specific process random parameter point; that is, under given design parameters, the performance value corresponding to this process random parameter point can characterize the target quantile in the overall performance distribution. Specifically, fixing the design parameters of all circuits, including the circuit design parameter set, under the influence of a large number of different process random parameters, the circuit performance will exhibit a cumulative distribution rather than a specific performance value. Finding a performance value in the cumulative distribution that results in a quantile at the target yield—that is, a performance value at this point in the distribution equal to the performance value corresponding to the target yield at the quantile—is the process random parameter point that allows the circuit to achieve that performance value, defined as the statistical angle of the current design parameters. During the optimization process, statistical angles correspond to design parameters and circuit performance parameters. Except for the initial trust region size, the trust region size obtained in each iteration has a corresponding relationship with the statistical angle. Different sets of design parameters correspond to different trust regions, and different iteration cycles correspond to different trust region sizes. The trust region size is customized and adaptively corrected by statistical angle drift, ensuring that the trust region can automatically optimize itself under each combination of design parameters. With the introduction of statistical angles, when solving the design parameters in each iteration, it is no longer necessary to consider the changes in circuit performance or even yield under different random process parameter values. It is only necessary to optimize the design parameters and then solve the latest statistical angle based on the design parameters obtained in the latest iteration. As long as the statistical angle and the target yield satisfy the preset relationship, the circuit corresponding to the design parameter set obtained under various random process parameter values ​​can meet the target yield requirements.

[0071] Preferably, the circuit simulator performs simulations for each round of design parameters and statistical angles, combines the simulation data to obtain samples, and updates the training circuit simulator. Specifically, in the k-th round of black-box optimization, the statistical angles used throughout the optimization of the design parameters from the k-th round remain unchanged. During the process, many candidate design parameter points are obtained. The statistical angle prediction performance values ​​of these candidate design parameter points are obtained through real simulation of the combined samples of design parameters and statistical angles, and used to train the model until the end of this round of optimization, thus obtaining the design parameters for the (k+1)-th round.

[0072] Optionally, step S2 specifically includes:

[0073] Step S21: Using the circuit design parameter set and the process random parameter set as input, obtain the circuit performance values ​​corresponding to the input based on the circuit simulator, and generate a circuit design sample set.

[0074] Using a defined set of circuit design parameters and a defined set of random process parameters as inputs to the circuit simulator, it outputs multiple sets of circuit performance values, which are expressed as follows: ,in, For circuit simulator, A set of design parameters in the circuit design parameter set. It is a set of process random parameters in the process random parameter set.

[0075] By utilizing the input and output relationships of a circuit simulator, a circuit design sample set is obtained. The circuit design sample set includes a circuit design parameter set, a process parameter set, and corresponding output circuit performance values.

[0076] Step S22: Calculate statistical angles based on the circuit design sample set.

[0077] A Gaussian process model was established based on a circuit design sample set.

[0078] The Gaussian process model, under a fixed set of circuit design parameters, describes the complex nonlinear mapping relationship between stochastic process parameters and circuit performance. Bayesian optimization is used to find the potential worst-performing point in the circuit design sample set, within a randomized set of process parameters with N samples. Optionally, N is greater than the initial number of samples. The Gaussian process model is simulated and updated to identify and verify the worst-case potential point. After finding and verifying the worst-case performance point, this point is removed from the search set. This process is repeated until the worst-case point in the set is found. Given a set of performance defects, construct a set of performance defects; where... The scale parameter is between [1,2], and Y is the yield with a value range of [0,1]. The potential point is the worst predicted point by the circuit performance parameter model during the above search process. The potential point includes multiple pieces of information, including at least a set of random process parameters and circuit performance values.

[0079] Based on the set of poorly performing circuits, and taking the difference between the current circuit yield and the target yield as the objective, a differential evolution algorithm is used to obtain the target failure rate, combined with a multi-objective standard deviation balancing term. Regarding the corresponding quantile performance values, this invention does not restrict the differential evolution algorithm; any conventional differential evolution algorithm in the field can be used. Compared to conventional kernel density estimation methods for extracting quantile performance values, this invention establishes an accurate relationship between the Monte Carlo set and the performance distribution through the above steps. The algorithm that best approximates the actual performance... The actual sampling points of the quantile performance value are used as the statistical angles of this performance. When the circuit outputs multiple performance values, a statistical angle is extracted for each value, specifically as follows:

[0080] ;

[0081] Where, arg This refers to the operation of finding the minimum value of the independent variable; it means finding the value f that makes the expression within the parentheses reach its minimum. This refers to the absolute value, used to measure the physical distance or deviation between the predicted performance and the target yield threshold, f. c The performance value is obtained through the differential evolution algorithm. Let Y be the cumulative probability density function of the performance under design parameter x. * For the target yield, λ refers to the balance coefficient, used to adjust the weight of the uncertainty penalty term in the model. This refers to the standard deviation operator, used for balancing the values ​​of multiple indicators. It refers to the performance normalized ratio vector, which represents the ratio of each individual performance indicator to its target threshold.

[0082] Step S2 obtains the mapping data of process parameter set-performance value through simulation and extracts statistical angles, transforming the high-dimensional stochastic yield optimization problem into a low-dimensional deterministic extremum solution problem, laying the data and theoretical foundation for subsequent low-overhead, high-reliability drift control iterative optimization.

[0083] The method provided by this invention establishes a corresponding relationship between circuit performance values ​​and yield. It replaces the traditional process angle representation by employing statistical angles to describe the changes in circuit performance under various random process parameter combinations with a fixed set of design parameters. After the circuit design is completed, the circuit performance parameters can be uniquely used to calculate the yield value, thus revealing the yield variation under the influence of random process parameters. During circuit optimization design, there is no need to repeatedly simulate and verify whether the circuit performance values ​​meet the conditions for each solution's design parameters under different random process parameter sets. As long as the circuit performance value at the statistical angle corresponding to the design parameters and the target threshold corresponding to the target yield meet the preset conditions, the circuits produced under various random process parameter set values ​​will meet the yield requirements.

[0084] Step S3: Iteratively optimize the circuit design parameter set based on statistical angles. In each cycle, determine the optimization starting point of the circuit design parameter set according to the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. Update the size of the trust region of the circuit design parameter set according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. Optimize the circuit design parameter set in the real-time trust region until the iteration stopping condition is met to obtain the optimal circuit design parameter set. The circuit generated with the optimal circuit design parameter set meets the target yield requirement under the perturbation of the process random parameter set.

[0085] Figure 2 A schematic diagram of the optimal circuit design parameter solution method based on statistical angle drift control proposed in this application is shown. (Refer to...) Figure 2 To meet the target yield requirement, iterative optimization of design parameters and extraction of corresponding statistical angles are required. In each iteration, the solution range of the design parameter set is constrained by the trust region, which corresponds to the circuit design parameter set and the iteration cycle. That is, different circuit design parameter sets correspond to different trust regions, except for the initial trust region. Different circuit design parameter sets refer to sets containing different circuit design parameters, which can be different in number or parameter type. It should be noted that the statistical angles mentioned in step S2 are a general term for statistical angle categories. During each iteration, the optimization solution range of the circuit design parameters for the current cycle, i.e., the latest trust region, is first calculated based on historical statistical angles (i.e., the statistical angles of the previous cycle). The circuit design parameters for the current cycle are then solved within this trust region. Next, the true statistical angles are calculated. Calculating the true statistical angles requires returning to the statistical angle extraction step. Based on the newly extracted statistical angles, the optimal combination of all design parameter values ​​in the circuit design parameter set is performed, thus requiring multiple iterations. Besides the initial trust region, the trust regions for different iteration cycles are also different. Optionally, the circuit design parameter set optimization process refers to optimizing the values ​​of each circuit design parameter in the circuit design parameter set. Specifically, it involves first calculating the corresponding statistical angle based on the current circuit design parameter set, then determining whether the target yield requirement is met based on the statistical angle. If not, it proceeds to the next iteration; if it is met, the current circuit design parameter set is taken as the optimal circuit design parameter set. In the next iteration, the drift degree of the statistical angle is first evaluated, and the latest trust region is obtained based on the drift degree. The new circuit design parameter set is then solved within the latest trust region, and the process returns to the step of calculating the corresponding statistical angle based on the current circuit design parameter set.

[0086] In step S3, the threshold is the performance value corresponding to the target yield. The optimization starting point refers to the initial position of the current circuit design parameter set selected at the beginning of each iteration optimization cycle. The initialized trust region refers to the radius of the trust region initialized in step S11, i.e., the initialized trust region is the entire optimization space. Statistical angle drift refers to the deviation between the performance predicted based on the previous statistical angle and the actual statistical angle performance re-extracted at the new design point, which is obtained by using the statistical angle extracted in the previous round to predict and simulate the performance of new candidate design points in the k-th Bayesian parameter optimization iteration. The optimal circuit design parameter set refers to the best combination of circuit design variables finally determined after iterative optimization based on statistical angle drift constraints.

[0087] Optionally, step S3 specifically includes:

[0088] Step S31: Determine whether the circuit design parameter set meets the target yield requirement based on the statistical angle.

[0089] The target yield requirement stipulates that the actual circuit performance value of the current circuit is not lower than the standard circuit performance value corresponding to the target yield. Specifically, the standard circuit performance value corresponding to the target yield is f. * The actual circuit performance value of the current circuit is f. c The target yield requirement is to meet the following conditions:

[0090] ;

[0091] If the yield requirement is met, then determine whether a higher yield requirement is needed. If there is no higher yield requirement, output the circuit design parameters corresponding to the current statistical angle. At this time, the circuit design parameters are the optimal circuit design parameter set. If there is a higher yield requirement, initialize the design parameter set, update the target yield, and return to step S1.

[0092] After calculating the statistical angle based on the latest iteration of circuit design parameters in each round, it is determined whether the circuit performance parameters meet the target yield requirements based on the statistical angle. If the target yield requirements are not met, the current optimization round of the statistical angle is recorded as the kth round, and the optimization proceeds to the k+1th round.

[0093] Step S32: Determine the optimization starting point of the circuit design parameter set based on the difference between the circuit performance value and the threshold corresponding to the circuit design parameter set.

[0094] Optionally, step S32 specifically includes:

[0095] 1. Calculate the target violation distance:

[0096] The set of simulation data with the worst circuit performance was selected based on the calculation of the statistical angle. Calculate the performance gap corresponding to each subset of simulation data;

[0097] Among them, the simulation dataset with the worst circuit performance. This refers to the set of performance gaps in step S22. The performance gap is calculated by subtracting the corresponding circuit performance value from the standard circuit performance value corresponding to the target yield in the subset of simulation data. Specifically, the standard circuit performance value corresponding to the target yield in the subset of simulation data is... The corresponding circuit performance value is ;

[0098] A positive part function is introduced to process the performance gap, retaining only the part greater than zero. The violation value when the actual performance meets the target is set to zero. The violation distance of the circuit design parameter set in the k-th round is obtained by summing all the corresponding performance gaps, as shown in the following figure:

[0099] ;

[0100] in, To violate distance, Due to performance gaps, The standard circuit performance values ​​corresponding to the target yield for the subset of simulation data. For statistical angles The corresponding design parameters below Performance values, The summation is performed after introducing a positive part function to account for the performance gap;

[0101] The weighted threshold for violating the target performance threshold is obtained by multiplying the weighting parameter by the number of violations in the simulation dataset with the worst circuit performance. Specifically, this is expressed as follows: ,in, are weight parameters and , This represents the number of samples in the simulation dataset where performance violations occur, i.e., the number of samples with a performance gap greater than zero.

[0102] The target violation distance is obtained by summing the weighted violation target performance threshold and the violation distance, specifically as follows:

[0103] ;

[0104] in, To achieve a goal that violates distance.

[0105] The target violation distance for the (k+1)th round is obtained using the same steps. .

[0106] 2. Determine the optimization starting point based on the target violation distance:

[0107] Specifically, k refers to the k-th round of Bayesian parameter optimization iteration in step S22, and the value range of k is... The integer is used to compare the target violation distance between the (k+1)th round and the kth round to determine the optimization starting point:

[0108] When the target violation distance in the (k+1)th round is less than the target violation distance under the optimization in the kth round, that is... When that time, the circuit design parameter set of the (k+1)th round is determined as the starting point for optimization;

[0109] When the target violation distance in the (k+1)th round is greater than the target violation distance under the optimization in the kth round, that is... When that time, the circuit design parameter set of the kth round is determined as the starting point for optimization.

[0110] Step S33: Obtain the statistical angle drift in adjacent periods.

[0111] Optionally, it specifically includes:

[0112] 1. Obtain the real-time circuit design parameters for the (k+1)th round of optimization calculation by calculating the statistical angles of the kth round. Statistical angles are extracted in step S22, and real-time statistical angles are calculated. The real-time circuit design parameters and real-time statistical angles are used as inputs to the circuit simulator to obtain the actual statistical angle performance parameters. ;

[0113] 2. Design parameters of real-time circuit for statistical angle prediction based on the k-th round The historical statistics of the kth round With predicted real-time circuit design parameters As input, the circuit simulator obtains the predicted statistical angle performance parameters. ;

[0114] 3. Based on the real-time statistical angle performance parameters and the predicted statistical angle performance parameters, the statistical angle drift is obtained.

[0115] The method provided by this invention relies on the statistical angle extraction process to obtain a set of poorly performing sampled simulation data, and measures the optimization quality of the current circuit design parameters and the reliability of the current trust region. The entire measurement process selects and weights different measurement methods according to the different optimization problems. As an optional embodiment, the measurement methods can include at least the number of violations, violation distance, maximum violation distance, and average violation distance, used to measure the current optimization effect. The entire measurement process does not require re-extracting or resimulating new information; it directly reuses the simulation data from the statistical angle extraction process to measure the overall degree of violation of the performance values ​​corresponding to the current circuit design parameter set relative to the target threshold. This provides a comparative basis for deciding whether to accept subsequent candidate design points, without adding any additional reference data extraction and generation process.

[0116] Step S34: Update the size of the trust region of the circuit design parameter set according to the statistical angle drift in adjacent periods.

[0117] The trust region size information includes at least a center point and a radius. The center point is the optimization starting point determined in the previous steps. The update size referred to in this step can specifically refer to the radius of the trust region of the updated circuit design parameter set.

[0118] Optionally, it specifically includes:

[0119] 1. Solve for the reliability metric index of the trust region.

[0120] Based on the historical statistical angle and historical circuit design parameter sets Historical statistical angle performance parameters were obtained. ;

[0121] The true statistical angle performance parameters Subtracting the historical statistical angle performance parameters yields the difference in the true performance parameters.

[0122] The predicted statistical angle performance parameters Subtracting the historical statistical angle performance parameters yields the difference in predicted performance parameters;

[0123] The reliability metric index of the trust region is obtained based on the ratio of the difference in actual performance parameters to the difference in predicted performance parameters; the reliability metric index of the trust region is used to describe the degree of statistical angle drift.

[0124] The reliability metric index of the trust region is specifically manifested as follows:

[0125] ;

[0126] in, For real-time statistical angle and real-time circuit design parameters The obtained true statistical angular performance parameters, Based on historical statistical perspective and real-time circuit design parameters The obtained predictive statistical angle performance parameters, Based on historical statistical perspective and historical circuit design parameters The obtained historical statistical angle performance parameters, This is a reliability metric index for the trust domain. The preset ideal value is 1. When the value is close to the preset ideal value, it indicates that the statistical angle drift is small and the current trust region has high reliability; when If the value is below the preset threshold, it indicates that the statistical angle prediction error within the current trust region is large. It is necessary to maintain or reduce the radius during the trust region update process to reduce the impact of statistical angle drift on the optimization process.

[0127] 2. Update the radius of the trust region based on the trust region reliability metric index and the violation distance.

[0128] when And the reliability index of the trust region meets the preset requirements, the optimization starting point of the (k+2)th round is determined to be valid, and the radius of the trust region is expanded according to the preset first ratio;

[0129] when If the reliability index of the trust region does not meet the preset requirements, the optimization starting point of the (k+2)th round is determined to be valid, and the radius of the trust region is maintained or reduced according to the preset second ratio.

[0130] when The optimization direction of the (k+2)th round is determined to be invalid, and the radius of the trust region is reduced by a preset third ratio; wherein the preset third ratio is greater than the preset second ratio and the preset first ratio, and the preset first ratio is greater than the preset second ratio.

[0131] Step S35: Based on the optimization starting point and the trust region radius, obtain the real-time trust region through sampling scoring.

[0132] Optionally, it specifically includes:

[0133] 1. Using the optimized starting point as the sampling starting point, perform sphere sampling with the updated trust region radius to obtain a set of candidate circuit design parameters;

[0134] 2. A surrogate model is constructed based on the simulation information from the k-th round of optimization. This surrogate model indicates the potential of the current circuit design parameter set, specifically evaluating the optimization potential of different combinations of circuit design parameter values ​​within the current local region. The surrogate model predicts the mean performance and uncertainty of each candidate circuit design parameter set. The mean performance refers to the predicted value of each candidate circuit design parameter set based on the surrogate model, and the predicted value is unique. The surrogate model is a Gaussian process surrogate model, constructed using a conventional Gaussian process surrogate model construction method, which will not be elaborated upon here. The mean performance and the uncertainty are added together to obtain a potential score.

[0135] 3. Select the top K% of the potential scores from largest to smallest to construct a minimum hyperrectangular region. The minimum hyperrectangular region is the minimum envelope region surrounding the sampling points. Use the minimum hyperrectangular region as the trust region for parameter optimization in the (k+2)th round. Optionally, the top K% is a predetermined value, and the value of K is in the range of 0 < K < 100.

[0136] Step S36: Optimize the circuit design parameters in the real-time trust domain until the iteration stopping condition is met to obtain the optimal circuit design parameters.

[0137] Optionally, it specifically includes:

[0138] In the trust region of the (k+2)th round of parameter optimization in step S35, a set of design parameters x is obtained by sampling. Optionally, the sampling can be a statistical angle obtained by low-difference sampling simulation. The corresponding circuit performance The design parameter x is used as input, and the corresponding circuit performance is used as output. A Gaussian model is fitted, and methods such as Thompson sampling and Pareto front are used to optimize and search for the optimal performance in the current cycle. When the circuit outputs multiple performance parameters (N... f When >1), a Gaussian process model is maintained for each statistical angle. The multi-objective optimization is set as a minimization problem, that is, optimizing the worst-performing point among multiple statistical angles to make it optimal, thereby optimizing the overall circuit performance. Specifically, this is manifested as follows:

[0139] ;

[0140] Where, x * For optimal circuit design parameters;

[0141] The optimization terminates when the simulator reaches its maximum number of simulation iterations or the circuit performance threshold for the target yield is reached, which manifests as follows:

[0142] ;

[0143] Among them, f cThe circuit performance values ​​of the design parameters during the iterative optimization process.

[0144] Before outputting the optimal circuit design parameters, determine whether to adjust the yield requirements:

[0145] If so, use the optimal circuit design parameters as the initial design parameters and return to the step of generating the circuit design sample set;

[0146] If not, output the optimal circuit design parameters.

[0147] Step S3 optimizes the iteration strategy by introducing a trust region update mechanism controlled by statistical angle drift. The trust region size is updated according to the calculated statistical angle drift, ensuring the reliability and stability of the optimization direction. The dynamic process-design coupling relationship is transformed into a controllable local parameter search process, which significantly improves the yield convergence efficiency while ensuring optimization stability.

[0148] In summary, this embodiment provides a circuit design parameter solving method based on statistical angle drift control. Firstly, from the perspective of the entire process, this application first determines the initial design parameter set and target yield, initializes the trust region, and then samples the process random parameter set. Using the initial design parameter set and the process random parameter set as input, the circuit performance values ​​corresponding to the input are obtained based on the circuit simulator, generating a circuit design sample set. Then, based on the circuit design sample set, statistical angles are calculated. The statistical angles are used to describe the relationship between process random parameters and circuit performance values. Next, the circuit design parameter set is iteratively optimized based on the statistical angles. In each cycle, the optimization starting point of the circuit design parameter set is determined based on the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. Finally, the circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met, obtaining the optimal circuit design parameter set. The core method of the entire process is to use statistical angle drift to constrain the trust region update. Throughout the entire process, the worst-case selection is guided by the Gaussian process posterior distribution, avoiding blind traversal of massive process parameters and significantly reducing simulation overhead. Statistical angle extraction simplifies complex process fluctuation scenarios into critical performance sampling points, allowing optimization to focus on extreme failure scenarios and avoiding distractions on qualified parameter points. Drift constraints add constraints to the iteration, preventing erroneous optimization caused by the reuse of old angles. This end-to-end design improves optimization efficiency and controls computational costs through Bayesian optimization, accurately identifies key bottlenecks affecting yield through statistical angle decoupling, ensures optimization reliability through drift constraints, and dynamically adapts to higher yield targets based on optimized parameters. Ultimately, it achieves the target yield with lower resource consumption and higher accuracy, and the optimized parameters exhibit strong stability under process fluctuations.

[0149] Secondly, this approach quantifies the overall violation gap between the current design and the target threshold by constructing a violation distance. Simultaneously, it constructs a reliability metric index for the trust region based on the difference between the statistical angle prediction performance and the reused actual performance. This index is used to jointly determine candidate point acceptance, optimal point update, and trust region expansion, maintenance, or contraction. Furthermore, it automates the calibration of iterative decisions through preset drift and reliability thresholds. The violation distance reuses simulation data from the statistical angle extraction stage, comprehensively considering the number of violation points and the normalized violation distance to accurately quantify the current design's compliance gap. The reliability metric index quantifies the credibility of statistical angle drift, avoiding misjudging false performance improvements caused by statistical angle drift as effective optimization. The joint judgment logic optimistically expands the domain to accelerate the search when performance improves and drift is controllable, and pessimistically contracts the domain to suppress risk when performance does not improve or drift exceeds limits. This approach can accurately identify the effectiveness of optimization directions, solving the problems of incorrect acceptance and invalid search caused by ignoring drift in existing statistical angle schemes, while reducing redundant simulation overhead.

[0150] Thirdly, the simulation data accumulated from the previous round of parameter optimization is reused to fit the surrogate model. A potential score is constructed based on the model's predicted mean performance and uncertainty. After screening high-potential candidate points, a minimum bounding hyperrectangle is constructed as the constraint range for the next round of trust regions. The potential score balances exploration and development by adding the mean performance and uncertainty, avoiding the trap of focusing only on local optima in known high-performance areas. The trust region shape of the minimum bounding hyperrectangle breaks free from the limitation of a fixed sphere, adaptively adjusting the search range to follow high-potential areas, avoiding invalid sampling. The reuse of historical simulation data breaks the limitation of existing schemes where simulation data is only used for single-point evaluation, significantly improving data utilization. This allows for precise identification of design spaces with high improvement potential in the next round of optimization, reducing the sampling of invalid candidate points.

[0151] Example 2:

[0152] Figure 3 This application provides an optimal circuit design parameter solving device based on statistical angular drift constraints, the device comprising:

[0153] Module 310 is used to determine the set of circuit design parameters and the set of random process parameters corresponding to the target yield of the finished circuit product.

[0154] Extraction module 320 is used to take the circuit design parameter set and the process random parameter set as input, obtain the circuit performance value corresponding to the input based on the circuit simulator, generate a circuit design sample set, and calculate statistical angles based on the circuit design sample set;

[0155] The optimization module 330 is used to iteratively optimize the circuit design parameter set based on statistical angles. In each cycle, the optimization starting point of the circuit design parameter set is determined according to the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. The circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met to obtain the optimal circuit design parameter set. The circuit generated by the optimal circuit design parameter set meets the target yield requirement under the perturbation of random process parameters.

[0156] In summary, this embodiment provides an optimal circuit design parameter solving device based on statistical angle drift constraints. Firstly, from the perspective of the entire process, this application first determines the initial design parameter set and target yield, initializes the trust region, and then samples the process random parameter set. Using the design parameter set and the process random parameter set as input, the circuit performance values ​​corresponding to the input are obtained based on a circuit simulator, generating a circuit design sample set. Then, statistical angles are calculated based on the circuit design sample set. Statistical angles are used to describe the relationship between process random parameters and circuit performance values. The circuit design parameter set is then iteratively optimized based on the statistical angles. In each cycle, the optimization starting point of the circuit design parameter set is determined based on the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. Finally, the circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met, obtaining the optimal circuit design parameter set. The core method of the entire process is to use statistical angle drift to constrain trust region updates. Throughout the entire process, the worst-case selection is guided by the Gaussian process posterior distribution, avoiding blind traversal of massive process parameters and significantly reducing simulation overhead. Statistical angle extraction simplifies complex process fluctuation scenarios into critical performance sampling points, allowing optimization to focus on extreme failure scenarios and avoiding distractions on qualified parameter points. Drift constraints add constraints to the iteration, preventing erroneous optimization caused by the reuse of old angles. This end-to-end design improves optimization efficiency and controls computational costs through Bayesian optimization, accurately identifies key bottlenecks affecting yield through statistical angle decoupling, ensures optimization reliability through drift constraints, and dynamically adapts to higher yield targets based on optimized parameters. Ultimately, it achieves the target yield with lower resource consumption and higher accuracy, and the optimized parameters exhibit strong stability under process fluctuations.

[0157] Secondly, this approach quantifies the overall violation gap between the current design and the target threshold by constructing a violation distance. Simultaneously, it constructs a reliability metric index for the trust region based on the difference between the statistical angle prediction performance and the reused actual performance. This index is used to jointly determine candidate point acceptance, optimal point update, and trust region expansion, maintenance, or contraction. Furthermore, it automates the calibration of iterative decisions through preset drift and reliability thresholds. The violation distance reuses simulation data from the statistical angle extraction stage, comprehensively considering the number of violation points and the normalized violation distance to accurately quantify the current design's compliance gap. The reliability metric index quantifies the credibility of statistical angle drift, avoiding misjudging false performance improvements caused by statistical angle drift as effective optimization. The joint judgment logic optimistically expands the domain to accelerate the search when performance improves and drift is controllable, and pessimistically contracts the domain to suppress risk when performance does not improve or drift exceeds limits. This approach can accurately identify the effectiveness of optimization directions, solving the problems of incorrect acceptance and invalid search caused by ignoring drift in existing statistical angle schemes, while reducing redundant simulation overhead.

[0158] Thirdly, the simulation data accumulated from the previous round of parameter optimization is reused to fit the surrogate model. A potential score is constructed based on the model's predicted mean performance and uncertainty. After screening high-potential candidate points, a minimum bounding hyperrectangle is constructed as the constraint range for the next round of trust regions. The potential score balances exploration and development by adding the mean performance and uncertainty, avoiding the trap of focusing only on local optima in known high-performance areas. The trust region shape of the minimum bounding hyperrectangle breaks free from the limitation of a fixed sphere, adaptively adjusting the search range to follow high-potential areas, avoiding invalid sampling. The reuse of historical simulation data breaks the limitation of existing schemes where simulation data is only used for single-point evaluation, significantly improving data utilization. This allows for precise identification of design spaces with high improvement potential in the next round of optimization, reducing the sampling of invalid candidate points.

[0159] Example 3:

[0160] Thirdly, embodiments of this application provide an electronic device, Figure 4 This is a block diagram illustrating an electronic device according to an exemplary embodiment. (e.g.) Figure 4 As shown, the electronic device may include a processor 11 and a memory 12 storing computer program instructions.

[0161] Specifically, the processor 11 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0162] The memory 12 may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory 12 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 12 may include removable or non-removable (or fixed) media. Where appropriate, the memory 12 may be internal or external to a data processing device. In a particular embodiment, the memory 12 is non-volatile memory. In a particular embodiment, the memory 12 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0163] The memory 12 can be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 11.

[0164] The processor 11 reads and executes the computer program instructions stored in the memory 12 to implement any of the circuit design parameter solving methods based on statistical angle drift control in the above embodiments.

[0165] In one embodiment, the electronic device may further include a communication interface 13 and a bus 10. Wherein, as... Figure 4 As shown, the processor 11, memory 12, and communication interface 13 are connected through bus 10 and communicate with each other.

[0166] The communication interface 13 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. The communication port 13 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0167] Bus 10 includes hardware, software, or both, that couples components of an electronic device together. Bus 10 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 10 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 10 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0168] The apparatus of this embodiment can be used to perform... Figure 1 The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0169] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0170] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0171] Fourthly, embodiments of this application provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the method provided in the first aspect for solving circuit design parameters based on statistical angle drift control.

[0172] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0173] In a possible implementation, this application can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, causes the terminal device to perform steps to implement the circuit design parameter solving method based on statistical angle drift control provided in the first aspect.

[0174] The program code for executing this application can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0175] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for solving circuit design parameters based on statistical angle drift control, characterized in that, The method includes: Determine the set of circuit design parameters and the set of random process parameters corresponding to the target yield of the finished circuit product; A fixed set of circuit design parameters is used as input. The circuit performance values ​​corresponding to the input are obtained based on the circuit simulator, a circuit design sample set is generated, and statistical angles are calculated based on the circuit design sample set. The circuit design parameter set is iteratively optimized based on statistical angles. In each cycle, the starting point for optimization of the circuit design parameter set is determined according to the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. The circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met to obtain the optimal circuit design parameter set. The circuit generated by the optimal circuit design parameter set meets the target yield requirement under the perturbation of the process random parameter set.

2. The method according to claim 1, characterized in that, The iterative optimization of the circuit design parameter set based on statistical angles includes: Based on the statistical angle, determine whether the circuit design parameter set meets the yield requirements; If not satisfied, calculate the target violation distance of the circuit design parameter set in the (k+1)th round; Calculate the real-time statistical angle in the (k+1)th round; calculate the circuit performance parameters corresponding to the historical statistical angle and the real-time statistical angle under the circuit design parameter set in the (k+1)th round respectively, and obtain the statistical angle drift; Based on the target violation distance, determine the starting point for optimization in the (k+2)th round; Based on the statistical angle drift, calculate the trust region reliability index, and update the trust region radius based on the trust region reliability index. Starting from the optimization point, the real-time trust region for the (k+2)th round is constructed using the updated trust region radius; The circuit design parameter set is optimized in the real-time trust domain, and the step of calculating the statistical angle based on the circuit design sample set is returned to output the optimal circuit design parameter set.

3. The method according to claim 1, characterized in that, The optimization starting point of the circuit design parameter set is determined based on the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold, including: Based on the calculation of the statistical angle, the set of simulation data with the worst circuit performance is selected, and the performance gap corresponding to each subset of simulation data is calculated. The violation distance of the current circuit design parameter set is obtained by summing the performance gaps. The weighted threshold for violating the target performance threshold is obtained by multiplying the weighting parameter by the number of violations of the target performance threshold in the simulation data set with the worst circuit performance. The target violation distance is obtained by summing the weighted violation target performance threshold and the violation distance.

4. The method according to claim 1, characterized in that, The method of updating the trust region of the circuit design parameter set based on the statistical angle drift in adjacent periods includes: By fixing the random parameter set of the process, the real-time circuit design parameter set and the predicted statistical angle performance parameters of the (k+1)th round of optimization are obtained. The predicted statistical angle performance parameters are calculated based on the real-time circuit design parameter set and the historical statistical angles corresponding to the kth round of optimization. The true statistical angle is recalculated based on the real-time circuit design parameter set. Calculate the performance parameters of the real statistical angle based on the real statistical angle and the real-time circuit design parameters; The statistical angle drift is obtained based on the actual statistical angle performance parameters and the predicted statistical angle performance parameters.

5. The method according to claim 1, characterized in that, The step of determining the optimization starting point of the circuit design parameter set based on the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold includes: The target violation distance in round (k+1) is compared with that in round (k) to determine the optimization starting point. When the target violation distance in round (k+1) is less than the target violation distance under optimization in round (k), the circuit design parameter set in round (k+1) is determined as the optimization starting point. When the target violation distance in round (k+1) is greater than the target violation distance under optimization in round (k), the circuit design parameter set in round (k) is determined as the optimization starting point.

6. The method according to claim 4, characterized in that, The step of obtaining the statistical angle drift based on the true statistical angle performance parameters and the predicted statistical angle performance parameters includes: Based on the historical statistical angle and historical circuit design parameter set, the historical statistical angle performance parameters are obtained; Subtracting the historical statistical angle performance parameters from the actual statistical angle performance parameters yields the difference in actual performance parameters. Subtracting the historical statistical angle performance parameters from the predicted statistical angle performance parameters yields the difference in predicted performance parameters. The reliability index of the trust region is obtained based on the ratio of the difference in actual performance parameters to the difference in predicted performance parameters; the reliability index of the trust region is used to describe the degree of statistical angle drift. The radius of the trust region is updated based on the trust region reliability metric index and the optimization starting point.

7. The method according to claim 6, characterized in that, The step of updating the radius of the trust region based on the trust region reliability metric index and the target violation distance includes: When the target violation distance of the circuit design parameter set in the (k+1)th round is less than the target violation distance under the optimization in the kth round, and the reliability index of the trust region meets the preset requirements, the optimization starting point of the (k+2)th round is determined to be valid, and the radius of the trust region is expanded by the first ratio. If the target violation distance of the circuit design parameter set in the (k+1)th round is less than the target violation distance under the optimization in the kth round, and the reliability index of the trust region does not meet the preset requirements, the optimization starting point of the (k+2)th round is determined to be valid, and the radius of the trust region is maintained or reduced according to the second ratio. When the degree of deviation of the target from the circuit design parameter set in the (k+1)th round is greater than the degree of deviation of the target from the optimization in the kth round, the optimization direction in the (k+2)th round is determined to be invalid, and the radius of the trust region is reduced by a third ratio; wherein the third ratio is greater than the second ratio.

8. The circuit design yield optimization method based on statistical angle drift constraints according to claim 2, characterized in that, The construction of the real-time trust region for the (k+2)th round using the updated trust region radius includes: The circuit design parameter set of the (k+1)th round of optimization is determined as the optimization starting point. Using the optimization starting point as the sampling starting point, sphere sampling is performed with the updated trust region radius to obtain a set of candidate circuit design parameter sets. A surrogate model is constructed based on the simulation information of the kth round of optimization to evaluate the score of each candidate circuit design parameter set. Select the top K% of the sampling points in the score to construct the smallest hyperrectangular region surrounding the sampling points, and use the smallest hyperrectangular region as the trust region for the (k+2)th round of optimization.

9. The method according to claim 2, characterized in that, Prior to the set of optimal output circuit design parameters, the following are included: Should the yield requirement be adjusted? If yes, use the optimal circuit design parameter set as the initial design parameter set and return to the step of generating the circuit design sample set.

10. A circuit design parameter solving device based on statistical angle drift control, characterized in that, The device includes: The determination module is used to determine the set of circuit design parameters and the set of random process parameters corresponding to the target yield of the finished circuit product. An extraction module is used to fix a set of circuit design parameters, take the set of circuit design parameters and the set of random process parameters as inputs, obtain the circuit performance values ​​corresponding to the inputs based on the circuit simulator, generate a set of circuit design samples, and calculate statistical angles based on the set of circuit design samples. The optimization module is used to iteratively optimize the circuit design parameter set based on statistical angles. In each cycle, the starting point for optimization of the circuit design parameter set is determined according to the change in the difference between the circuit performance value corresponding to the circuit design parameter set and the threshold. The size of the trust region of the circuit design parameter set is updated according to the statistical angle drift in adjacent cycles to obtain the real-time trust region. The circuit design parameter set is optimized in the real-time trust region until the iteration stopping condition is met to obtain the optimal circuit design parameter set. The circuit generated by the optimal circuit design parameter set meets the target yield requirement under the perturbation of the process random parameter set.

Images