A method and apparatus for estimating circuit failure probability under random process disturbances

By constructing a failure domain and conditional sampling guided by a rotation matrix, the low sampling efficiency and stability problems of high-dimensional small failure probability estimation in existing technologies are solved, achieving high-precision and high-efficiency circuit failure probability estimation and improving the stability and reliability of integrated circuits.

CN123115976APending Publication Date: 2026-11-17ZHEJIANG JIESHU LAW TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611617056.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In high-dimensional, low-failure-probability estimation scenarios, existing technologies lack the utilization of prior geometric information of the failure domain, resulting in blind sampling direction and coupled parameter updates. This leads to low sampling efficiency, large estimation variation coefficients, and unstable convergence, making it difficult to meet the high-precision and high-efficiency requirements of large-scale integrated circuit yield analysis.

Method used

By constructing a failure domain, the optimal design point and failure direction of random process disturbance parameters are solved, a rotation matrix is ​​constructed for conditional sampling, intermediate events are generated, and the target failure probability estimate is calculated. The geometric information of the failure boundary is used to guide the sampling direction and scaling scale, thereby achieving efficient and accurate failure probability estimation.

Benefits of technology

It improves the accuracy and comprehensiveness of predicting the failure probability of integrated circuits under the influence of random process parameters, ensuring the stability and reliability of integrated circuits, and reducing computing costs and resource waste.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115976A_ABST
    Figure CN123115976A_ABST
Patent Text Reader

Abstract

This application provides a method and apparatus for estimating the failure probability of a circuit under random process disturbances, belonging to the field of integrated circuit design. The method includes: constructing a failure domain of the circuit to be verified based on the distribution of random process disturbance parameters; solving for the optimal design point, reliability index, and failure direction of the parameters in a standard normal space towards the failure domain; constructing a rotation matrix based on the failure direction to align the first coordinate axis with it; calculating a standard deviation scaling factor based on the reliability index; combining the scaling factor and the rotation matrix for conditional sampling to generate multiple intermediate events containing candidate parameter values; and calculating the target failure probability estimate corresponding to the candidate values ​​based on the conditional probability of the intermediate events. This invention significantly improves the accuracy and efficiency of failure probability estimation and reduces computational overhead through rotation alignment and scaling sampling.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of integrated circuit design, and in particular to a method for estimating the probability of circuit failure under random process disturbances. Background Technology

[0002] As integrated circuit manufacturing processes advance into the deep submicron and nanometer stages, the random fluctuations in process parameters such as doping concentration, gate length, channel width, and device mismatch have an increasingly significant impact on chip performance. Circuit yield optimization has become a crucial link between physical design and design for manufacturability, and its estimation efficiency and convergence stability determine the performance, power consumption, and production cost of chip products. In applications such as high-performance computing, mobile communications, and automotive electronics, due to unavoidable random doping fluctuations and line edge roughness during wafer fabrication, circuit performance indicators such as delay, gain, and leakage current exhibit a statistical distribution around their nominal values. Designers need to accurately estimate the failure probability under extreme process perturbations during the design phase to ensure that mass production yields meet stringent requirements. If this process can be automated with high efficiency and accuracy, it can not only significantly shorten the design iteration cycle but also reduce reliance on expensive physical tape-out verification, thereby lowering R&D costs.

[0003] Existing methods for estimating small failure probabilities under extreme process disturbances mainly employ two approaches: The first is based on Monte Carlo and its improved large-scale sampling simulation, which generates massive random samples in the original process distribution, calls circuit simulators such as SPICE point by point, and counts the proportion of failure samples; the second is based on the sequential sampling method of generalized subset simulation, which introduces an intermediate failure threshold and amplifies the standard deviation of the input variables to transform the small failure probability, which is difficult to estimate directly, into a product of a series of intermediate conditional probabilities that are easier to estimate, thereby reducing simulation overhead.

[0004] However, the above two methods have the following problems: 1. Traditional Monte Carlo and its improved methods require a large number of simulations to obtain a stable estimate when the probability of target failure is very small, which is costly; 2. Improved methods such as importance sampling are highly dependent on the setting of the sampling distribution. If the sampling distribution is not set properly, a large number of irrelevant samples will be generated, leading to fluctuations in the estimate; 3. The sampling efficiency of existing generalized subset simulation methods is highly dependent on the accuracy of important failure direction identification; 4. The initial amplification factor of existing generalized subset simulation methods is usually set according to an empirical range, and the subsequent amplification factor is adjusted simultaneously with the intermediate threshold, lacking a clear scale basis related to the distance to the failure boundary; 5. The intermediate event parameter algorithm of existing generalized subset simulation methods involves high-dimensional optimization, and the solution is not unique, making it difficult to guarantee the variance reduction property, which has a negative impact on the stability of the calculation results. Summary of the Invention

[0005] In view of this, this application provides a method and apparatus for estimating the circuit failure probability under random process disturbances, in order to solve the problems in the prior art that, in high-dimensional small failure probability estimation scenarios, the lack of utilization of prior geometric information of the failure domain, blind sampling direction and coupled parameter updates lead to low sampling efficiency, large estimation variation coefficients and unstable convergence, making it difficult to meet the high-precision and high-efficiency requirements of large-scale integrated circuit yield analysis.

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

[0007] The first aspect of this application provides a method for estimating the circuit failure probability under random process disturbances, the method comprising:

[0008] The failure domain of the circuit to be verified is constructed based on the distribution of random process disturbance parameters. The failure domain is where the performance parameters of the circuit to be verified do not meet the target conditions.

[0009] In the standard normal space of the circuit to be verified, the optimal design point, reliability index and failure direction of the random process disturbance parameters are solved in the direction of the failure domain;

[0010] A rotation matrix is ​​constructed based on the failure direction, and the first coordinate axis processed by the rotation matrix is ​​aligned with the failure direction.

[0011] Calculate the scaling factor of the standard deviation based on the reliability index;

[0012] Conditional sampling is performed based on the scaling factor and the rotation matrix to generate multiple intermediate events, wherein the intermediate events include at least candidate values ​​of random process disturbance parameters;

[0013] The estimated target failure probability corresponding to the candidate value is calculated based on the conditional probability of the multiple intermediate events.

[0014] In a preferred embodiment of this application, the step of constructing the failure domain of the circuit to be verified based on the distribution of random process disturbance parameters includes:

[0015] Based on the distribution of the random process disturbance parameters and the simulator interface parameters, the performance parameters of the circuit to be verified are obtained;

[0016] Using the condition that the performance parameters of the circuit to be verified are less than or equal to the target threshold, the set of random process disturbance parameter distributions that satisfy the condition is constructed as the failure domain of the circuit to be verified.

[0017] In another preferred embodiment of this application, the step of solving for the optimal design point, reliability index, and failure direction of the stochastic process disturbance parameters in the standard normal space of the circuit to be verified towards the failure domain includes:

[0018] Using the origin of the failure domain as the initial point, an iterative optimization algorithm is used to search for the failure boundary point closest to the initial point within the failure domain, and the failure boundary point is determined as the optimal design point for the random process disturbance parameters.

[0019] Calculate the reliability index based on the optimal design point;

[0020] The direction from the initial point to the optimal design point is determined as the failure direction.

[0021] In another preferred embodiment of this application, the step of constructing a rotation matrix based on the failure direction, and aligning the first coordinate axis processed by the rotation matrix with the failure direction, includes:

[0022] The Schmidt orthogonalization method is used, with the failure direction as the first coordinate axis basis vector of the rotating coordinate system;

[0023] Construct coordinate axis basis vectors orthogonal to the failure direction, and combine the first coordinate axis basis vector with the coordinate axis basis vector to form an orthogonal rotation matrix.

[0024] In another preferred embodiment of this application, the step of calculating the scaling factor of the standard deviation based on the reliability index includes:

[0025] Multiplying the reliability index by the correction term yields the standard deviation;

[0026] The scaling factor of the standard deviation is calculated based on the reliability index, and the scaling factor is used to scale the standard deviation.

[0027] In another preferred embodiment of this application, the step of calculating the scaling factor of the standard deviation based on the reliability index further includes:

[0028] Obtain the failure boundary geometry correction term;

[0029] The correction term is obtained by adding 1 to the failure boundary geometry correction term.

[0030] In another preferred embodiment of this application, the conditional sampling based on the scaling factor and the rotation matrix includes:

[0031] Generate random samples based on the standard deviation and failure direction of the current layer;

[0032] The performance results of the random sample are obtained by calling the simulator interface parameters.

[0033] In another preferred embodiment of this application, the generation of a plurality of intermediate events, wherein the intermediate events include at least candidate values ​​of random process perturbation parameters, including:

[0034] Based on the sample performance results of the current layer, the intermediate threshold of the current layer is determined according to the sample performance ranking and the preset conditional probability.

[0035] Update the standard deviation based on the characteristics of the random sample;

[0036] Updating the intermediate events includes first determining the current layer's intermediate threshold, and then updating the standard deviation;

[0037] Select a seed sample from the updated intermediate event sample, and perform MCMC conditional sampling with the seed sample as the initial state to obtain candidate values;

[0038] The MCMC conditional sampling process generates intermediate events for the next layer based on the candidate values ​​and the samples from the previous layer when the intermediate threshold of the sampled samples does not reach the target threshold, and updates the standard deviation. When the intermediate threshold of the sampled samples reaches the target threshold, the iteration stops, and the multiple intermediate events generated during the iteration process are obtained.

[0039] In another preferred embodiment of this application, calculating the target failure probability estimate corresponding to the candidate value based on the conditional probability of the plurality of intermediate events includes:

[0040] Calculate the conditional probability of the candidate value corresponding to the plurality of intermediate events, and multiply the conditional probabilities together to obtain the estimated value of the target failure probability.

[0041] A second aspect of this application provides a circuit failure probability estimation device under random process disturbances, the device comprising:

[0042] A construction module is used to construct the failure domain of the circuit to be verified based on the distribution of random process disturbance parameters. The failure domain is where the performance parameters of the circuit to be verified do not meet the target conditions.

[0043] The solver module is used to solve for the optimal design point, reliability index, and failure direction of the random process disturbance parameters in the standard normal space of the circuit to be verified, in the direction of the failure domain.

[0044] A construction module is used to construct a rotation matrix based on the failure direction, wherein the first coordinate axis processed by the rotation matrix is ​​aligned with the failure direction;

[0045] The solver module is used to calculate the scaling factor of the standard deviation based on the reliability index;

[0046] The generation module is used to perform conditional sampling based on the scaling factor and the rotation matrix to generate multiple intermediate events, wherein the intermediate events include at least candidate values ​​of random process disturbance parameters;

[0047] The solution module is used to calculate the target failure probability estimate corresponding to the candidate value based on the conditional probability of the multiple intermediate events.

[0048] This application provides a high-dimensional small failure probability estimation method based on coefficient of variation reduction, which has the following advantages:

[0049] In existing technologies, the estimation of failure probability is often inaccurate regarding whether parameters in the boundary region of the process random parameter set will lead to circuit performance failure. Specifically, it's difficult to predict whether random disturbances in extreme states will cause circuit failure, resulting in poor estimation accuracy for low failure probabilities in high-dimensional cases. The method provided in this application searches for the optimal design point of the process random parameters at the nearest boundary throughout the entire failure domain. Based on the direction between the optimal design point and the origin of the failure domain, as well as the reliability information of the optimal design point's location, the sampling direction and scaling scale are guided. This guides sampling towards the boundary, with different scaling areas at different locations, resulting in a more reasonable sampling distribution covering the area near the failure boundary. Through numerical analysis of the failure boundary geometry, a matching sampling direction and scaling scale are found, guiding sampling at the failure boundary location. This predicts the actual failure probability evolution trend under the current sampling distribution, accurately predicting whether the integrated circuit will fail under the influence of process random parameters. Furthermore, the prediction covers the process random parameters in the boundary region, improving the comprehensiveness of the prediction and ensuring the stability and reliability of the integrated circuit. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating a high-dimensional small failure probability estimation method based on coefficient of variation reduction, according to an exemplary embodiment.

[0051] Figure 2 This is a flowchart illustrating the MCMC sampling process according to an exemplary embodiment.

[0052] Figure 3 This is a block diagram of an apparatus according to an exemplary embodiment.

[0053] Figure 4 This is a block diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0054] 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.

[0055] 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.

[0056] 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."

[0057] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0058] Example 1:

[0059] Figure 1 This is a flowchart of an embodiment of a high-dimensional small failure probability estimation method based on coefficient of variation reduction provided in this application. Please refer to... Figure 1 The method provided in this embodiment includes steps S1 to S6:

[0060] Step S1: Construct the failure domain of the circuit to be verified based on the distribution of random process disturbance parameters. The failure domain is the space formed by random process disturbance parameters whose performance parameters of the circuit to be verified do not meet the target conditions.

[0061] In step S1, the failure domain refers to the set of all parameter combinations that cause the circuit performance to fail to meet the target threshold in the random process parameter space.

[0062] Optionally, step S1 specifically includes:

[0063] The initial parameters are read, preferably including at least: a circuit netlist, the performance indicators to be observed, a target threshold, a distribution of random process disturbance parameters, and simulator interface parameters. Specifically, the circuit netlist includes at least the devices included in the circuit, device parameters, and device connection structures; specifically, it can be topology and device dimensions. The performance indicators to be observed include at least the analog performance indicators and level indicators in the circuit; specifically, they can be gain, frequency, and current. The target threshold includes at least the analog performance threshold in the circuit; specifically, a gain <60dB is considered acceptable. The random process disturbance parameters include at least the parameter type that causes drift and the drift type; specifically, the drift parameter type can be the width and length of the transistor, and the drift type can be Gaussian or uniform distribution. The simulator refers to simulators such as SPICE, used to provide the mapping relationship between input parameters and output performance values.

[0064] It is worth noting that whether the performance parameters of the circuit to be verified meet the target conditions in this step of the application is initially judged based on the simulator. The performance parameters f(u) of the circuit to be verified are obtained based on the distribution of the random process disturbance parameters and the simulator interface parameters.

[0065] In standard normal space, the condition for constructing a failure domain is that the performance parameters of the circuit to be verified are less than or equal to the target threshold. Specifically, the condition for constructing a failure domain is... The performance parameters of the circuit to be verified are: The target threshold is The conditions for constructing the failure domain are expressed in the following form:

[0066] ;

[0067] in, ;

[0068] The set of random process disturbance parameter distributions that satisfy the conditions is used to construct the failure domain of the circuit to be verified.

[0069] Step S1 constructs a failure domain by combining the distribution of random process disturbance parameters with circuit performance failure criteria. This achieves a precise mapping of physical circuit design requirements to standard normal space geometric constraints, clarifies failure boundaries, and provides a standardized benchmark for subsequent geometric prior extraction and directional sampling. This avoids blind sampling and resource waste caused by the ambiguity of the target area definition in traditional random sampling.

[0070] Step S2: Solve for the optimal design point, reliability index and failure direction of the random process disturbance parameters in the standard normal space of the circuit to be verified, in the direction of the failure domain.

[0071] In step S2, the optimal design point is the point in standard normal space that lies on the failure boundary and is closest to the origin. The reliability index is the Euclidean distance from the origin to the optimal design point. The failure direction is the unit vector pointing from the origin to the optimal design point.

[0072] Optionally, step S2 specifically includes:

[0073] Using the origin of the failure domain as the initial point, an iterative optimization algorithm searches for the failure boundary point closest to the initial point within the failure domain, and determines the failure boundary point as the optimal design point for the random process disturbance parameters. The iterative optimization algorithm includes at least a failure domain optimal design point search method and a central difference method, wherein the failure domain optimal design point search method is used to search for failure boundary points. The iterative optimization algorithm uses the central difference method to obtain the gradient of the limit state function, and obtains the performance values ​​of the limit state function at the current point and the disturbance point through batch processing and parallel simulation. Based on the current iteration point, a center point, positive disturbance points of each dimension, and negative disturbance points are constructed, wherein the current iteration point includes at least the failure boundary point. The batch processing combines the center point, positive disturbance points of each dimension, and negative disturbance points into a two-dimensional batch simulation matrix, and obtains the function values ​​and central difference gradients in parallel by calling the simulator interface at once.

[0074] The reliability index is in norm form of the optimal design point. Specifically, the reliability index is... The optimal design point is The norm symbol is The reliability index meets the following conditions:

[0075] ;

[0076] Among them, norm symbol Used to solve for the modulus;

[0077] The direction from the initial point to the optimal design point is defined as the failure direction, which is the ratio of the optimal design point to the reliability index. Specifically, the failure direction is... The failure direction is manifested as follows:

[0078] ;

[0079] Step S2 performs iterative optimization search within the failure domain to obtain the failure boundary point (optimal design point) closest to the origin, and calculates the reliability index and failure direction based on this point. This achieves the analysis of key geometric features of high-dimensional failure boundaries, transforming unbiased sampling of random process parameters into directed guidance and blind exploration into exploration of the most likely failure path. This solves the problems of low sampling efficiency and direction loss caused by unknown failure domain location in circuit simulation.

[0080] Step S3: Construct a rotation matrix based on the failure direction, process the first coordinate axis based on the rotation matrix, and align the first coordinate axis processed by the rotation matrix with the failure direction.

[0081] In step S3, the rotation matrix refers to an orthogonal transformation tool used to twist the standard normal coordinate system to a coordinate system state coaxial with the failure direction.

[0082] Optionally, step S3 specifically includes:

[0083] A first coordinate system is established in the failure domain, and the first coordinate system has a first coordinate axis. Since the coordinate axis in the standard normal space usually makes an angle with the failure direction, if sampling is performed directly in the original coordinate system, random disturbances will be scattered in various dimensions, making it difficult to extend efficiently along the failure boundary. By constructing an orthogonal rotation matrix, the coordinate system is rotated so that the first coordinate axis after rotation is aligned with the failure direction. The construction of the orthogonal rotation matrix includes:

[0084] The Schmidt orthogonalization method is used, with the failure direction as the first coordinate axis basis vector of the rotating coordinate system; the remaining coordinate axis basis vectors orthogonal to the failure direction are constructed; and the first coordinate axis basis vector and the remaining coordinate axis basis vectors are combined column-wise to construct an orthogonal rotation matrix.

[0085] Step S3 constructs an orthogonal rotation matrix based on the failure direction and aligns the rotated first coordinate axis with the failure direction, thereby achieving rigid torsion of the standard normal coordinate system towards the principal axis of the failure domain. By directly aligning the first coordinate axis with the failure direction, the spatial sampling step no longer needs to focus on the failure direction, decoupling the high-dimensional sampling process from the failure direction, simplifying the sampling calculation process, and providing a geometric reference for subsequent directional standard deviation amplification only in the failure-related direction while maintaining compact sampling in other directions. This effectively avoids the blind sampling diffusion and waste of computational resources caused by traditional full-dimensional uniform scaling.

[0086] Step S4: Calculate the scaling factor of the standard deviation based on the reliability index.

[0087] In step S4, the standard deviation is used as a scale parameter to control the spread of the sampling distribution, including a scaling factor and a correction term.

[0088] Optionally, step S4 specifically includes:

[0089] Acquire geometric observation index values of the failure boundary of the failure domain, where the geometric observation index values at least include principal curvature, mean curvature, local quadric surface parameters, boundary distances in different directions, and near-failure sample distance statistics. Specifically, the condition of the failure domain is a limit state function, and for the limit state function the second-order partial derivative is solved to obtain a Hessian matrix, and the Hessian matrix is subjected to feature decomposition in the main failure direction, where the main failure direction is the failure direction in step S2, the obtained eigenvalues are the principal curvatures, and the average value of the eigenvalues is the mean curvature; the local quadric surface parameters may be acquired by taking a plurality of sample points near the failure boundary, where the vicinity of the failure boundary is represented as -σ<g(u)<0 and 0<g(u)<σ, σ is a preset threshold, and the standards for different indexes are different, and optionally, σ may be 0.1; the sample points are approximately fitted by a quadric surface equation, and the coefficients of the quadric surface equation are solved by a least square method, where the coefficients are the local quadric surface parameters; for the different directions, optionally, the sample points may have n variables, the variables may be resistance value, power supply voltage and the like, the n variables constitute n coordinate axes of the n-dimensional space of the rotation matrix, and the positive and negative directions of each coordinate axis are the different directions; the boundary distances in different directions may be acquired by, along a specific coordinate axis direction, increasing the variable value on the basis of the current sample point, so that it is equal to 0, and the variable value in each direction is the boundary distance corresponding to each direction; the near-failure sample distance statistics may be acquired by taking a plurality of sample points near the failure boundary, calculating Euclidean distances from the selected plurality of sample points near the failure boundary to the origin, and solving an average value or solving the nearest distance to serve as the statistics of the boundary distance. The vicinity refers to a first preset range from the boundary, and a failure boundary geometric correction term is determined based on the geometric observation index values of the failure boundary wherein the principal curvature, the mean curvature, the local quadric surface parameters, the boundary distances in different directions or the near-failure sample distance statistics are obtained through observation and calculation, and are used for describing the features of the failure boundary geometric correction term;

[0090] The failure boundary geometric correction term is added by 1 to obtain the correction term, wherein is the failure boundary geometric correction term; the reliability index is multiplied by the correction term to correct the standard deviation, which is specifically represented as:

[0091] ;

[0092] wherein, is the standard deviation, is the reliability index, is a correction term;

[0093] at this time, the failure direction of the first coordinate axis , the remaining directions are forcibly set to avoid sample waste caused by full-dimensional scaling; the reliability index serves as a scaling factor for correcting the standard deviation , and the scaling factor is used to scale the failure boundary scale.

[0094] It should be noted that the reliability index is calculated only once, that is, it is used to set the initial value, and the reliability index is no longer used when updating the standard deviation in subsequent steps.

[0095] In step S4, the scaling factor of the standard deviation is calculated according to the reliability index and the geometric correction term of the failure boundary, and the scaling factor is used to scale the failure boundary scale, so as to realize the calibration of the subsequent sampling scale, achieve the effect of matching the step length of the sampling distribution along the main failure axis direction with the distance of the failure domain and the bending degree of the boundary, avoid the situation that the unit standard deviation is difficult to cross the safe area to reach the far-end failure boundary due to insufficient step length, and prevent sampling divergence and resource waste caused by blind amplification, thereby effectively reducing the coefficient of variation of failure probability estimation and improving the stability of high-dimensional convergence.

[0096] Step S5: Perform conditional sampling according to the scaling factor and the rotation matrix to generate a plurality of intermediate events, where the intermediate events at least include candidate values of random process disturbance parameters.

[0097] In step S5, the intermediate event refers to a constructed sample state. Specifically, the sample state is a sample state about to fail, and optionally, the condition of being about to fail may be -0.1<g(u)≤0; specifically, the conditional sampling is MCMC sampling, the input of the conditional sampling is samples satisfying the intermediate event, and the output is samples of each layer satisfying the intermediate event; the update of the intermediate threshold occurs before the conditional sampling, and is used to gradually approach the real failure boundary to prevent sampling failure caused by an excessively large threshold; the update of the standard deviation occurs after the update of the intermediate threshold and before the conditional sampling, the standard deviation is updated according to the situation of the random samples after conditional sampling, the update of the standard deviation is used to update the scaling factor, and the situation of the random samples refers to the positive and negative condition of the failure domain condition g(u) of each sampling point in the random samples. The candidate value refers to a set of specific parameter combinations generated by the sampling algorithm in the random process disturbance parameter space for testing whether a failure occurs.

[0098] Optionally, step S5 specifically includes:

[0099] The scaling standard deviation of the current layer is kept constant. Random samples are generated based on the scaling standard deviation, and the random sampling method is not limited. The sample distribution is expanded along the failure direction in a rotating coordinate system. The random samples are mapped back to the standard normal space, and simulation is performed using simulator parameters to obtain the performance results of the random samples. Based on the performance results of the random samples of the current layer, the number of performance results is obtained and sorted from largest to smallest. An intermediate threshold for the current layer is determined from largest to smallest according to a preset conditional probability. The intermediate threshold is the target threshold for the random samples of the current layer. The standard deviation is updated according to the distribution of the random samples. The updated standard deviation is based on the hyperparameters of the algorithm, such as the current layer and the target coefficient of variation. The standard deviation is updated to scale the failure boundary. By updating the threshold first and then the standard deviation, the intermediate event progression is matched with the current sampling distribution. The intermediate event refers to the failure determination event between the inside and outside of the failure domain, used to approximate the true failure boundary of the failure domain. The distribution of the random sample is the sign of the failure domain condition g(u) at each sampling point in the random sample. Seed samples are selected from the updated intermediate event samples, and MCMC conditional sampling is performed with the seed samples as the initial state. The sample set of each layer that satisfies the intermediate event is the candidate value. Optionally, the MCMC conditional sampling can adopt the Metropolis-Hastings type conditional sampling method, and the Metropolis-Hastings type conditional sampling method is a publicly disclosed sampling method, which will not be explained here.

[0100] Among them, such as Figure 2 As shown, the MCMC conditional sampling process includes generating the intermediate event of the next layer based on the candidate value and the previous layer sample when the intermediate threshold of the sampled sample does not reach the target threshold, and outputting the conditional probability at this time; updating the standard deviation according to the current layer, the coefficient of variation target, and other internal hyperparameters of the algorithm; stopping the iteration when the intermediate threshold of the sampled sample reaches the target threshold, and obtaining the multiple intermediate events generated during the iteration process. Specifically, the conditional probability refers to the probability of generating the next layer intermediate event under the condition that the previous layer intermediate event occurs; optionally, the solution method of the conditional probability includes: based on the sample set generated by MCMC conditional sampling, calling a circuit simulator to obtain the performance value corresponding to each sample, comparing the simulated performance value with the updated intermediate threshold at this time, if the simulated performance value is less than the updated intermediate threshold, then the sample at this time is determined to be a failed sample, obtaining the number of failed samples in the sample set, and dividing the number of failed samples by the total number of samples to obtain the conditional probability.

[0101] Step S5 constructs a directional scaling sampling distribution by applying the scaling factor to the first coordinate axis after processing with the rotation matrix, and generates candidate values ​​of random process disturbance parameters under this distribution to constitute intermediate events. This achieves the aggregation of samples in the failure-related direction and the compression in the irrelevant direction, which enables the sample points to extend efficiently along the main axis of the failure domain and accurately cover the area near the failure boundary. This reduces blind exploration and invalid simulation in high-dimensional space, thereby significantly improving the convergence speed and computational efficiency of small failure probabilities while ensuring the unbiasedness of the estimation.

[0102] Step S6: Calculate the target failure probability estimate corresponding to the candidate value based on the conditional probabilities of the multiple intermediate events.

[0103] In step S6, the target failure probability estimate refers to the numerical result that characterizes the likelihood of the circuit's performance failing under random process disturbances.

[0104] Optionally, step S6 specifically includes:

[0105] The target failure probability estimate is obtained by multiplying the conditional probabilities together. Specifically, the target failure probability estimate is... The first conditional probability is The number of conditional probabilities is The estimated probability of target failure satisfies the following condition:

[0106] ;

[0107] When outputting the estimated failure probability of the target, statistical information (such as the intermediate threshold of each layer, scaling standard deviation, conditional probability, sample size, optimal design point, reliability index, and important failure direction) is also output.

[0108] Step S6 multiplies the conditional probabilities of intermediate events generated layer by layer by the simulation of the generalized subset, thereby achieving unbiased estimation and statistical uncertainty quantification of the high-dimensional minimal failure probability. This transforms the small failure probability that is difficult to sample directly into the product of multiple easily convergent probability events, significantly reducing the total number of simulation calls and obtaining a final yield evaluation result with both high accuracy and low coefficient of variation, providing circuit designers with a basis for quantifying failure risk.

[0109] In summary, this embodiment provides a method for estimating the circuit failure probability under random process disturbances.

[0110] Firstly, this application, from the perspective of the entire process, first conducts a search for the optimal design point of the failure domain in the preprocessing stage to obtain the failure boundary point closest to the origin, reliability indicators, and important failure directions. Then, it constructs a rotating coordinate system aligned with the direction of the optimal design point, initializes the scaling standard deviation based on the reliability indicators and the geometric correction term of the failure boundary, and then updates the intermediate thresholds and scaling coefficients layer by layer using a threshold-first strategy. It recursively derives the conditional probabilities of each layer through MCMC conditional sampling, and finally multiplies the conditional probabilities of each layer to obtain the target small failure probability estimate. In the entire process, the search for the optimal design point of the failure domain accurately locates the representative direction connecting the high-probability region and the failure domain by searching for the failure boundary point closest to the origin, avoiding the problem of ineffective sample diffusion in irrelevant dimensions due to inaccurate direction identification in existing technologies. The directional scaling constructs a rotating coordinate system based on this direction, amplifying the standard deviation only along the failure-related direction, concentrating limited simulation resources on regions more likely to be close to the failure boundary, significantly improving sampling effectiveness. This end-to-end design not only improves the accuracy of direction recognition by guiding the process through the optimal design point, but also reduces the complexity and instability of joint parameter search through the threshold-priority update mechanism, ultimately achieving the estimation of small failure probability of high-dimensional circuits with fewer samples and higher stability.

[0111] Secondly, an initial scaling scale is determined based on the distance to the failure boundary and its geometric or statistical information. The initial scaling coefficient is constructed by combining the reliability index β with the geometric correction term of the failure boundary, and is adaptively adjusted through this correction term. This initialization mechanism uses a theoretically or statistically based failure distance scale instead of empirically or arbitrarily set scaling coefficients. This avoids excessive scaling, which would cause samples to concentrate near the origin and require additional intermediate layers, or excessive scaling, which would cause samples to spread excessively into irrelevant regions. The geometric correction term of the failure boundary can be determined based on principal curvature, average curvature, local quadratic surface parameters, etc. When the boundary is close to a plane, the correction term approaches zero. When the boundary nonlinearity increases, the initial scaling scale is adjusted according to the degree of curvature, so that the sampling distribution forms a more reasonable spatial match with the target failure boundary. This reduces trial-and-error parameter tuning and the number of intermediate layers, improves the effectiveness and stability of the probability correction factor estimation for each layer, and reduces the overall computational cost.

[0112] Thirdly, a threshold-first intermediate event update mechanism is adopted, which prioritizes determining intermediate thresholds based on the performance ranking of current layer samples and preset conditional probabilities, and then updates the scaling standard deviation according to the current layer sampling situation. This mechanism matches the advancement of intermediate events with the adjustment of sampling distribution, avoiding insufficient effective seeds due to excessively rapid threshold advancement, and also avoiding excessive subset layers and error accumulation due to excessively slow advancement. Simultaneously, the threshold-first criterion is applied to a generalized subset simulation sampling framework that does not rely on a Bayesian surrogate model, avoiding additional uncertainty introduced by surrogate model prediction errors. Intermediate events are updated directly based on circuit simulation results and current layer sample statistics, preserving the reliability of real simulation while introducing a parameter update order more conducive to controlling the coefficient of variation. This reduces the complexity and instability of joint parameter search in high-dimensional space, makes the update objectives of each stage clear, and improves the stability and computational efficiency of the estimation results.

[0113] Example 2:

[0114] like Figure 3 As shown in the figure, this application provides a high-dimensional small failure probability estimation device based on coefficient of variation reduction. The device includes:

[0115] The construction module 310 is used to construct the failure domain of the circuit to be verified based on the distribution of random process disturbance parameters. The failure domain is where the performance parameters of the circuit to be verified do not meet the target conditions.

[0116] The solver module 330 is used to solve for the optimal design point, reliability index and failure direction of the random process disturbance parameters in the standard normal space of the circuit to be verified in the direction of the failure domain;

[0117] Construction module 310 is used to construct a rotation matrix based on the failure direction, wherein the first coordinate axis processed by the rotation matrix is ​​aligned with the failure direction;

[0118] Solver module 330 is used to calculate the scaling factor of the standard deviation based on the reliability index;

[0119] The generation module 320 is used to perform conditional sampling based on the scaling factor and the rotation matrix to generate multiple intermediate events, wherein the intermediate events include at least candidate values ​​of random process disturbance parameters;

[0120] The solver module 330 is used to calculate the target failure probability estimate corresponding to the candidate value based on the conditional probability of the multiple intermediate events.

[0121] In summary, this embodiment provides a circuit failure probability estimation device under random process disturbances.

[0122] Firstly, this application, from the perspective of the entire process, first conducts a search for the optimal design point of the failure domain in the preprocessing stage to obtain the failure boundary point closest to the origin, reliability indicators, and important failure directions. Then, it constructs a rotating coordinate system aligned with the direction of the optimal design point, initializes the scaling standard deviation based on the reliability indicators and the geometric correction term of the failure boundary, and then updates the intermediate thresholds and scaling coefficients layer by layer using a threshold-first strategy. It recursively derives the conditional probabilities of each layer through MCMC conditional sampling, and finally multiplies the conditional probabilities of each layer to obtain the target small failure probability estimate. In the entire process, the search for the optimal design point of the failure domain accurately locates the representative direction connecting the high-probability region and the failure domain by searching for the failure boundary point closest to the origin, avoiding the problem of ineffective sample diffusion in irrelevant dimensions due to inaccurate direction identification in existing technologies. The directional scaling constructs a rotating coordinate system based on this direction, amplifying the standard deviation only along the failure-related direction, concentrating limited simulation resources on regions more likely to be close to the failure boundary, significantly improving sampling effectiveness. This end-to-end design not only improves the accuracy of direction recognition by guiding the process through the optimal design point, but also reduces the complexity and instability of joint parameter search through the threshold-priority update mechanism, ultimately achieving the estimation of small failure probability of high-dimensional circuits with fewer samples and higher stability.

[0123] Secondly, an initial scaling scale is determined based on the distance to the failure boundary and its geometric or statistical information. The initial scaling coefficient is constructed by combining the reliability index β with the geometric correction term of the failure boundary, and is adaptively adjusted through this correction term. This initialization mechanism uses a theoretically or statistically based failure distance scale instead of empirically or arbitrarily set scaling coefficients. This avoids excessive scaling, which would cause samples to concentrate near the origin and require additional intermediate layers, or excessive scaling, which would cause samples to spread excessively into irrelevant regions. The geometric correction term of the failure boundary can be determined based on principal curvature, average curvature, local quadratic surface parameters, etc. When the boundary is close to a plane, the correction term approaches zero. When the boundary nonlinearity increases, the initial scaling scale is adjusted according to the degree of curvature, so that the sampling distribution forms a more reasonable spatial match with the target failure boundary. This reduces trial-and-error parameter tuning and the number of intermediate layers, improves the effectiveness and stability of the probability correction factor estimation for each layer, and reduces the overall computational cost.

[0124] Thirdly, a threshold-first intermediate event update mechanism is adopted, which prioritizes determining intermediate thresholds based on the performance ranking of current layer samples and preset conditional probabilities, and then updates the scaling standard deviation according to the current layer sampling situation. This mechanism matches the advancement of intermediate events with the adjustment of sampling distribution, avoiding insufficient effective seeds due to excessively rapid threshold advancement, and also avoiding excessive subset layers and error accumulation due to excessively slow advancement. Simultaneously, the threshold-first criterion is applied to a generalized subset simulation sampling framework that does not rely on a Bayesian surrogate model, avoiding additional uncertainty introduced by surrogate model prediction errors. Intermediate events are updated directly based on circuit simulation results and current layer sample statistics, preserving the reliability of real simulation while introducing a parameter update order more conducive to controlling the coefficient of variation. This reduces the complexity and instability of joint parameter search in high-dimensional space, makes the update objectives of each stage clear, and improves the stability and computational efficiency of the estimation results.

[0125] Example 3:

[0126] 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.

[0127] 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.

[0128] 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.

[0129] The memory 12 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 11.

[0130] The processor 11 reads and executes computer program instructions stored in the memory 12 to achieve... Figure 4 Any of the circuit failure probability estimation methods under random process disturbances in the above embodiments.

[0131] In one embodiment, the electronic device may further include a communication interface 13 and a bus 10. As shown, the processor 11, memory 12, and communication interface 13 are connected via the bus 10 and communicate with each other.

[0132] 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.

[0133] 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.

[0134] 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.

[0135] 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.

[0136] 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.

[0137] Fourthly, embodiments of this application provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the circuit failure probability estimation method provided in the first aspect under a random process disturbance.

[0138] 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.

[0139] In a possible implementation, this application can also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform steps implementing the circuit failure probability estimation method under random process disturbances provided in the first aspect.

[0140] The program code for executing this application can be written in any combination of one or more programming languages, and 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.

[0141] 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 estimating the circuit failure probability under random process disturbances, characterized in that, The method includes: The failure domain of the circuit to be verified is constructed based on the distribution of random process disturbance parameters. The failure domain is where the performance parameters of the circuit to be verified do not meet the target conditions. In the standard normal space of the circuit to be verified, the optimal design point, reliability index and failure direction of the random process disturbance parameters are solved in the direction of the failure domain; A rotation matrix is ​​constructed based on the failure direction, and the first coordinate axis processed by the rotation matrix is ​​aligned with the failure direction. Calculate the scaling factor of the standard deviation based on the reliability index; Conditional sampling is performed based on the scaling factor and the rotation matrix to generate multiple intermediate events, wherein the intermediate events include at least candidate values ​​of random process disturbance parameters; The estimated target failure probability corresponding to the candidate value is calculated based on the conditional probability of the multiple intermediate events.

2. The method according to claim 1, characterized in that, The construction of the failure domain of the circuit to be verified based on the distribution of random process disturbance parameters includes: Based on the distribution of the random process disturbance parameters and the simulator interface parameters, the performance parameters of the circuit to be verified are obtained; Using the condition that the performance parameters of the circuit to be verified are less than or equal to the target threshold, the set of random process disturbance parameter distributions that satisfy the condition is constructed as the failure domain of the circuit to be verified.

3. The method according to claim 1, characterized in that, The process of solving for the optimal design point, reliability index, and failure direction of the stochastic process disturbance parameters in the standard normal space of the circuit to be verified, towards the failure domain, includes: Using the origin of the failure domain as the initial point, an iterative optimization algorithm is used to search for the failure boundary point closest to the initial point within the failure domain, and the failure boundary point is determined as the optimal design point for the random process disturbance parameters. Calculate the reliability index based on the optimal design point; The direction from the initial point to the optimal design point is determined as the failure direction.

4. The method according to claim 1, characterized in that, The step of constructing a rotation matrix based on the failure direction, wherein the first coordinate axis processed by the rotation matrix is ​​aligned with the failure direction, includes: The Schmidt orthogonalization method is used, with the failure direction as the first coordinate axis basis vector of the rotating coordinate system; Construct coordinate axis basis vectors orthogonal to the failure direction, and combine the first coordinate axis basis vector with the coordinate axis basis vector to form an orthogonal rotation matrix.

5. The method according to claim 1, characterized in that, The scaling factor for calculating the standard deviation based on the reliability index includes: Multiplying the reliability index by the correction term yields the standard deviation; The scaling factor of the standard deviation is calculated based on the reliability index, and the scaling factor is used to scale the standard deviation.

6. The method according to claim 1, characterized in that, The scaling factor for calculating the standard deviation based on the reliability index further includes: Obtain the failure boundary geometry correction term; The correction term is obtained by adding 1 to the failure boundary geometry correction term.

7. The method according to claim 1, characterized in that, The conditional sampling based on the scaling factor and the rotation matrix includes: Generate random samples based on the standard deviation and failure direction of the current layer; The performance results of the random sample are obtained by calling the simulator interface parameters.

8. The method according to claim 1, characterized in that, The generation of multiple intermediate events, wherein each intermediate event includes at least candidate values ​​for random process perturbation parameters, including: Based on the sample performance results of the current layer, the intermediate threshold of the current layer is determined according to the sample performance ranking and the preset conditional probability. Update the standard deviation based on the characteristics of the random sample; Updating the intermediate events includes first determining the current layer's intermediate threshold, and then updating the standard deviation; Select a seed sample from the updated intermediate event sample, and perform MCMC conditional sampling with the seed sample as the initial state to obtain candidate values; The MCMC conditional sampling process generates intermediate events for the next layer based on the candidate values ​​and the previous layer samples when the intermediate threshold of the sampled samples does not reach the target threshold, and updates the standard deviation. When the intermediate threshold of the sampled samples reaches the target threshold, the iteration stops, and the multiple intermediate events generated during the iteration process are obtained.

9. The method according to claim 1, characterized in that, The step of calculating the target failure probability estimate corresponding to the candidate value based on the conditional probability of the plurality of intermediate events includes: Calculate the conditional probability of the candidate value corresponding to the plurality of intermediate events, and multiply the conditional probabilities together to obtain the estimated value of the target failure probability.

10. A device for estimating the probability of circuit failure under random process disturbances, characterized in that, The device includes: A construction module is used to construct the failure domain of the circuit to be verified based on the distribution of random process disturbance parameters. The failure domain is where the performance parameters of the circuit to be verified do not meet the target conditions. The solver module is used to solve for the optimal design point, reliability index, and failure direction of the random process disturbance parameters in the standard normal space of the circuit to be verified, in the direction of the failure domain. A construction module is used to construct a rotation matrix based on the failure direction, wherein the first coordinate axis processed by the rotation matrix is ​​aligned with the failure direction; The solver module is used to calculate the scaling factor of the standard deviation based on the reliability index; The generation module is used to perform conditional sampling based on the scaling factor and the rotation matrix to generate multiple intermediate events, wherein the intermediate events include at least candidate values ​​of random process disturbance parameters; The solution module is used to calculate the target failure probability estimate corresponding to the candidate value based on the conditional probability of the multiple intermediate events.