Circuit yield analysis method, device, storage medium and electronic device
By using a pre-trained radial basis regression model and a multi-layer perceptron neural network, combining uniform and normal distribution sampling, replacing SPICE simulation, the problem of time-consuming and cost-effective SRAM circuit yield analysis is solved, achieving high efficiency and accuracy Circuit quality assessment.
Patent Information
- Application Number
- CN202411112521.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-14
AI Technical Summary
In the prior art, SRAM circuit yield analysis relies on costly and time-consuming SPICE simulation, making it difficult to efficiently conduct circuit quality evaluation.
Using a pre-trained radial basis regression model and multi-layer perceptron neural network, the sample set obtained by uniform sampling and normal distribution sampling is predicted, and the circuit yield is determined based on the failure threshold. Traditional SPICE simulation method.
It significantly shortens the circuit yield analysis time, improves analysis efficiency and accuracy, reduces costs, and ensures the accuracy of prediction results.
Smart Images

Figure CN118627451B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of circuit technology, and in particular to a circuit yield analysis method, device, storage medium, and electronic device. Background Art
[0002] Static Random Access Memory (SRAM) is a memory with static access function, which can save the data stored in it without refreshing the circuit. The yield of SRAM is an important indicator to measure the manufacturing quality of SRAM. It reflects the proportion of SRAM that can meet specific standards or specifications during the manufacturing process. The yield directly affects the cost, reliability and market competitiveness of SRAM.
[0003] At present, the yield analysis of SRAM is usually carried out by using SPICE simulation software to perform millions of Monte Carlo simulations on the SRAM circuit. However, due to the high cost of SPICE-level simulation, it takes a very long time to use this method to perform yield analysis on SRAM. Summary of the invention
[0004] In view of this, embodiments of the present application provide a circuit yield analysis method, device, storage medium, and electronic device to at least partially solve the above problems.
[0005] According to a first aspect of an embodiment of the present application, a circuit yield analysis method is provided, comprising: obtaining circuit data of each target circuit included in a target circuit set, wherein the circuit data of each target circuit includes process parameters of a MOS tube in the target circuit; inputting the circuit data of each target circuit into a pre-trained target radial basis regression model, and outputting intermediate prediction results corresponding to each target circuit, wherein the target radial basis regression model is trained by a first sample circuit set obtained by uniform sampling; inputting each intermediate prediction result into a pre-trained target multilayer perceptron neural network, and outputting prediction results corresponding to each target circuit, wherein the target multilayer perceptron neural network is trained by a second sample circuit set obtained by normal distribution sampling; and determining the yield of the target circuit in the target circuit set according to the prediction results and failure thresholds corresponding to each target circuit.
[0006] According to the second aspect of the embodiment of the present application, a circuit yield analysis device is provided, comprising an acquisition module, a first prediction module, a second prediction module and an analysis module. The acquisition module is used to acquire the circuit data of each target circuit contained in the target circuit set, wherein the circuit data of each target circuit includes the process parameters of the MOS tube in the target circuit; the first prediction module is used to input the circuit data of each target circuit into a pre-trained target radial basis regression model, and output the intermediate prediction results corresponding to each target circuit, and the target radial basis regression model is obtained by training the first sample circuit set obtained by uniform sampling; the second prediction module is used to input each intermediate prediction result into a pre-trained target multi-layer perceptron neural network, and output the prediction results corresponding to each target circuit, and the target multi-layer perceptron neural network is obtained by training the second sample circuit set obtained by normal distribution sampling; the analysis module is used to determine the yield of the target circuit in the target circuit set according to the prediction results corresponding to each target circuit and the failure threshold.
[0007] According to a third aspect of an embodiment of the present application, a computer storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method described in the first aspect is implemented.
[0008] According to the fourth aspect of an embodiment of the present application, there is provided an electronic device, including a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus; the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect.
[0009] The embodiments of the present application provide a circuit yield analysis method, device, storage medium and electronic device, wherein the circuit yield analysis method comprises: obtaining circuit data of each target circuit included in a target circuit set, wherein the circuit data of each target circuit comprises process parameters of a MOS tube in the target circuit; inputting the circuit data of each target circuit into a pre-trained target radial basis regression model, and outputting intermediate prediction results corresponding to each target circuit, wherein the target radial basis regression model is trained by a first sample circuit set obtained by uniform sampling; inputting each intermediate prediction result into a pre-trained target multi-layer perceptron neural network, and outputting prediction results corresponding to each target circuit, wherein the target multi-layer perceptron neural network is trained by a second sample circuit set obtained by normal distribution sampling; and determining the yield of the target circuit in the target circuit set according to the prediction results corresponding to each target circuit and the failure threshold.
[0010] In the present application, the target radial basis regression model is obtained by training the first sample circuit set obtained by uniform sampling, and the target multilayer perceptron neural network is obtained by training the second sample circuit set obtained by normal distribution sampling. Then, the process parameters of the MOS tube in the target circuit are predicted directly by the complete model constructed by the target radial basis regression model and the target multilayer perceptron neural network to obtain the prediction result, and then whether the target circuit fails is determined based on the prediction result and the preset failure threshold, so as to determine the yield of the circuit in the target circuit data set. By replacing SPICE simulation, the present application shortens the yield analysis time, improves the analysis efficiency, and reduces the cost at the same time.
[0011] In addition, in the embodiment of the present application, the intermediate prediction results of each target circuit are first obtained according to the target radial basis regression model, and then the intermediate prediction results are corrected by the target multilayer perceptron neural network to obtain the final prediction results of each target circuit. On the one hand, since the training samples for training the target radial basis regression model are obtained based on uniform sampling of the process parameter variables, it can be ensured that the training samples can uniformly cover the entire fluctuation space of the process parameter variables, so that in the inference stage, whether for qualified circuits whose actual circuit responses meet the requirements or critical circuits whose actual circuit responses are near the failure threshold, the target radial basis regression model can output relatively accurate prediction values; on the other hand, considering that the circuit data of the target circuit to be analyzed in the actual inference stage usually conforms to the normal distribution sampling law, if the prediction results are directly obtained by the target radial basis regression model, there will be a problem of inconsistency between the distribution of the model training samples and the distribution of the variables to be predicted in the model inference stage, thereby limiting the accuracy of the prediction results. Based on the above reasons, in the embodiment of the present application, a target multilayer perceptron neural network obtained by normal distribution sampling training is further introduced, so that the prediction results output by the target radial basis regression model are further corrected by the network to correct the prediction result errors caused by different data distributions. In summary, the embodiment of the present application can effectively improve the accuracy of the circuit yield analysis results. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0013] Figure 1 is a flow chart of the steps of a circuit yield analysis method according to an exemplary embodiment of the present application;
[0014] Figure 2is a flowchart of a circuit yield analysis method according to another exemplary embodiment of the present application;
[0015] Figure 3 is a structural block diagram of a circuit yield analysis device according to an exemplary embodiment of the present application;
[0016] Figure 4 It is a schematic structural diagram of an electronic device according to an exemplary embodiment of the present application. DETAILED DESCRIPTION
[0017] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.
[0018] Before describing the circuit yield analysis method according to the embodiment of the present application, the application scenario of the circuit yield analysis method is briefly described to facilitate understanding.
[0019] Static Random Access Memory (SRAM) is a memory with static access function, which can save the data stored in it without refreshing the circuit. The yield of SRAM is an important indicator to measure the manufacturing quality of SRAM. It reflects the proportion of SRAM that can meet specific standards or specifications during the manufacturing process. The yield directly affects the cost, reliability and market competitiveness of SRAM.
[0020] In the prior art, the yield analysis of SRAM is usually performed by simulating the circuit of SRAM using SPICE simulation software, and then the yield of SRAM is obtained based on the simulation circuit analysis. However, it takes a long time to perform the yield analysis of SRAM using this method, and the cost of using SPICE simulation is high. In view of this, the embodiments of the present application provide a circuit yield analysis method, device, storage medium and electronic device to solve the above technical problems.
[0021] Reference Figure 1 , shows a step flow chart of a circuit yield analysis method according to an exemplary embodiment of the present application.
[0022] The circuit yield analysis method of this embodiment specifically includes the following steps:
[0023] S102: Acquire circuit data of each target circuit included in the target circuit set.
[0024] The circuit data of each target circuit includes process parameters of all MOS tubes in the target circuit.
[0025] Exemplarily, the target circuit set may be an SRAM circuit set, and each target circuit in the target circuit set has the same circuit structure. Each target circuit may include multiple MOS tubes, and each MOS tube may include multiple process parameters, such as threshold voltage, gate-drain resistance, current density, electron mobility, etc. Among them, the threshold voltage refers to the voltage value at which the MOS tube starts to conduct when the gate voltage reaches a certain value. The gate-drain resistance refers to the resistance value between the gate and the drain. Current density refers to the current flowing through a unit area. Electron mobility refers to the speed at which electrons move under the action of an electric field. The circuit data of each target circuit is an input variable , is a multidimensional matrix.
[0026] S104 , inputting the circuit data of each target circuit into a pre-trained target radial basis regression model, and outputting an intermediate prediction result corresponding to each target circuit.
[0027] The target radial basis regression model is obtained by training the first sample circuit set obtained by uniform sampling.
[0028] Exemplarily, the circuit data of the target circuit is calculated by the target radial basis regression model, and the output intermediate prediction result is a one-dimensional number. The intermediate prediction result represents the performance index of the target circuit, such as circuit delay, etc. In order to enable the collected sample data to cover the entire fluctuation space of the circuit data of the target circuit, this embodiment uses the first sample circuit set obtained by uniform sampling for model training to obtain the target radial basis regression model. Uniform sampling can be implemented by a uniform sampling generator, and the uniform sampling generator is implemented by the Latin hypercube algorithm. Through the uniform sampling generator, a first sample circuit data set that obeys a uniform distribution is generated.
[0029] In an optional implementation, the target radial basis regression model includes a target linear function and a target radial basis function. The circuit data of each target circuit is input into a pre-trained target radial basis regression model, and an intermediate prediction result corresponding to each target circuit is output, including: inputting the circuit data of each target circuit into the target linear function for linear fitting, and outputting an initial prediction result; inputting the initial prediction result into the target radial basis function for result correction, and outputting an intermediate prediction result corresponding to each target circuit.
[0030] Exemplarily, the calculation formula of the target radial basis regression model is as follows:
[0031]
[0032]
[0033]
[0034] in, is the target linear function, is the target radial basis function, is the input variable The dimension of It is the product of the number of MOS tubes in a target circuit and the number of process parameters of each MOS tube. yes A single variable (i.e. a single process parameter) The coefficient of is a constant, yes The corresponding weight coefficient is, is the number of interpolation samples of the target radial basis function, that is, the number of training samples of the target radial basis regression model, It is an adjustment factor used to control the granularity of RBF interpolation, that is, the sample density, which is a preset fixed value. The total number of target circuits included in the target circuit set.
[0035] The circuit data of each target circuit The target linear function is input for linear fitting calculation, and the initial prediction result obtained is a one-dimensional prediction value; the initial prediction result is input into the target radial basis function for result correction calculation, and the intermediate prediction result corresponding to each target circuit is obtained.
[0036] In this implementation, a target linear function and a target radial basis function are used to construct a target radial basis model. During the training process, the input process parameter variables can be linearly fitted first, and then corrections can be made based on the obtained linear fitting results, thereby reducing the number of training samples and accelerating model convergence.
[0037] S106, inputting each intermediate prediction result into a pre-trained target multi-layer perceptron neural network, and outputting the prediction result corresponding to each target circuit.
[0038] The target multilayer perceptron neural network is trained by a second sample circuit set obtained by sampling from a normal distribution.
[0039] For example, the target multilayer perceptron neural network can be constructed as a 3-layer neural network, where the number of neurons in the first input layer can be equal to the input variable The number of neurons in the middle hidden layer can be twice the number of neurons in the input layer, and because it is a regression model, the number of neurons in the third output layer is 1. Since the circuit data of the target circuit obeys the normal distribution in the actual application scenario of the present method, therefore, when performing multi-layer perceptron neural network training, the present embodiment uses the second sample circuit set obtained by normal distribution sampling to train the target multi-layer perceptron neural network, thereby compensating for the prediction deviation caused by the distribution of the training samples used in the training of the target radial basis regression model being different from the data distribution in the actual application scenario, and improving the accuracy of the model prediction. The normal distribution sampling method of the present embodiment can be to convert the first sample circuit set obtained by uniform sampling into the second sample circuit set obeying the normal distribution through the Monte Carlo algorithm, or to use other algorithms to implement normal distribution sampling, and the present embodiment does not limit this.
[0040] S108. Determine the yield of the target circuits in the target circuit set according to the prediction results and failure thresholds corresponding to the target circuits.
[0041] Exemplarily, the failure threshold can be flexibly set by those skilled in the art according to actual conditions, and this embodiment does not limit this. The prediction results corresponding to each target circuit are compared with the failure threshold, the number of non-failed circuits in each target circuit is determined, and the ratio of the number of non-failed circuits to the total number of target circuits in the target circuit set is calculated to obtain the yield of the target circuit in the target circuit set.
[0042] In an optional implementation, the yield of the target circuits in the target circuit data set is determined based on the prediction results and failure thresholds corresponding to each target circuit, including: determining a target number of target circuits whose prediction results are less than the failure threshold based on the prediction results and failure thresholds corresponding to each target circuit; and determining the yield of the target circuits in the target circuit set based on the target number and the total number of target circuits in the target circuit set.
[0043] Exemplarily, the yield calculation formula of the target circuit is as follows:
[0044]
[0045] in, is the total number of target circuits contained in the target circuit set, can be the failure threshold, Indicates the prediction results corresponding to each target circuit. When the prediction result corresponding to the target circuit is greater than When the target circuit is a failed circuit, the prediction result corresponding to the target circuit is less than , it indicates that the target circuit is a non-failure circuit. Function Is the indicator function of failure, that is:
[0046]
[0047] In this implementation, the target number of target circuits whose prediction results are less than the failure threshold, that is, the number of non-failed circuits in each target circuit, is determined directly through the prediction results and failure thresholds corresponding to each target circuit. Then, the ratio of the target number to the total number of target circuits in the target circuit set is calculated to obtain the yield of the target circuits in the target circuit set. The calculation process is simple, which improves the efficiency of circuit yield analysis.
[0048] In this embodiment, the target radial basis regression model is obtained by training the first sample circuit set obtained by uniform sampling, and the target multilayer perceptron neural network is obtained by training the second sample circuit set obtained by normal distribution sampling. Then, the process parameters of the MOS tube in the target circuit are predicted directly by the complete model constructed by the target radial basis regression model and the target multilayer perceptron neural network to obtain the prediction result, and then whether the target circuit fails is determined based on the prediction result and the preset failure threshold, so that the yield of the circuit in the target circuit data set can be determined. This application shortens the yield analysis time, improves the analysis efficiency, and reduces the cost by replacing SPICE simulation.
[0049] In addition, in the embodiment of the present application, the intermediate prediction results of each target circuit are first obtained according to the target radial basis regression model, and then the intermediate prediction results are corrected by the target multilayer perceptron neural network to obtain the final prediction results of each target circuit. On the one hand, since the training samples for training the target radial basis regression model are obtained based on uniform sampling of the process parameter variables, it can be ensured that the training samples can uniformly cover the entire fluctuation space of the process parameter variables, so that in the inference stage, whether for qualified circuits whose actual circuit responses meet the requirements or critical circuits whose actual circuit responses are near the failure threshold, the target radial basis regression model can output relatively accurate prediction values; on the other hand, considering that the circuit data of the target circuit to be analyzed in the actual inference stage usually conforms to the normal distribution sampling law, if the prediction results are directly obtained by the target radial basis regression model, there will be a problem of inconsistency between the distribution of the model training samples and the distribution of the variables to be predicted in the model inference stage, thereby limiting the accuracy of the prediction results. Based on the above reasons, in the embodiment of the present application, a target multilayer perceptron neural network obtained by normal distribution sampling training is further introduced, so that the prediction results output by the target radial basis regression model are further corrected by the network to correct the prediction result errors caused by different data distributions. In summary, the embodiment of the present application can effectively improve the accuracy of the circuit yield analysis results.
[0050] Reference Figure 2, shows a flow chart of the steps of a circuit yield analysis method according to another exemplary embodiment of the present application. As shown in the figure, this embodiment mainly shows a specific implementation scheme of the pre-training process of the target radial basis regression model of the above embodiment. This embodiment mainly includes the following steps:
[0051] S202: Calculate and obtain a first target parameter in a target linear function according to the first sample circuit set and a pre-constructed minimization loss function.
[0052] The first sample circuit set includes sample circuit data of each first sample circuit and corresponding first sample circuit simulation results, and the sample circuit data of each first sample circuit includes first sample process parameters of the MOS tube in the first sample circuit. The minimization loss function includes a preset linear function in a preset radial basis regression model and an L1 regularization term corresponding to the parameters in the preset linear function.
[0053] Exemplarily, the preset linear function refers to a function with undetermined parameters. , the constructed minimization loss function can be as follows:
[0054]
[0055] in, is the number of training samples of the target radial basis regression model, This is a first sample circuit simulation result, such as a read access delay simulation result of an SRAM. for A collection of parameters, including The coefficients in and constants , is the L1 regularization term, is a preset fixed value, which can be flexibly set by those skilled in the art according to actual conditions, and this embodiment does not limit this. When , there is no regularization term, therefore, in this embodiment Not 0.
[0056] By solving the minimization loss function through the first sample circuit set, we get The solution is the first target parameter, including the coefficient and constants The solution.
[0057] S204: Determine a second target coefficient in a target radial basis function based on the first target parameter and the first sample circuit set.
[0058] For example, according to The formula and The formula for the second objective coefficient is as follows:
[0059]
[0060] ;
[0061] in, is a first sample circuit simulation result corresponding to each first sample circuit, is the output result of the target linear function corresponding to each first sample circuit, is the circuit data of each first sample circuit input The corresponding weight coefficient is, is the number of training samples of the target radial basis regression model, that is, the number of first sample circuits.
[0062] In an optional implementation, based on the first target parameter and the first sample circuit set, a second target coefficient of the target radial basis function is determined, including: based on the first target parameter, filtering the first sample process parameter to obtain a filtering parameter, the filtering parameter refers to the parameter remaining after filtering out the first sample process parameter whose coefficient in the first target parameter is 0 from the first sample process parameter; inputting the filtering parameter, the output result of the target linear function corresponding to the filtering parameter, and the simulation result of the first sample circuit into a preset radial basis function in a preset radial basis regression model, and calculating the second target coefficient of the target radial basis function.
[0063] For example, when solving the minimization loss function, since the L1 regularization term corresponding to the parameter in the preset linear function is added, the coefficient in the first target parameter finally obtained is There is a case where the coefficient is 0, and the first sample process parameter corresponding to the coefficient of 0 can be understood as a process parameter that has little impact on circuit performance. Therefore, when performing the second target coefficient When solving the problem, you can first input the first sample process parameter variable The first sample process parameter with a coefficient of 0 in the first target parameter is filtered out to obtain the filter parameter, that is, the number of dimensions of the filter parameter is less than the first sample process parameter variable of the original input The number of dimensions is calculated, and then the filtering parameters are input into the solution formula of the second target coefficient for solution.
[0064] In this implementation, by calculating the first target parameter of the L1 regularization term corresponding to the parameter in the preset linear function, the dimension reduction processing of the first sample process parameter can be achieved, so that when calculating the second target coefficient of the target radial basis function, the amount of calculated data is reduced, and the calculation efficiency is improved. In addition, in the process of training the target radial basis regression model, the number of training samples needs to be much larger than the dimension of the input variable. Through this implementation, the number of training samples can be reduced and the training efficiency can be improved.
[0065] In an optional implementation, the filter parameters, the output results of the target linear function corresponding to the filter parameters, and the first sample circuit simulation results are input into a preset radial basis function in a preset radial basis regression model to calculate a second target coefficient of the target radial basis function, including: inputting the filter parameters, the output results of the target linear function corresponding to the filter parameters, and the first sample circuit simulation results into the preset radial basis function in the preset radial basis regression model to obtain a target calculation formula for the second target coefficient of the target radial basis function; vectorizing the calculation items corresponding to the filter parameters in the target calculation formula to obtain vectorized calculation items, and calculating the vectorized calculation items through a dedicated matrix calculation library to obtain calculation results of the calculation items corresponding to the filter parameters; solving the target calculation formula based on the calculation results to obtain the second target coefficient of the target radial basis function.
[0066] Exemplarily, for the solution process of the second target coefficient, the solution formula of the second target coefficient can be transformed to obtain the target calculation formula for the second target coefficient of the target radial basis function, as follows:
[0067]
[0068]
[0069]
[0070] Furthermore, the calculation items corresponding to the filter parameters can be Perform vectorization to obtain vectorized calculation items:
[0071]
[0072] The vectorized calculation items are calculated by a dedicated matrix calculation library to obtain the calculation results of the calculation items corresponding to the filter parameters; the target calculation formula is solved based on the calculation results to obtain the second target coefficient of the target radial basis function.
[0073] In this implementation, by vectorizing the calculation items corresponding to the filter parameters in the target calculation formula to obtain vectorized calculation items, and calculating the vectorized calculation items through a dedicated matrix calculation library to obtain the calculation results of the calculation items corresponding to the filter parameters, the calculation efficiency can be improved, thereby improving the model training efficiency.
[0074] S206: Determine a target radial basis regression model according to the first target parameter in the target linear function and the second target coefficient in the target radial basis function.
[0075] Exemplarily, a target linear function is determined according to the calculated first target parameter, a target radial basis function is determined according to the calculated second target coefficient, and a target radial basis regression model is constructed according to the target linear function and the target radial basis function.
[0076] In this embodiment, by calculating the first target parameter of the L1 regularization term corresponding to the parameter in the preset linear function, it is possible to reduce the amount of calculated data when subsequently calculating the second target coefficient of the target radial basis function, improve calculation efficiency, and reduce training samples.
[0077] In an optional implementation, the second sample circuit set includes circuit data of each second sample circuit and a corresponding second sample circuit simulation result, the sample circuit data of each second sample circuit includes second sample process parameters of the MOS tube in the second sample circuit, and the pre-training process of the target multilayer perceptron neural network includes: inputting the second sample circuit set into a target radial basis regression model for prediction to obtain sample prediction values corresponding to each second sample circuit; using the sample prediction values as training samples and the second sample circuit simulation results as training labels to train the preset multilayer perceptron neural network to obtain the target multilayer perceptron neural network.
[0078] Exemplarily, the preset multilayer perceptron neural network is trained using the second sample circuit set obtained by normal distribution sampling. Since in actual application scenarios, the target multilayer perceptron neural network input is the intermediate prediction result obtained by the target radial basis regression model prediction, therefore, when performing the preset multilayer perceptron neural network training, it is necessary to first input the second sample circuit set into the target radial basis regression model for prediction to obtain the sample prediction value corresponding to each second sample circuit, and then the sample prediction value corresponding to each second sample circuit and the corresponding second sample circuit simulation result are combined into a training data set to train the preset multilayer perceptron neural network to obtain the target multilayer perceptron neural network.
[0079] In this implementation, the second sample circuit set is first input into the target radial basis regression model for prediction to obtain sample prediction values corresponding to each second sample circuit, and then the sample prediction values are used as training samples, and the second sample circuit simulation results are used as training labels to train the preset multilayer perceptron neural network to obtain the target multilayer perceptron neural network, thereby ensuring the prediction accuracy of the target multilayer perceptron neural network.
[0080] Reference Figure 3 , shows a structural block diagram of a circuit yield analysis device according to an exemplary embodiment of the present application.
[0081] The circuit yield analysis device of this embodiment includes an acquisition module 302 , a first prediction module 304 , a second prediction module 306 and an analysis module 308 .
[0082] Among them, the acquisition module 302 is used to obtain the circuit data of each target circuit included in the target circuit set, wherein the circuit data of each target circuit includes the process parameters of the MOS tube in the target circuit; the first prediction module 304 is used to input the circuit data of each target circuit into a pre-trained target radial basis regression model, and output the intermediate prediction results corresponding to each target circuit, and the target radial basis regression model is trained by a first sample circuit set obtained by uniform sampling; the second prediction module 306 is used to input each intermediate prediction result into a pre-trained target multilayer perceptron neural network, and output the prediction results corresponding to each target circuit, and the target multilayer perceptron neural network is trained by a second sample circuit set obtained by normal distribution sampling; the analysis module 308 is used to determine the yield of the target circuit in the target circuit set according to the prediction results and failure thresholds corresponding to each target circuit.
[0083] In an optional implementation, the target radial basis regression model includes a target linear function and a target radial basis function, and the first prediction module 304 is also used to: input the circuit data of each target circuit into the target linear function for linear fitting, and output an initial prediction result; input the initial prediction result into the target radial basis function for result correction, and output an intermediate prediction result corresponding to each target circuit.
[0084] In an optional implementation, the first sample circuit set includes sample circuit data of each first sample circuit and a corresponding first sample circuit simulation result, and the sample circuit data of each first sample circuit includes a first sample process parameter of a MOS tube in the first sample circuit. The circuit yield analysis device of this embodiment also includes a first pre-training module, which is used to: calculate the first target parameter in the target linear function according to the first sample circuit set and a pre-constructed minimization loss function, wherein the minimization loss function includes a preset linear function in a preset radial basis regression model and an L1 regularization term corresponding to the parameter in the preset linear function; determine the second target coefficient in the target radial basis function based on the first target parameter and the first sample circuit set; determine the target radial basis regression model according to the first target parameter in the target linear function and the second target coefficient in the target radial basis function.
[0085] In an optional implementation, the first pre-training module is also used to: filter the first sample process parameters based on the first target parameters to obtain filtering parameters, where the filtering parameters refer to the parameters remaining after filtering out the first sample process parameters whose coefficients are 0 in the first target parameters; input the filtering parameters, the output results of the target linear function corresponding to the filtering parameters, and the first sample circuit simulation results into a preset radial basis function in a preset radial basis regression model to calculate the second target coefficient of the target radial basis function.
[0086] In an optional implementation, the first pre-training module is also used to: input the filtering parameters, the output results of the target linear function corresponding to the filtering parameters, and the first sample circuit simulation results into a preset radial basis function in a preset radial basis regression model to obtain a target calculation formula for a second target coefficient of the target radial basis function; vectorize the calculation items corresponding to the filtering parameters in the target calculation formula to obtain vectorized calculation items, and calculate the vectorized calculation items through a dedicated matrix calculation library to obtain calculation results of the calculation items corresponding to the filtering parameters; solve the target calculation formula based on the calculation results to obtain the second target coefficient of the target radial basis function.
[0087] In an optional implementation, the second sample circuit set includes circuit data of each second sample circuit and a corresponding second sample circuit simulation result, and the sample circuit data of each second sample circuit includes a second sample process parameter of the MOS tube in the second sample circuit. The circuit yield analysis device of this embodiment also includes a second pre-training module, which is used to: input the second sample circuit set into a target radial basis regression model for prediction to obtain sample prediction values corresponding to each second sample circuit; use the sample prediction values as training samples and the second sample circuit simulation results as training labels to train a preset multilayer perceptron neural network to obtain a target multilayer perceptron neural network.
[0088] In an optional implementation, the analysis module 308 is further used to: determine a target number of target circuits whose prediction results are less than the failure threshold based on the prediction results and failure thresholds corresponding to each target circuit; and determine the yield of the target circuits in the target circuit set based on the target number and the total number of target circuits in the target circuit set.
[0089] The circuit yield analysis device of this embodiment is used to implement the corresponding circuit yield analysis methods in the aforementioned multiple method embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be described in detail here. In addition, the functional implementation of each module in the circuit yield analysis device of this embodiment can refer to the description of the corresponding parts in the aforementioned method embodiments, which will not be described in detail here.
[0090] Reference Figure 4 , shows a schematic diagram of the structure of an electronic device according to an exemplary embodiment of the present application. The specific embodiment of the present invention does not limit the specific implementation of the electronic device.
[0091] like Figure 4 As shown, the electronic device may include: a processor (processor) 401 , a memory (memory) 403 , a communication bus 404 , and a communication interface (Communications Interface) 405 .
[0092] Wherein: the processor 401 , the memory 403 and the communication interface 405 communicate with each other via the communication bus 404 .
[0093] The communication interface 405 is used to communicate with other electronic devices or servers.
[0094] The processor 401 is used to execute the program 402, and specifically can execute the steps of any circuit yield analysis method in the above embodiments.
[0095] Specifically, the program 402 may include program codes, which include computer operation instructions.
[0096] The processor 401 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.
[0097] The memory 403 is used to store the program 402. The memory 403 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
[0098] Program 402 can be specifically used to enable processor 401 to execute to implement the steps of any circuit yield analysis method described in the embodiment. The specific implementation of each step in program 402 can refer to the corresponding description of the steps and units executed by any circuit yield analysis method in the above steps, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above-mentioned method embodiment.
[0099] An embodiment of the present application further provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements a circuit yield analysis method as described in any one of the above-mentioned method embodiments.
[0100] An embodiment of the present application also provides a computer program product, including computer instructions, which instruct a computing device to execute operations corresponding to the circuit yield analysis method described in any one of the above-mentioned multiple method embodiments.
[0101] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present application.
[0102] The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or implemented as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or implemented as a computer code originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and to be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a special-purpose processor or programmable or special-purpose hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown here, the execution of the code converts the general-purpose computer into a special-purpose computer for executing the method shown here.
[0103] Those of ordinary skill in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present application.
[0104] The above implementation methods are only used to illustrate the embodiments of the present application, and are not limitations on the embodiments of the present application. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the embodiments of the present application. Therefore, all equivalent technical solutions also belong to the scope of the embodiments of the present application. The scope of patent protection of the embodiments of the present application should be limited by the claims.
Claims
1. A circuit yield analysis method, It is characterized in that include: Acquire circuit data of each target circuit included in the target circuit set, wherein the circuit data of each target circuit includes process parameters of the MOS tube in the target circuit; Inputting the circuit data of each target circuit into the target linear function in the pre-trained target radial basis regression model for linear fitting, and outputting an initial prediction result; inputting the initial prediction result into the target radial basis function in the target radial basis regression model for result correction, and outputting an intermediate prediction result corresponding to each target circuit, wherein the target radial basis regression model is trained by a first sample circuit set obtained by uniform sampling; the first sample circuit set includes sample circuit data of each first sample circuit and corresponding first sample circuit simulation results, and the sample circuit data of each first sample circuit includes a first sample process parameter of a MOS tube in the first sample circuit; Inputting each of the intermediate prediction results into a pre-trained target multi-layer perceptron neural network, and outputting a prediction result corresponding to each of the target circuits, wherein the target multi-layer perceptron neural network is trained by a second sample circuit set obtained by sampling a normal distribution; Determining the yield of the target circuits in the target circuit set according to the prediction results and failure thresholds corresponding to the target circuits; The calculation formula of the target radial basis regression model is: in, is the target linear function, is the target radial basis function, is the input variable The dimension of yes A single variable The coefficient of is a constant, yes The corresponding weight coefficient is, is the number of training samples of the target radial basis regression model, is a preset fixed value. is the total number of target circuits included in the target circuit set; The pre-training process of the target radial basis regression model includes: calculating a first target parameter in the target linear function according to the first sample circuit set and a pre-constructed minimization loss function; determining a second target coefficient in the target radial basis function based on the first target parameter and the first sample circuit set; determining the target radial basis regression model according to the first target parameter in the target linear function and the second target coefficient in the target radial basis function; The minimization loss function is: in, is the simulation result of the first sample circuit, is the first target parameter, and the first target parameter includes The coefficients in and the constant , is the L1 regularization term, It is a preset fixed value.
2. The circuit yield analysis method according to claim 1, It is characterized in that The step of determining a second target coefficient of the target radial basis function based on the first target parameter and the first sample circuit set comprises: Based on the first target parameter, filtering the first sample process parameter to obtain a filtering parameter, wherein the filtering parameter refers to a parameter remaining after filtering out the first sample process parameter whose coefficient is 0 in the first target parameter from the first sample process parameter; The filter parameters, the output results of the target linear function corresponding to the filter parameters and the first sample circuit simulation results are input into a preset radial basis function in the target radial basis regression model to calculate a second target coefficient of the target radial basis function.
3. The circuit yield analysis method according to claim 2, It is characterized in that The step of inputting the filter parameter, the output result of the target linear function corresponding to the filter parameter, and the first sample circuit simulation result into a preset radial basis function in the target radial basis regression model to calculate a second target coefficient of the target radial basis function includes: Inputting the filtering parameter, the output result of the target linear function corresponding to the filtering parameter, and the first sample circuit simulation result into a preset radial basis function in the target radial basis regression model to obtain a target calculation formula for a second target coefficient of the target radial basis function; Vectorizing the calculation item corresponding to the filter parameter in the target calculation formula to obtain the vectorized calculation item, and calculating the vectorized calculation item through a dedicated matrix calculation library to obtain the calculation result of the calculation item corresponding to the filter parameter; The target calculation formula is solved based on the calculation result to obtain the second target coefficient of the target radial basis function.
4. The circuit yield analysis method according to claim 1, It is characterized in that The second sample circuit set includes circuit data of each second sample circuit and corresponding second sample circuit simulation results, the sample circuit data of each second sample circuit includes second sample process parameters of the MOS tube in the second sample circuit, and the pre-training process of the target multilayer perceptron neural network includes: Inputting the second sample circuit set into the target radial basis regression model for prediction to obtain a sample prediction value corresponding to each of the second sample circuits; The sample prediction value is used as a training sample, and the second sample circuit simulation result is used as a training label to train a preset multilayer perceptron neural network to obtain the target multilayer perceptron neural network.
5. The circuit yield analysis method according to claim 1, It is characterized in that The step of determining the yield of the target circuit in the target circuit data set according to the prediction result and the failure threshold corresponding to each target circuit includes: Determine a target number of target circuits whose prediction results are less than the failure threshold according to the prediction results and failure thresholds corresponding to the target circuits; The yield of the target circuits in the target circuit set is determined according to the target number and the total number of target circuits in the target circuit set.
6. A circuit yield analysis device, It is characterized in that include: An acquisition module, used to acquire circuit data of each target circuit included in the target circuit set, wherein the circuit data of each target circuit includes process parameters of the MOS tube in the target circuit; A first prediction module is used to input the circuit data of each target circuit into a target linear function in a pre-trained target radial basis regression model for linear fitting, and output an initial prediction result; input the initial prediction result into the target radial basis function in the target radial basis regression model for result correction, and output an intermediate prediction result corresponding to each target circuit, wherein the target radial basis regression model is trained by a first sample circuit set obtained by uniform sampling; the first sample circuit set includes sample circuit data of each first sample circuit and a corresponding first sample circuit simulation result, and the sample circuit data of each first sample circuit includes a first sample process parameter of a MOS tube in the first sample circuit; The calculation formula of the target radial basis regression model is: in, is the target linear function, is the target radial basis function, is the input variable The dimension of yes A single variable The coefficient of is a constant, yes The corresponding weight coefficient is, is the number of training samples of the target radial basis regression model, is a preset fixed value. is the total number of target circuits included in the target circuit set; The pre-training process of the target radial basis regression model includes: calculating a first target parameter in the target linear function according to the first sample circuit set and a pre-constructed minimization loss function; determining a second target coefficient in the target radial basis function based on the first target parameter and the first sample circuit set; determining the target radial basis regression model according to the first target parameter in the target linear function and the second target coefficient in the target radial basis function; The minimization loss function is: in, is the simulation result of the first sample circuit, is the first target parameter, and the first target parameter includes The coefficients in and the constant , is the L1 regularization term, is a preset fixed value; A second prediction module is used to input each of the intermediate prediction results into a pre-trained target multi-layer perceptron neural network, and output a prediction result corresponding to each of the target circuits, wherein the target multi-layer perceptron neural network is trained by a second sample circuit set obtained by sampling a normal distribution; The analysis module is used to determine the yield of the target circuits in the target circuit set according to the prediction results and failure thresholds corresponding to each target circuit.
7. An electronic device, It is characterized in that include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method according to any one of claims 1 to 5.
8. A computer storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.
