A confidence analysis method for semiconductor device simulation results based on sample data

Abstract

The present invention is a confidence analysis method for semiconductor device simulation results based on sample data, comprising the following steps: Step 1: according to the source of actual sample space data, respectively set the confidence weight of the actual sample space data; Step 2: according to the device structure parameters to be simulated, set the neighborhood radius; Step 3: according to the similarity function, substitute the structural parameters of the space points within the neighborhood radius and the structural parameters of the simulation sample into the similarity function, and calculate the similarity between the structural parameters of the space points within the neighborhood radius and the structural parameters of the simulation sample; Step 4: according to the similarity calculation result, calculate and obtain the confidence of the neighborhood sample space relative to the simulation sample. The present invention introduces the confidence of sample data into the confidence formula, effectively improves the accuracy of confidence evaluation, and by controlling the value of the neighborhood, not only a small range of neighborhood space can be considered, but also the overall data interval can be considered, which can effectively guide the design process.

Description

Technical Field

[0001] The present invention belongs to a result analysis method in the field of artificial intelligence, and specifically relates to a method for performing confidence analysis on simulation results of semiconductor devices using known sample data. Background Art

[0002] For designers, effective design result evaluation methods have important practical significance. In recent years, with the explosion of data and the improvement of computer hardware equipment, modeling technology with machine learning algorithms as the core has continued to develop, and models that use machine learning algorithms to construct nonlinear mapping relationships between input and output have been widely studied and applied. The construction of machine learning models depends largely on data. The quality and quantity of data are key factors in determining the effectiveness of the model. Similarly, it also has an important impact on the confidence of the prediction results.

[0003] At present, both in academia and industry, the evaluation of simulation results based on machine learning algorithms mainly comes from the model construction process, and only considers the correlation between data. Taking the Gaussian process regression algorithm for confidence analysis as an example, it uses the entire data set as reference information and only considers the correlation between data. However, data from different sources also have confidence, and there are no reports on evaluation schemes that consider the confidence of data sources. Therefore, there is an urgent need for a more accurate and comprehensive solution to achieve a more effective confidence analysis method. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a semiconductor device simulation result confidence analysis method with low cost, low environmental requirements and high reliability. The method can consider not only a small range of neighborhood space, but also the overall data interval.

[0005] In order to achieve the above object, the present invention is achieved through the following technical solutions:

[0006] The present invention is a semiconductor device simulation result confidence analysis method based on sample data, which comprises the following steps:

[0007] Step 1: According to the actual sample space data source, set its confidence weight C(X i );

[0008] Step 2: Based on the simulation structure parameter X0, set the neighborhood radius R;

[0009] Step 3: Calculate and obtain the spatial point X in the neighborhood i The similarity g(X i ,X0);

[0010] Step 4: Based on the above similarity function, use the formula Calculate the confidence of the neighborhood sample space relative to the simulated sample.

[0011] Furthermore, in step 1, the confidence of the sample space data is determined according to the data source, which includes but is not limited to experimental data, simulation data or literature research data, etc.

[0012] Furthermore, in step 2, when R is set smaller than the sample space boundary, it includes the sample space within the neighborhood of R. When R is set larger than the sample space boundary, it includes the entire sample space.

[0013] Furthermore, the function for measuring the similarity between the spatial points in the neighborhood and the simulated samples in step 3 may be a Gaussian kernel function, Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, angle cosine, Hamming distance, Jaccard similarity coefficient, Jaccard distance, correlation coefficient, correlation distance, Lang's distance, oblique space distance, exponential similarity coefficient, non-parametric similarity (maximum and minimum similarity), information entropy, and other similarity calculation methods that are not limited to such methods.

[0014] Furthermore, in step 4, the confidence of the simulated sample is determined using formula (1):

[0015]

[0016] Among them, C(X i ) is the confidence of the neighborhood sample space data, i is the different sample data in the neighborhood, and N is the total number of sample data in the neighborhood.

[0017] The beneficial effects of the present invention are:

[0018] The present invention proposes for the first time an analysis method for the confidence of semiconductor device simulation results taking into account sample data. It not only considers the correlation between the simulation samples and the sample database, but also introduces the confidence of the sample data into the confidence formula, thereby effectively improving the accuracy of the confidence evaluation.

[0019] The present invention proposes for the first time a confidence analysis method that takes neighborhood into consideration. For the simulation results of a machine learning simulation model constructed based on data information, the present invention can consider not only a small-scale neighborhood space but also the overall data interval by controlling the value of the neighborhood R.

[0020] The present invention is applicable to any machine learning model built based on data information, including regression models and classification models. The analysis scheme can provide guidance for designers' design results and effectively guide the design process.

[0021] The present invention uses a confidence analysis method based on known sample data to simply and directly give the reliability of the simulation results based on the machine learning model given the simulation device structure. This method supplements and expands the traditional machine learning simulation solution and provides better guidance for designers. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of principle analysis of the present invention.

[0023] Figure 2 It is a flow chart of confidence analysis of simulation results according to an embodiment of the present invention.

[0024] Figure 3 It is a schematic diagram of a slice of a three-dimensional SOI LDMOS structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will disclose the embodiments of the present invention with drawings. For the purpose of clear description, many practical details will be described together in the following description. However, it should be understood that these practical details should not be used to limit the present invention. That is, in some embodiments of the present invention, these practical details are not necessary. In addition, for the purpose of simplifying the drawings, some conventional structures and components will be depicted in a simple schematic manner in the drawings.

[0026] The present invention is a semiconductor device simulation result confidence analysis method based on sample data. The analysis method is applicable to a machine learning regression simulation model and a machine learning classification simulation model that rely on data for training. The method comprises the following steps:

[0027] Step 1. First, set different confidence weights according to the data source of the sample database. For example, the confidence weight of data source 1 is C1=0.9, and the confidence weight of data source 2 is C2=0.95.

[0028] Step 2. According to the structural parameters of the device to be simulated, take X0 as an example and set the neighborhood value to R.

[0029] Step 3. Get the sample points in the neighborhood space R. Assume that there are 4 sample points, 3 of which are from data source 1, namely X1, X2, and X3, and 1 is from data source 2, namely X4. Define g(X i ,X0) is any sample structure X in the space i The similarity kernel function between the simulated sample structure X0 is calculated as g(X1,X0), g(X2,X0), g(X3,X0), and g(X4,X0). When the Gaussian kernel function is used, When the cosine angle function is used, When using Euclidean distance,

[0030] Step 4. According to the formula Substituting the sample points in step 3 into it, we can get the confidence level for the simulated structure X0:

[0031]

[0032] The following is a device sample data set and data structure data to be simulated, combined with the accompanying drawings. The calculation of the confidence level of the simulation result of the silicon-on-insulator (SOI) lateral double-diffused metal oxide semiconductor field effect transistor (LDMOS) device structure on an insulating substrate is used as an example for analysis.

[0033] SOI LDMOS is a type of integrated power device. The three-dimensional ring structure is its basic structure in practical applications. Figure 3 A SOI LDMOS slice structure is given, where 1 is a heavily doped region with a first conductivity type, 2 is a heavily doped region with a second conductivity type, 3 is a channel region with a first conductivity type, 4 is a top silicon drift region with a second conductivity type, 5 is a heavily doped region with a second conductivity type, 6 is an insulating buried layer on a supporting substrate, and 7 is a supporting substrate.

[0034] A known simulated SOI LDMOS structure is used as an example to perform confidence analysis using the method proposed in the present invention.

[0035] (1) Figure 3 The schematic diagram of the structure of two-dimensional SOI LDMOS is given, and the training sample data in the neighborhood space is obtained through device simulation, experiment or literature research. Taking the simulation structure X0 = [1e15, 25, 3, 1] as an example, its neighborhood value R is set to R = [±2e14, ±5, ±1, ±0.5]. By calculating the sample points in the simulation structure and space, 4 points can be obtained, of which 3 points are data source 1, namely X1 = [1.2e15, 22, 3.5, 0.8], X2 = [8e14, 23, 3, 1.5], and X3 = [9e14, 18, 2.5, 1], and 1 point is data source 2, X4 = [1e15, 23, 2, 1.4].

[0036] (2) By substituting the above structural points into the similarity kernel function, when the Gaussian kernel function is used, g(X1,X0)=0.9, g(X2,X0)=0.92, g(X3,X0)=0.82, g(X4,X0)=0.74; when the cosine angle function is used, g(X1,X0)=0.99, g(X2,X0)=0.99, g(X3,X0)=0.99, g(X4,X0)=0.97; when the Euclidean distance function is used, g(X1,X0)=0.61, g(X2,X0)=0.57, g(X3,X0)=0.86, g(X4,X0)=1.09.

[0037] (3) According to the known data sources, the confidence weights of the data are determined as C1 = 0.9 and C2 = 0.95. Substituting the above results into formula (3), the confidence of the final prediction result can be obtained. The Gaussian kernel function Q(X0) = 0.9 is used; the cosine angle function Q(X0) = 0.91 is used; and the Euclidean distance is Q(X0) = 0.91.

[0038] The present invention proposes a confidence analysis method for a simulation model constructed by a data-based machine learning algorithm. By utilizing the correlation and confidence weight between data in the neighborhood space and the simulation structure, compared with the traditional intelligent simulation model that can only provide the performance of the simulation structure, the confidence of the simulation result can be provided while providing the performance of the simulation structure, thereby providing more reference information for designers and facilitating the realization of higher-precision structural design.

[0039] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A semiconductor device simulation result confidence analysis method based on sample data, characterized in that: The semiconductor device simulation result confidence analysis method comprises the following steps: Step 1: According to the actual sample space data source, set the confidence weight of the actual sample space data respectively, where X i Represents the structural parameters of the device; Step 2: Set the neighborhood radius R according to the device structure parameter X0 to be simulated; Step 3: According to the similarity function, the actual sample space structure parameter X of the space point within the neighborhood radius R in step 2 is i Substitute the simulated sample structure parameter X0 into the similarity function to calculate the structure parameter X of the spatial point within the neighborhood radius R i The similarity g(X i ,X0); Step 4: Based on the similarity calculation results in step 3, calculate the confidence of the neighborhood sample space relative to the simulation sample.

2. The method for analyzing the confidence of semiconductor device simulation results based on sample data according to claim 1, characterized in that: In step 4, the calculation formula of the confidence of the neighborhood sample space relative to the simulation sample is: Among them, C(X i ) is the confidence of the neighborhood sample space data, i is the different sample data in the neighborhood, and N is the total number of sample data in the neighborhood.

3. The method for analyzing the confidence of semiconductor device simulation results based on sample data according to claim 1, characterized in that: The function for measuring the similarity between the spatial point in the neighborhood and the simulation sample in the step 3 is one of Gaussian kernel function, Euclidean distance, angle cosine, Manhattan distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, Hamming distance, Jaccard similarity coefficient, Jaccard distance, correlation coefficient, correlation distance, Lang's distance, oblique space distance, exponential similarity coefficient, non-parametric similarity, and information entropy.

4. The method for analyzing the confidence of semiconductor device simulation results based on sample data according to claim 3, characterized in that: The similarity function in step 3 is a Gaussian kernel function Where x, y are structural parameters, and σ is the inverse of the x dimension.

5. The method for analyzing the confidence of semiconductor device simulation results based on sample data according to claim 1, characterized in that: In step 2, when the neighborhood radius R is set smaller than the sample space boundary, it includes the sample space within the R neighborhood, and when R is set larger than the sample space boundary, it includes the entire sample space.

6. The method for analyzing the confidence of semiconductor device simulation results based on sample data according to claim 1, characterized in that: The confidence weight of the sample space data in step 1 is determined according to a data source, which includes but is not limited to experimental data, simulation data or literature research data.

7. A semiconductor device simulation result confidence analysis method based on sample data according to any one of claims 1 to 6, characterized in that: This analysis method is applicable to machine learning regression simulation models and machine learning classification simulation models that rely on data for training.

Images