A power device breakdown voltage calculation method based on electric field dimension reduction

Abstract

The present application belongs to the field of power semiconductor device design, and specifically provides a power device breakdown voltage calculation method based on electric field dimension reduction, to solve the technical problems of waste of computing resources and low efficiency of engineering evaluation caused by global high-density grid discretization in the prior art; the present application realizes adaptive dimension reduction extraction of global data points by identifying the geometric curvature and gradient characteristics of the distribution curve of internal physical quantities (such as electric field) of the device; under the premise of ensuring that the mutation nodes (such as electric field peak points) determining the physical mechanism can be accurately controlled through threshold conditions, the redundant calculation load of the gentle area is greatly reduced, thereby improving the evaluation efficiency of the avalanche breakdown characteristics of advanced power devices under the premise of ensuring the evaluation accuracy.

Description

Technical Field

[0001] This invention belongs to the field of power semiconductor device design, specifically providing a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction. Background Technology

[0002] The increasing demand for power semiconductors in new energy vehicles, data centers, and industrial control has placed more stringent requirements on the voltage withstand capability of devices, leading to increasingly complex electric field distributions within these devices. In advanced voltage withstand layer designs such as multiple floating islands (MFI) and super junctions (SJ), the mainstream design philosophy for achieving higher breakdown voltages (BV) is to make the electric field intensity more uniform across space, giving the electric field partially smooth and linear characteristics.

[0003] In predicting device breakdown voltage, by means of Numerical calculation methods for integrals are fundamental in the industry. These integrals consist of nested path integrals and contain complex functions such as exponents. Currently, in the evaluation of avalanche breakdown characteristics of the aforementioned complex devices, whether using analytical models or numerical simulations, a globally high-density uniform grid is commonly used for spatial discretization. However, this approach has certain limitations in engineering applications. First, computational redundancy is significant. With uniform grid partitioning (typically containing thousands to tens of thousands of nodes), a large number of nodes are distributed in regions with gentle electric field changes, contributing little to the collision ionization integral results while still consuming considerable computational and storage resources. Second, it is difficult to balance computational efficiency and accuracy. When the overall grid density is reduced, local electric field peaks caused by multi-layered floating island structures are easily overlooked, leading to significant avalanche integral errors. While maintaining a high-density grid helps improve accuracy, it significantly increases the time required for analytical calculations or numerical iterations, affecting the efficiency of evaluation in engineering research and development. Summary of the Invention

[0004] The purpose of this invention is to provide a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction, in order to solve the technical problems of wasted computational resources and low engineering evaluation efficiency caused by global high-density grid discretization in the prior art. This invention achieves adaptive dimensionality reduction extraction of global data points by identifying the geometric curvature and gradient characteristics of the distribution curve of physical quantities (such as electric field) inside the device. Under the premise of ensuring that the abrupt nodes that determine the physical mechanism (such as electric field peak points) can be accurately controlled through threshold conditions, the redundant computational load in the smooth region is greatly reduced, thereby improving the evaluation efficiency of avalanche breakdown characteristics of advanced power devices while ensuring evaluation accuracy.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for calculating the breakdown voltage of power devices based on electric field dimension reduction, characterized by the following steps:

[0007] Step 1: Establish a physical model of the target semiconductor device and solve for the reference dense electric field distribution under zero bias conditions;

[0008] Step 2: Set the absolute error threshold, and adaptively extract the feature nodes in the reference total electric field distribution based on the absolute error threshold to obtain the reference sparse electric field distribution;

[0009] Step 3: Set the initial search domain of the test voltage according to the reference dense electric field distribution, and complete the test voltage initialization;

[0010] Step 4: Under the current test voltage, update the sparse electric field distribution based on linear offset, and reconstruct it into a dense electric field distribution through linear interpolation;

[0011] Step 5: Based on the reconstructed dense electric field distribution, the avalanche collision ionization model is used to determine whether the target semiconductor device has reached the breakdown condition.

[0012] Step 6: Based on the judgment result, update the test voltage search domain and the current test voltage using the binary search method, and perform iterative updates until the cutoff condition is met. The test voltage at the cutoff point is the static avalanche breakdown voltage of the target semiconductor device.

[0013] Furthermore, in step 1, the input parameters of the physical model of the target semiconductor device include: device physical structure parameters, material doping parameters, and a one-dimensional initial spatial grid sequence.

[0014] With the applied bias boundary condition set to 0, the input parameters are substituted into the physical model to calculate the one-dimensional initial spatial mesh sequence. Transverse electric field components of each dense grid node under zero bias With longitudinal electric field component The reference dense electric field distribution of the target semiconductor device under zero bias condition is calculated. :

[0015] .

[0016] Furthermore, the physical structural parameters of the device include: the total thickness of the drift region. Device half-cell width P-type floating island thickness and the width of the P-type floating island Material doping parameters include: N-type doping concentration in the drift region. With respect to the P-type doping concentration in the floating island region .

[0017] Furthermore, when the target semiconductor device is a silicon carbide (SiC) device, the physical model adopts a two-dimensional analytical physical model based on the Poisson equation and double Fourier series expansion.

[0018] Furthermore, the specific process of step 2 is as follows:

[0019] Set absolute error threshold : , To preset the relative error tolerance ratio, As a reference dense electric field distribution, This represents the maximum background electric field value of the reference dense electric field distribution;

[0020] Based on the aforementioned absolute error threshold, the feature nodes in the reference total electric field distribution are adaptively extracted using the DP algorithm to obtain the feature grid sequence. Corresponding feature grid sequence The baseline sparse electric field distribution is extracted, where the transverse electric field component is: Longitudinal electric field component .

[0021] Furthermore, the specific process of step 3 is as follows:

[0022] From the reference sparse electric field distribution Extracting the maximum background electric field value:

[0023] ,

[0024] Combined with the critical breakdown electric field constant of the target semiconductor material Total thickness of the drift region and preset upper limit redundancy coefficient Calculate the initial estimate of the theoretical pressure resistance limit. :

[0025] ,

[0026] Set the search domain to , As the lower bound for the search, initialized to ; The upper limit of the search is initialized to... The initialization of the trial voltage is then: .

[0027] Furthermore, the specific process of step 4 is as follows:

[0028] Based on the current test voltage The baseline sparse electric field distribution is migrated and updated to obtain the updated transverse electric field components. With longitudinal electric field component :

[0029] ,

[0030] ,

[0031] in, For the characteristic grid sequence, The transverse electric field component of the reference sparse electric field distribution is: The longitudinal electric field component is the reference sparse electric field distribution. The total thickness of the drift region of the target semiconductor device;

[0032] Calculate the total electric field distribution of the test :

[0033] ;

[0034] Then, a linear interpolation method is used to determine the total electric field distribution of the experiment. From sparse feature sequences Mapping to a one-dimensional initial spatial grid sequence Thus, the current test voltage is obtained. Reconstructed electric field distribution .

[0035] Furthermore, the specific process of step 5 is as follows:

[0036] Reconstructing the electric field distribution Substituting into the Chynoweth collisional ionization empirical model, the electron ionization rate of each feature grid is calculated. With hole ionization rate :

[0037] , ,

[0038] in, , , , The collision ionization coefficient is constant. This represents the electric field value corresponding to the feature grid.

[0039] The total electron integral is obtained by performing Townsend avalanche multiplication discrete numerical integration using the trapezoidal integral method. Total integral of holes :

[0040] ,

[0041] ,

[0042] Based on the integral extreme value Determine whether the target semiconductor device has reached the breakdown condition: If If the current test voltage reaches or exceeds the breakdown condition, then it is determined that the current test voltage has reached or exceeded the breakdown condition; if If the current test voltage does not meet the breakdown condition, it is determined that the current test voltage does not meet the breakdown condition.

[0043] Furthermore, the specific process of step 6 is as follows:

[0044] Current test voltage When the breakdown condition is met or exceeded, update the upper limit of the test voltage search domain. : ;

[0045] Current test voltage If the breakdown condition is not met, update the lower limit of the test voltage search domain. : ;

[0046] The voltage search interval is gradually narrowed with each iteration. When the length of the search interval meets the convergence accuracy condition under a preset threshold, the loop ends and the static avalanche breakdown voltage of the target device is output. .

[0047] Based on the above technical solution, the beneficial effects of the present invention are as follows:

[0048] This invention provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction. First, a reference electric field distribution of the target semiconductor device on a high-density spatial grid is obtained through analytical physical models or numerical simulations. Based on this, the Douglas-Peucker algorithm from computer graphics is introduced to adaptively compress the initial electric field distribution data. By setting a global relative error threshold, redundant nodes in regions with gentle electric field changes are automatically ignored based on the deviation between the original data points and the piecewise linear approximation, retaining only feature points with deviations exceeding the threshold. The processed feature nodes effectively characterize regions with significant changes in the internal electric field distribution of the device. This thinning method can characterize the spatial regions with the most drastic changes in internal physical quantities and the most pronounced nonlinear characteristics while reducing redundant nodes in regions with gentle changes. In subsequent solution processes, calculations are performed only on the extracted feature grid, avoiding repeated traversal of the entire high-density grid as in traditional methods. In parameter optimization iteration, the feature nodes are updated by linear offset based on the principle of electric field superposition. Subsequently, an interpolation method is used to reconstruct the feature data to the original spatial grid and substitute it into the collisional ionization integral model (Chynoweth). The model is used to determine the breakdown voltage, thereby achieving efficient calculation of the static avalanche breakdown voltage of the device. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the power device breakdown voltage calculation method based on electric field dimensionality reduction in this invention.

[0050] Figure 2 This is a diagram showing the accuracy consistency of the power device breakdown voltage calculation method based on electric field dimension reduction in Embodiment 1 of the present invention.

[0051] Figure 3 This is a diagram showing the accuracy consistency of the power device breakdown voltage calculation method based on electric field dimension reduction in Embodiment 2 of the present invention.

[0052] Figure 4 This is a diagram showing the accuracy consistency of the power device breakdown voltage calculation method based on electric field dimension reduction in Embodiment 3 of the present invention.

[0053] Figure 5 This is a diagram showing the accuracy consistency of the power device breakdown voltage calculation method based on electric field dimensionality reduction in Embodiment 4 of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0055] Example 1

[0056] This embodiment provides a method for calculating the breakdown voltage of power devices based on electric field dimension reduction, the process of which is as follows: Figure 1 As shown, this method effectively integrates physical prior domain locking, the principle of electric field superposition, and DP adaptive dimensionality reduction, achieving a dramatic leap in computational speed while maintaining high accuracy. Taking the calculation of the static avalanche breakdown voltage (BV) of silicon carbide (SiC) multilayer floating island (MFI) power devices as an example, the specific steps include:

[0057] Step 1: Establish a physical model of the target semiconductor device and solve for the reference dense electric field distribution under zero bias conditions;

[0058] First, a physical model of the target semiconductor device is established, and its high-resolution reference electric field distribution is solved under no external bias condition. The input parameters of the model include: device physical structure parameters, material doping parameters, and a one-dimensional initial spatial grid coordinate sequence. Among them, the device physical structure parameters may include: the total thickness of the drift region. Device half-cell width P-type floating island thickness and the width of the P-type floating island Material doping parameters may include: N-type doping concentration in the drift region. With respect to the P-type doping concentration in the floating island region The one-dimensional initial spatial grid coordinate sequence can be denoted as: The total number of its nodes is denoted as The standard setting is 4000~10000 uniformly discrete nodes;

[0059] For silicon carbide (SiC) devices, a two-dimensional analytical physical model based on the Poisson equation and double Fourier series expansion is introduced, and the applied bias boundary condition is set to 0 before the voltage iteration cycle. After substituting the above input parameters into the analytical physical model, the one-dimensional initial space grid coordinate sequence is calculated. Transverse electric field components of each dense grid node under zero bias With longitudinal electric field component Thus, the reference dense electric field distribution of the device under zero bias voltage is obtained. :

[0060] ,

[0061] Step 2: Set the absolute error threshold, and adaptively extract the feature nodes in the reference total electric field distribution based on the absolute error threshold to obtain the reference sparse electric field distribution;

[0062] To reduce redundant nodes in regions with gentle electric field gradients, the Douglas-Peucker curve simplification algorithm (DP) can be used to simplify the baseline total electric field distribution. Perform adaptive feature extraction;

[0063] Set absolute error threshold Its relative error tolerance ratio compared to the peak value of the reference total electric field and the preset relative error tolerance ratio (Values ​​range from 0 to 1) Related, specifically represented as: , Represents the baseline dense electric field distribution The maximum background electric field value;

[0064] Based on the aforementioned absolute error threshold, the DP algorithm can compress feature nodes in regions of approximately linear or gently changing total electric field distribution, while retaining feature nodes in electric field extrema regions caused by multiple floating island structures, regions with significant curvature changes, or other key change locations. After dimensionality reduction, a feature grid sequence can be output. The number of its nodes is denoted as , usually satisfy ;

[0065] Furthermore, the transverse electric field components of the baseline sparse electric field distribution on the feature grid sequence are extracted and saved. With longitudinal electric field component This data is used as the reference electric field data in subsequent voltage iteration updates to reduce the computational burden of repeatedly calling the complete analytical model.

[0066] Step 3: Based on the reference dense electric field distribution Set the initial search domain for the test voltage and complete the test voltage initialization;

[0067] Combining the critical electric field characteristics of the device material and the zero-bias background electric field, the initial search interval for the bisection method voltage optimization is estimated; firstly, based on the reference sparse electric field distribution... Extracting the maximum background electric field value:

[0068] ,

[0069] Subsequently, the critical breakdown electric field constant of the target semiconductor material was considered. Total thickness of device drift region and the preset physical prior redundancy coefficient An initial estimate of the theoretical pressure resistance limit can be calculated:

[0070] ,

[0071] Therefore, the initial search domain of the bisection voltage optimization loop can be set to That is, setting the initial search lower limit. Initial search limit After entering the binary search loop, the trial voltage for each iteration is initialized to the median value of the current search interval.

[0072] ,

[0073] Step 4: Under the current test voltage, update the sparse electric field distribution based on linear offset, and reconstruct it into a dense electric field distribution through linear interpolation;

[0074] During the iterative calculation of the bisection method for optimization, the electric field distribution can be updated quickly using the superposition relationship of electric fields to reduce the repeated calculation of higher-order analytical formulas.

[0075] Based on the current test voltage It can be found in the feature grid sequence The zero-bias reference electric field is offset and updated; the transverse electric field component can be approximately maintained at the zero-bias reference value, and the longitudinal electric field component can be superimposed with the longitudinal uniform electric field component caused by the applied bias.

[0076] ,

[0077] ,

[0078] in, This represents the longitudinal additional uniform electric field generated by the applied bias voltage;

[0079] Based on the updated electric field components, the trial total electric field distribution on the feature grid sequence is calculated:

[0080] ,

[0081] Subsequently, a linear interpolation method can be used to determine the total electric field distribution of the test. From sparse feature coordinate sequence Mapping to the initial spatial coordinate sequence Thus, the current test voltage is obtained. The high-resolution reconstructed electric field distribution is denoted as . The reconstructed electric field distribution can be used for subsequent collision ionization integral calculations;

[0082] Step 5: Based on the reconstructed dense electric field distribution, the avalanche collision ionization model is used to determine whether the target semiconductor device has reached the breakdown condition.

[0083] In obtaining the reconstructed electric field distribution Then, based on the avalanche collision ionization model, it is determined whether the device has reached the breakdown condition; specifically, it can be... Substituting into the Chynoweth collisional ionization empirical model, the electron ionization rate of each feature grid is calculated. With hole ionization rate :

[0084] , ,

[0085] in, , , , The collision ionization coefficient is constant. This represents the electric field value corresponding to the feature grid.

[0086] In dense grid sequence The trapezoidal integral method is used to perform Townsend avalanche multiplication discrete numerical integration to obtain the total electron integral. Total integral of holes :

[0087] ,

[0088] ,

[0089] Based on the integral extreme value The relationship between the threshold value and the avalanche breakdown determination threshold (typically set to 1) is used to determine whether the target semiconductor device has reached the breakdown condition; if If so, the current test voltage can be considered to have reached or exceeded the breakdown condition; if If so, it can be considered that the current test voltage has not reached the breakdown judgment condition;

[0090] Step 6: Based on the judgment result, update the test voltage search domain and the current test voltage using the binary search method, and perform iterative updates until the cutoff condition is met. The test voltage at the cutoff point is the static avalanche breakdown voltage of the target semiconductor device.

[0091] When the current test voltage reaches or exceeds the breakdown condition, update the upper limit of the test voltage search domain to [value]. ;

[0092] If the current test voltage does not meet the breakdown criteria, update the lower limit of the test voltage search domain to... ;

[0093] By iterating in the above manner, the voltage search interval is gradually reduced. When the length of the search interval meets the convergence accuracy condition under the preset threshold, the loop ends and the static avalanche breakdown voltage of the target device is output. .

[0094] Through the above steps, the benchmark pre-calculation, DP algorithm dimensionality reduction, and linear offset interpolation reconstruction method in this embodiment can significantly simplify the computational load. Taking the static avalanche breakdown voltage (BV) calculation of silicon carbide (SiC) multilayer floating island (MFI) power devices as an example, the device structure parameters, control parameters, and physical model parameters are shown in Table 1, and the accuracy consistency results are as follows. Figure 2 As shown;

[0095] Table 1

[0096]

[0097] Depend on Figure 2 As can be seen, the thick gray line represents the reference electric field waveform calculated by the traditional 4000-node full analysis method; the solid red line is the waveform reconstructed by the hybrid acceleration algorithm of this invention; the red dots represent a small number of feature data points (about 40-100) adaptively extracted by the DP algorithm of this invention. It can be seen that the extracted feature data points almost error-free lock all electric field peaks and extreme points caused by the multi-layer floating islands, the reconstructed waveform is highly consistent with the reference waveform, and the final calculated breakdown voltage result has negligible error compared with the traditional method, but the overall running time is significantly accelerated by tens of times; the full analysis method is used as a comparative example, and the performance comparison between this embodiment and the comparative example is shown in Table 2.

[0098] Table 2

[0099]

[0100] In summary, based on Table 2 and Figure 2As can be seen, the present invention provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction, which can achieve a significant leap in calculation speed while ensuring high accuracy.

[0101] Example 2

[0102] This embodiment provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction. The process is the same as in Embodiment 1, and it also uses the static avalanche breakdown voltage (BV) calculation of a silicon carbide (SiC) multilayer floating island (MFI) power device as an example. Compared to the conventional withstand voltage rating (around 2300V) in Embodiment 1, the total thickness of the drift region of the device in this embodiment is... Increase to (Right now With a withstand voltage rating of 3300V, it can be used in rail transit traction inverters or high-power frequency converters.

[0103] In order to achieve effective electric field distribution control in a thicker drift region, this embodiment increases the number of P-type floating island layers to 3. This multi-layer floating island structure results in a more complex and dense sawtooth multi-peak characteristic in the electric field distribution inside the device. If a traditional analytical algorithm is used, the computational load will increase exponentially due to the extremely high dependence of the ionization integral on the node density.

[0104] Specifically, the device structure parameters, control parameters, and physical model parameters in this embodiment are shown in Table 3 (parameters not listed are consistent with those in Embodiment 1), and the accuracy consistency results are as follows: Figure 3 As shown;

[0105] Table 3

[0106]

[0107] In this embodiment, the calculated zero-bias reference electric field has three distinct main peaks. Through adaptive extraction using the DP algorithm, the original 5000 full grid points are simplified to 108 feature nodes, with data points accounting for only 2.16% of the original total. This demonstrates that the present invention can pinpoint the electric field abrupt change points induced by the floating island structure when dealing with multi-layer floating island structures, while ignoring linearly distributed redundant regions. Furthermore, by locking the search domain, the search space is automatically shrunk to... This avoids a large number of ineffective high-pressure iterations; similarly, the full analysis method is used as a comparative example, and the performance comparison between this embodiment and the comparative example is shown in Table 4.

[0108] Table 4

[0109]

[0110] In summary, based on Table 4 and... Figure 3As can be seen, the present invention provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction, which can achieve a significant leap in calculation speed while ensuring high accuracy.

[0111] Example 3

[0112] This embodiment provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction. The process is the same as in Embodiment 1, and it also uses the calculation of the static avalanche breakdown voltage (BV) of a silicon carbide (SiC) multilayer floating island (MFI) power device as an example. In this embodiment, the total thickness of the drift region of the device is... Increase to ,Right now With a withstand voltage rating of 6500V, it can be applied in ultra-high voltage fields, such as flexible direct current transmission (HVDC) or high-speed rail power networks.

[0113] For SiC devices at the 6500V level, the drift region thickness Need to be significantly increased to ( To achieve charge balance, the number of internal P-type floating islands is increased to 4 or more. In physical space, to eliminate the Gibbs oscillation error caused by the double Fourier series expansion, it is necessary not only to increase the total electric field data points to 8000, but also to increase the number of Fourier high-frequency expansion terms to more than 120 (set to 120 in this embodiment). This setting causes the traditional full analysis method to perform a huge double loop accumulation in each search iteration, and the computational complexity expands exponentially, facing computational difficulties.

[0114] Specifically, the device structure parameters, control parameters, and physical model parameters in this embodiment are shown in Table 5 (unlisted parameters are consistent with those in Embodiment 1), and the accuracy consistency results are as follows: Figure 4 As shown;

[0115] Table 5

[0116]

[0117] In this embodiment, good feature extraction performance and computational efficiency are maintained even under high-density grid and numerical perturbation conditions. For high-frequency numerical noise extraction, the DP algorithm can still identify physical mutation points induced by multi-layer floating islands when facing a reference electric field with numerical jitter composed of 8000 grid points, and adaptively reduce the dimensionality to 155 feature nodes, with a data compression ratio as high as 98.06%. For computational decoupling, this invention strictly limits the time-consuming double Fourier calculation to the initial zero-bias reference extraction stage (only executed once). In the subsequent 15 optimization iterations, only millisecond-level linear mapping is performed based on 155 points, avoiding the defect of traditional methods that repeatedly call higher-order formulas in loops.

[0118] Through adaptive calculation, the initial physical search limit for the device was determined to be 18501.7 V. Using the same full-quantity analysis method as a comparative example, the performance comparison between this embodiment and the comparative example is shown in Table 6.

[0119] Table 6

[0120]

[0121] In summary, based on Table 6 and... Figure 4 As can be seen, the present invention provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction, which can achieve a significant leap in calculation speed while ensuring high accuracy.

[0122] Example 4

[0123] This embodiment provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction. The process is the same as in Embodiment 1, and it also uses the calculation of the static avalanche breakdown voltage (BV) of a silicon carbide (SiC) multilayer floating island (MFI) power device as an example. In this embodiment, the total thickness of the drift region of the device is... Increase to ,Right now With a withstand voltage rating of 10,000V, it can be applied to ultra-high voltage scenarios, such as smart grid solid-state transformers (SST) and deep-sea wind power grid connection systems.

[0124] To achieve charge balance, the thickness of the drift region Need to increase to (Right now The number of internal P-type floating islands is increased to 5. In order to eliminate the Gibbs oscillation error of the double Fourier series expansion and ensure accurate capture at the abrupt change in the shape of the local electric field, the number of full electric field data points needs to be increased to 12,000 and the number of Fourier high-frequency expansion terms needs to be increased to 120th order. The traditional full analytical method needs to repeatedly perform double repeating loop calculations in the iterative search, which takes a long time for a single solution and faces time waste.

[0125] Specifically, the device structure parameters, control parameters, and physical model parameters in this embodiment are shown in Table 7 (parameters not listed are consistent with those in Embodiment 1), and the accuracy consistency results are as follows: Figure 5 As shown;

[0126] Table 7

[0127]

[0128] In this embodiment, for adaptive feature node number compression, the DP algorithm can identify the physical abrupt change points induced by the 5-layer floating islands, adaptively compressing 12,000 reference electric field nodes to 359 feature nodes, achieving a data compression ratio of 97.01%; for adaptive locking of the physical search domain, based on the extracted reference maximum electric field extremum, combined with the critical electric field of 4H-SiC material... The system dynamically calculates and locks the upper limit of the optimization search domain to 31137.5 V, replacing the traditional manual static setting and improving the rationality of the initial interval. For decoupling of computational load, the time-consuming double Fourier calculation is limited to the initial zero bias reference electric field extraction stage (executed only once). In the subsequent 15 optimization iterations, linear mapping and interpolation are performed only based on 359 feature nodes, avoiding the repetitive computational time consumption of the traditional method of repeatedly calling higher-order formulas.

[0129] Through adaptive calculation, the initial physical search limit of the device was locked at 31137.5 V. The full analysis method was also used as a comparison. The performance comparison between this embodiment and the comparison is shown in Table 8.

[0130] Table 8

[0131]

[0132] In summary, based on Table 8 and... Figure 5 As can be seen, the present invention provides a method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction, which can achieve a significant leap in calculation speed while ensuring high accuracy.

[0133] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. A method for calculating the breakdown voltage of power devices based on electric field dimension reduction, characterized in that, Includes the following steps: Step 1: Establish a physical model of the target semiconductor device and solve for the reference dense electric field distribution under zero bias conditions; Step 2: Set the absolute error threshold, and adaptively extract the feature nodes in the reference total electric field distribution based on the absolute error threshold to obtain the reference sparse electric field distribution; Step 3: Set the initial search domain of the test voltage according to the reference dense electric field distribution, and complete the test voltage initialization; Step 4: Under the current test voltage, update the sparse electric field distribution based on linear offset, and reconstruct it into a dense electric field distribution through linear interpolation; Step 5: Based on the reconstructed dense electric field distribution, the avalanche collision ionization model is used to determine whether the target semiconductor device has reached the breakdown condition. Step 6: Based on the judgment result, update the test voltage search domain and the current test voltage using the binary search method, and perform iterative updates until the cutoff condition is met. The test voltage at the cutoff point is the static avalanche breakdown voltage of the target semiconductor device.

2. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 1, characterized in that, In step 1, the input parameters of the physical model of the target semiconductor device include: device physical structure parameters, material doping parameters, and a one-dimensional initial spatial grid sequence; With the applied bias boundary condition set to 0, the input parameters are substituted into the physical model to calculate the one-dimensional initial spatial mesh sequence. Transverse electric field components of each dense grid node under zero bias With longitudinal electric field component The reference dense electric field distribution of the target semiconductor device under zero bias condition is calculated. : 。 3. The method for calculating the breakdown voltage of power devices based on electric field dimensionality reduction according to claim 2, characterized in that, Device physical structure parameters include: total thickness of the drift region Device half-cell width P-type floating island thickness and the width of the P-type floating island Material doping parameters include: N-type doping concentration in the drift region. With respect to the P-type doping concentration in the floating island region .

4. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 2, characterized in that, When the target semiconductor device is a silicon carbide (SiC) device, the physical model adopts a two-dimensional analytical physical model based on the Poisson equation and double Fourier series expansion.

5. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 1, characterized in that, The specific process of step 2 is as follows: Set absolute error threshold : , To preset the relative error tolerance ratio, As a reference dense electric field distribution, This represents the maximum background electric field value of the reference dense electric field distribution; Based on the aforementioned absolute error threshold, the feature nodes in the reference total electric field distribution are adaptively extracted using the DP algorithm to obtain the feature grid sequence. Corresponding feature grid sequence The baseline sparse electric field distribution is extracted, where the transverse electric field component is: Longitudinal electric field component .

6. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 1, characterized in that, The specific process of step 3 is as follows: From the reference sparse electric field distribution Extracting the maximum background electric field value: ， Combined with the critical breakdown electric field constant of the target semiconductor material Total thickness of the drift region and preset upper limit redundancy coefficient Calculate the initial estimate of the theoretical pressure resistance limit. : ， Set the search domain to , As the lower bound for the search, initialized to ; The upper limit of the search is initialized to... The initialization of the trial voltage is then: .

7. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 1, characterized in that, The specific process of step 4 is as follows: Based on the current test voltage The baseline sparse electric field distribution is migrated and updated to obtain the updated transverse electric field components. With longitudinal electric field component : ， ， in, For the characteristic grid sequence, The transverse electric field component of the reference sparse electric field distribution is: The longitudinal electric field component is the reference sparse electric field distribution. The total thickness of the drift region of the target semiconductor device; Calculate the total electric field distribution of the test : ； Then, a linear interpolation method is used to determine the total electric field distribution of the experiment. From sparse feature sequences Mapping to a one-dimensional initial spatial grid sequence Thus, the current test voltage is obtained. Reconstructed electric field distribution .

8. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 1, characterized in that, The specific process of step 5 is as follows: Reconstructing the electric field distribution Substituting into the Chynoweth collisional ionization empirical model, the electron ionization rate of each feature grid is calculated. With hole ionization rate : ， ， in, , , , The collision ionization coefficient is constant. This represents the electric field value corresponding to the feature grid. The total electron integral is obtained by performing Townsend avalanche multiplication discrete numerical integration using the trapezoidal integral method. Total integral of holes : ， ， Based on the integral extreme value Determine whether the target semiconductor device has reached the breakdown condition: If If the current test voltage reaches or exceeds the breakdown condition, then it is determined that the current test voltage has reached or exceeded the breakdown condition; if If the current test voltage does not meet the breakdown condition, it is determined that the current test voltage does not meet the breakdown condition.

9. The method for calculating the breakdown voltage of power devices based on electric field dimension reduction according to claim 1, characterized in that, The specific process of step 6 is as follows: Current test voltage When the breakdown condition is met or exceeded, update the upper limit of the test voltage search domain. : ; Current test voltage If the breakdown condition is not met, update the lower limit of the test voltage search domain. : ; The voltage search interval is gradually narrowed with each iteration. When the length of the search interval meets the convergence accuracy condition under a preset threshold, the loop ends and the static avalanche breakdown voltage of the target device is output. .

Images