Semiconductor device high-frequency electromagnetic characteristic and thermal stability collaborative optimization design method

Abstract

The invention provides a semiconductor device high-frequency electromagnetic characteristic and thermal stability collaborative optimization design method, which relates to the technical field of semiconductor device design, and comprises the following steps: arranging electromagnetic field detection points at grid nodes of a device, establishing a mapping relation between electromagnetic field energy loss and temperature, and analyzing thermal distribution characteristics by using a multi-scale decomposition network. The temperature change trend is predicted in combination with the recurrent neural network, the material parameter mapping relation is constructed, the dielectric constant and the conductivity distribution are optimized, collaborative optimization of the electromagnetic property and the thermal stability is achieved, and the performance stability and reliability of the semiconductor device are effectively improved.

Description

Technical Field

[0001] The present invention relates to semiconductor device design technology, and in particular to a method for collaboratively optimizing the high-frequency electromagnetic characteristics and thermal stability of a semiconductor device. Background Art

[0002] Semiconductor devices generate a large amount of heat when operating at high frequencies. Traditional device design methods often separate electromagnetic and thermal characteristics and optimize them separately, resulting in a large deviation between the performance of the device during actual operation and the design expectations, affecting the reliability and service life of the device.

[0003] Existing technologies typically use simplified thermal models to predict device temperatures. These models fail to accurately reflect the conversion of electromagnetic field energy loss into heat, and lack precise identification of hotspots and analysis of temperature evolution patterns, resulting in a lack of effective theoretical guidance for device structure optimization. Furthermore, the correlation between thermal conductivity characteristics and material parameters is not fully considered during the optimization of material parameters, making it difficult to achieve coordinated optimization of device electromagnetic properties and thermal stability.

[0004] Therefore, there is an urgent need for a design method that can accurately describe the relationship between electromagnetic field energy loss and temperature distribution, and at the same time optimize the configuration of material parameters based on thermal conductivity characteristics, so as to improve the performance stability of semiconductor devices under high-frequency working conditions. Summary of the Invention

[0005] The embodiments of the present invention provide a method for collaboratively optimizing the high-frequency electromagnetic characteristics and thermal stability of a semiconductor device, which can solve the problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for collaboratively optimizing the high-frequency electromagnetic characteristics and thermal stability of a semiconductor device, comprising: Arrange electromagnetic field detection points at the grid nodes of semiconductor devices, collect the electric field strength and magnetic field strength of the electromagnetic field detection points, calculate the electromagnetic field energy loss of the grid nodes based on the electric field strength and magnetic field strength, use piecewise linear interpolation method to establish a mapping relationship between electromagnetic field energy loss and temperature, and convert the electromagnetic field energy loss into heat source power density based on the mapping relationship; Input the heat source power density into the multi-scale decomposition network to obtain heat distribution characteristic data, determine the location information of the hot spot area based on the heat distribution characteristic data, establish the time series temperature data of the hot spot area, input the time series temperature data into the recursive neural network to obtain the temperature change trend of the hot spot area, and generate a temperature gradient matrix based on the temperature change trend; The temperature gradient matrix is ​​decomposed to obtain heat conduction characteristics. Based on the heat conduction characteristics, the device area is divided and heat conduction channels between the areas are established. The heat conduction channels are associated with the temperature gradient to construct a material parameter mapping relationship. The dielectric constant distribution and electrical conductivity distribution are generated through parameter sensitivity analysis and boundary continuity optimization. The dielectric constant distribution and conductivity distribution are used as initial parameters, and material parameters are configured for each area of ​​the semiconductor device. The new temperature distribution is calculated based on the configuration results, and the new temperature distribution is compared with the preset temperature threshold. When the temperature distribution meets the preset threshold requirements, the current dielectric constant distribution and conductivity distribution are determined as the optimized structural parameters of the semiconductor device.

[0007] In an optional embodiment, Arrange electromagnetic field detection points at the grid nodes of semiconductor devices, collect the electric field strength and magnetic field strength of the electromagnetic field detection points, and calculate the electromagnetic field energy loss of the grid nodes based on the electric field strength and magnetic field strength. Arranging electromagnetic field detection points at the grid nodes, the electromagnetic field detection points including electric field detectors and magnetic field detectors, sampling output signals of the electric field detectors and the magnetic field detectors to obtain electromagnetic field sampling data; Filtering the electromagnetic field sampling data to obtain the electric field intensity and magnetic field intensity of the electromagnetic field detection point; Acquiring temperature data of the semiconductor device, determining material parameters based on the temperature data, calculating a dielectric loss value based on the electric field strength, calculating a magnetic loss value based on the magnetic field strength, and calculating a conduction loss value based on the electric field strength and the material parameters; The dielectric loss value, the magnetic loss value, and the conduction loss value are added together to obtain the electromagnetic field energy loss of the grid node.

[0008] In an optional embodiment, The mapping relationship between electromagnetic field energy loss and temperature is established using the piecewise linear interpolation method. The electromagnetic field energy loss is converted into heat source power density according to the mapping relationship, including: Adaptively filter the electromagnetic field energy loss data, dynamically adjust the filter parameters according to the signal change amplitude, obtain the filtered electromagnetic field energy loss data, calculate the distribution probability of the filtered electromagnetic field energy loss data, determine the inflection point of the probability change as the segmentation point, and divide the range of the electromagnetic field energy loss into multiple sub-intervals; In each subinterval, the correlation coefficient between the electromagnetic field energy loss data and the temperature data is calculated, the correlation coefficient is used as a weight value, and a piecewise linear interpolation method is used to establish a subinterval mapping function for the weighted data; Calculate the slope difference of the mapping function of adjacent subintervals at the segmentation point, adjust the mapping function coefficient according to the slope difference, make the mapping function of adjacent subintervals continuous at the segmentation point, and obtain the piecewise linear mapping relationship between electromagnetic field energy loss and temperature; Real-time electromagnetic field energy loss data of the semiconductor device is obtained, the real-time electromagnetic field energy loss data is substituted into the piecewise linear mapping relationship to obtain a temperature prediction value, a corresponding material parameter correction factor is obtained according to the temperature prediction value, and the real-time electromagnetic field energy loss data is multiplied by the material parameter correction factor to obtain the heat source power density.

[0009] In an optional embodiment, Input the heat source power density into the multi-scale decomposition network to obtain the heat distribution characteristic data. According to the heat distribution characteristic data, the location information of the hot spot area is determined. The time series temperature data of the hot spot area is established, including: Inputting the heat source power density into a multi-scale decomposition network to perform signal decomposition to obtain a decomposition coefficient matrix group, and dividing the decomposition coefficient matrix group into multiple scale levels; Calculating local statistical features of decomposition coefficient matrices at multiple scale levels to obtain characteristic variance matrices and characteristic mean matrices, constructing a dynamic decision function based on the characteristic variance matrix and characteristic mean matrix, and performing a comparison operation between the decomposition coefficient matrix group and the dynamic decision function to obtain heat distribution characteristic data; The positional relationship of the hotspot area is calculated based on the heat distribution characteristic data, and the positional relationship is analyzed and processed using a density clustering method to obtain the position information of the hotspot area. Temperature data is collected based on the position information of the hotspot area, and the temperature data is sorted in chronological order to establish the time series temperature data of the hotspot area.

[0010] In an optional embodiment, Input the time series temperature data into the recurrent neural network to obtain the temperature change trend of the hot spot area. The temperature gradient matrix generated according to the temperature change trend includes: Perform feature enhancement processing on time series temperature data, extract temperature value, temperature change, temperature change rate and hot spot area volume, and construct temperature feature vector; Calculating the temperature difference between the hotspot area and the adjacent area based on the temperature feature vector, obtaining the shape characteristics and position information of the hotspot area, and generating a spatial correlation feature; Input the temperature feature vector and spatial correlation feature into a two-layer recursive neural network, extract the short-term temperature change feature and temperature evolution law through the first layer network, input the short-term temperature change feature and temperature evolution law into the second layer network, establish the long-term temperature dependency, and obtain the temperature change trend of the hot spot area; The amplitude and direction of the spatial temperature gradient are calculated according to the temperature change trend, the gradient singular point distribution information is extracted, and the amplitude and direction of the spatial temperature gradient and the singular point distribution information are combined to generate a temperature gradient matrix.

[0011] In an optional embodiment, The temperature gradient matrix is ​​decomposed to obtain heat conduction characteristics. Based on the heat conduction characteristics, the device area is divided and heat conduction channels between the areas are established. The heat conduction channels are associated with the temperature gradient to construct a material parameter mapping relationship. The final dielectric constant distribution and conductivity distribution are generated through parameter sensitivity analysis and boundary continuity optimization, including: performing eigenvalue decomposition and eigenvector decomposition on the temperature gradient matrix to obtain heat conduction direction characteristic data, generating a gradient intensity distribution map based on the heat conduction direction characteristic data, dividing the device into regions according to the gradient intensity distribution map to obtain material parameter priority region data, and constructing heat conduction channels between adjacent regions based on the material parameter priority region data; Performing an associative operation on the heat conduction channel and the temperature gradient matrix to establish a mapping relationship between the temperature gradient and the dielectric constant and a mapping relationship between the temperature and the conductivity, and calculating the adaptive weight of the material parameters according to the mapping relationship; Calculating the parameter sensitivity of each region using the adaptive weight, comparing the parameter sensitivity with a preset threshold to obtain a parameter update interval, and generating local parameter adjustment data according to the parameter update interval; Performing boundary continuity analysis on the local parameter adjustment data, calculating parameter gradients and boundary features of adjacent regions, constructing boundary feature vectors, classifying boundary types based on the boundary feature vectors, calculating regional comprehensive weights based on the obtained boundary types, performing layered smoothing on the parameters, and generating a parameter smoothing transition scheme; Physical constraints of materials are imposed on the parameter smooth transition scheme, a mapping relationship between the temperature field and the material parameters is established, and the dielectric constant distribution and the conductivity distribution of each region of the device are calculated.

[0012] In an optional embodiment, Performing boundary continuity analysis on the local parameter adjustment data, calculating the parameter gradient and boundary characteristics of adjacent regions, constructing boundary feature vectors, classifying boundary types based on the boundary feature vectors, calculating regional comprehensive weights based on the obtained boundary types, performing layered smoothing on the parameters, and generating a parameter smoothing transition scheme include: Using a multi-directional differential template to analyze local parameter adjustment data, obtain parameter gradient information, calculate the normal gradient, tangential gradient and boundary curvature based on the parameter gradient information, and extract boundary features; Normalizing the boundary features, and combining the normalized parameter gradient amplitude, gradient direction, boundary curvature, and parameter jump variable to construct a boundary feature vector; Establishing a boundary classification matrix based on the boundary feature vector, determining a first boundary classification according to the parameter gradient amplitude, determining a second boundary classification in combination with the gradient direction continuity, and determining a third boundary classification by the boundary curvature feature; Constructing a weight association matrix based on the boundary classification, inputting boundary classification weights, regional area weights, parameter importance weights, and location correlation weights into the weight association matrix to calculate the regional comprehensive weight; A smoothing coefficient is selected based on the comprehensive weight of the area, a first smoothing coefficient is used for the first boundary classification area, a second smoothing coefficient is used for the second boundary classification area, and a third smoothing coefficient is used for the third boundary classification area. The boundary continuity of the smoothing result is evaluated, and when the boundary continuity meets the preset conditions, a parameter smoothing transition scheme is output.

[0013] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0014] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0015] In this embodiment, by establishing a mapping relationship between electromagnetic field energy loss and temperature, a close coupling of electromagnetic field analysis and thermal analysis is achieved, overcoming the optimization inconsistency problem caused by independent analysis of electromagnetic characteristics and thermal characteristics in traditional design methods, making the design results more accurate and reliable. The combination of multi-scale decomposition network and recursive neural network is used to process the time series temperature data of the hot spot area, which can effectively capture the hot spot distribution and temperature evolution trend of semiconductor devices under high-frequency working conditions, improve the accuracy of thermal stability prediction, make thermal management strategies more targeted, and significantly improve the thermal reliability of the device. Through parameter sensitivity analysis and boundary continuity optimization, a mapping relationship between material parameters and temperature gradients is established, and adaptive optimization of device structural parameters is achieved, so that the optimized semiconductor device has better thermal stability while maintaining good electromagnetic characteristics, extending the service life of the device and improving the overall performance and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1Schematic diagram of the process of the collaborative optimization design method for high-frequency electromagnetic characteristics and thermal stability of semiconductor devices according to an embodiment of the present invention; Figure 2 This is a comparison diagram of the efficiency of the electromagnetic field energy loss data processing method according to an embodiment of the present invention; Figure 3 This is a comparison chart of the temperature gradient analysis and material parameter optimization effects of an embodiment of the present invention. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0018] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0019] Figure 1 FIG. 1 is a flow chart of a method for collaboratively optimizing the design of high-frequency electromagnetic characteristics and thermal stability of a semiconductor device according to an embodiment of the present invention. Figure 1 As shown, the method includes: Arrange electromagnetic field detection points at the grid nodes of semiconductor devices, collect the electric field strength and magnetic field strength of the electromagnetic field detection points, calculate the electromagnetic field energy loss of the grid nodes based on the electric field strength and magnetic field strength, use piecewise linear interpolation method to establish a mapping relationship between electromagnetic field energy loss and temperature, and convert the electromagnetic field energy loss into heat source power density based on the mapping relationship; Input the heat source power density into the multi-scale decomposition network to obtain heat distribution characteristic data, determine the location information of the hot spot area based on the heat distribution characteristic data, establish the time series temperature data of the hot spot area, input the time series temperature data into the recursive neural network to obtain the temperature change trend of the hot spot area, and generate a temperature gradient matrix based on the temperature change trend; The temperature gradient matrix is ​​decomposed to obtain heat conduction characteristics. Based on the heat conduction characteristics, the device area is divided and heat conduction channels between the areas are established. The heat conduction channels are associated with the temperature gradient to construct a material parameter mapping relationship. The dielectric constant distribution and electrical conductivity distribution are generated through parameter sensitivity analysis and boundary continuity optimization. The dielectric constant distribution and conductivity distribution are used as initial parameters, and material parameters are configured for each area of ​​the semiconductor device. The new temperature distribution is calculated based on the configuration results, and the new temperature distribution is compared with the preset temperature threshold. When the temperature distribution meets the preset threshold requirements, the current dielectric constant distribution and conductivity distribution are determined as the optimized structural parameters of the semiconductor device.

[0020] In an optional embodiment, arranging electromagnetic field detection points at grid nodes of the semiconductor device, collecting electric field strength and magnetic field strength at the electromagnetic field detection points, and calculating the electromagnetic field energy loss of the grid nodes based on the electric field strength and magnetic field strength includes: Arranging electromagnetic field detection points at the grid nodes, the electromagnetic field detection points including electric field detectors and magnetic field detectors, sampling output signals of the electric field detectors and the magnetic field detectors to obtain electromagnetic field sampling data; Filtering the electromagnetic field sampling data to obtain the electric field intensity and magnetic field intensity of the electromagnetic field detection point; Acquiring temperature data of the semiconductor device, determining material parameters based on the temperature data, calculating a dielectric loss value based on the electric field strength, calculating a magnetic loss value based on the magnetic field strength, and calculating a conduction loss value based on the electric field strength and the material parameters; The dielectric loss value, the magnetic loss value, and the conduction loss value are added together to obtain the electromagnetic field energy loss of the grid node.

[0021] For example, electromagnetic field detection points are arranged at grid nodes of semiconductor devices. Each electromagnetic field detection point consists of a group of electric field detectors and magnetic field detectors. The electric field detectors collect changes in the amplitude of the electric field signal at that node, while the magnetic field detectors synchronously acquire changes in the spatial magnetic field amplitude. The output signals of these two types of sensors are sampled at a high frequency in the time domain, with a sampling frequency of no less than 100 trillion times per second, to ensure that all fluctuation characteristics of the electromagnetic response of high-frequency devices during operation are captured. The sampled raw data includes unstable noise and background disturbance signals. To improve data accuracy, a sliding window filtering method is used for processing. Each window contains 128 consecutive data points. A dynamic threshold range is set to filter out high-frequency noise and electromagnetic background interference, and the effective electromagnetic intensity values ​​of the electric and magnetic field detection points are output. For the electric field intensity, its instantaneous peak value and time average value are recorded; for the magnetic field intensity, its maximum fluctuation amplitude and dominant direction are extracted.

[0022] To meet the requirements of thermal response coupling modeling, temperature sensors are deployed at the spatial locations corresponding to the aforementioned electromagnetic sampling nodes to collect temperature data during semiconductor device operation. The temperature collection interval is set to 0.5 seconds to ensure that the heat accumulation process is fully covered. The collected temperature data is used to identify the device's thermal state during its operating phase. Combined with the material parameter mapping information for different temperatures in the material datasheet, parameters such as relative permittivity, magnetic permeability, and electrical conductivity at the current node are determined. These parameters serve as the basis for calculating electromagnetic field energy loss. Dielectric loss is determined by combining the electric field strength and relative permittivity data; magnetic loss is determined by the relationship between magnetic field strength and magnetic permeability; and conduction loss is calculated based on the relationship between electric field strength and electrical conductivity. These three loss quantities are uniformly normalized and weighted superimposed to obtain the electromagnetic field energy loss per unit time for that grid node.

[0023] Taking a high-frequency RF switch device with a structural dimension of 8 mm x 8 mm as an example, a 64 x 64 grid node network was established on its surface, with each node deploying a composite electromagnetic field detection point and a temperature sensor. Under operating conditions of 5 GHz and a channel power consumption of 0.5 W, a typical node detected a peak electric field strength of 280 V / m and a magnetic field strength of 3.2 A / m in the center area, corresponding to a temperature of 78°C. Based on data from its material library, the relative dielectric constant at this point is 4.5, the electrical conductivity is 1.6 S / m, and the magnetic permeability is 1.2. Based on the above physical quantity relationships, the dielectric loss, magnetic loss, and conduction loss are estimated to be approximately 0.013 W, 0.005 W, and 0.007 W, respectively. Adding these three together, the electromagnetic field energy loss per unit time for this node is 0.025 W.

[0024] After obtaining the energy losses of all grid nodes, a heat source power density distribution map is constructed in the form of a two-dimensional matrix and fed into a multi-scale decomposition network model as input. Based on a hierarchical residual structure and a spatial attention mechanism, the multi-scale decomposition network extracts the spatial characteristics and gradient change boundaries of hotspot-dense areas in the heat source distribution. The heat distribution feature data is compressed using convolution operations and then reconstructed into a recognizable hotspot distribution map through a cross-scale fusion module. Significant areas of heat distribution are marked as potential high-temperature clusters, and their heat accumulation changes are further extracted to generate time series temperature data corresponding to each hotspot.

[0025] Time series temperature data is fed into a recurrent neural network for processing. The network utilizes bidirectional memory cells, which simultaneously capture both the rate of temperature rise and the trend of temperature drop fluctuations. For each hotspot, a series of temperature change predictions for the next five seconds is output. The difference between the current temperature and the predicted value is used to construct a local temperature gradient matrix. Each temperature gradient matrix contains information on the direction and rate of temperature change at 16 grid points within the hotspot, serving as input for heat transfer trends.

[0026] After obtaining the temperature gradient matrix, the eigendecomposition method is used to convert it into the dominant heat conduction eigenvectors. These vectors represent the main propagation direction and intensity of heat in space. Through cluster analysis of the heat conduction eigenvectors, the entire device structure is divided into multiple regional units, each unit area has similar heat conduction characteristics. The heat exchange relationship between regions is modeled as a heat conduction channel. The heat conduction channel length, channel cross-sectional area and boundary thermal resistance are comprehensively evaluated to establish a description of the heat conduction flux of each channel.

[0027] By mapping the thermal channels to the temperature gradient matrix, a material parameter mapping network is constructed. This network automatically adjusts the dielectric constant and conductivity distribution within a region to maximize thermal conductivity. Before configuring material parameters, the sensitivity of each parameter is analyzed to identify key parameters that influence the uniformity of heat diffusion. Parameter continuity at boundaries is then adjusted to avoid heat flow distortion caused by sudden parameter changes. Ultimately, a dielectric constant and conductivity distribution map is generated for each region.

[0028] The above parameter distribution is used as input to update the material configuration of each functional area of ​​the semiconductor device. The temperature distribution of the device under high-frequency operating conditions is recalculated according to the new parameters to obtain the steady-state thermal field map of the entire device. The highest temperature in the thermal field map is compared with the set temperature threshold. If the temperature of all nodes is below the threshold, it means that the parameter configuration scheme meets the optimization requirements. At this time, the current dielectric constant distribution map and conductivity distribution map are confirmed as the final collaborative optimization design results of the semiconductor device, completing the comprehensive optimization process between high-frequency electromagnetic characteristics and thermal stability. If the calculation result does not meet the set threshold, the parameter sensitivity evaluation and regional material parameter update steps are restarted until the optimization is completed.

[0029] In this embodiment, the electric and magnetic field intensities are accurately obtained through electromagnetic field detection points, and the energy loss is calculated in combination with temperature data to achieve an effective mapping of electromagnetic behavior to thermal effects. The introduction of multi-scale decomposition and recursive network structure makes the identification of hot spots and the prediction of temperature trends more accurate, providing data support for subsequent material parameter adjustments. Through heat conduction characteristic analysis and regional division, a thermal management-oriented material distribution optimization mechanism is established to effectively reduce the risk of local temperature rise. Ultimately, a device design scheme with both high-frequency response capability and thermal stability is obtained, which improves the operational reliability and service life of the device.

[0030] In an optional embodiment, a piecewise linear interpolation method is used to establish a mapping relationship between electromagnetic field energy loss and temperature, and converting the electromagnetic field energy loss into heat source power density according to the mapping relationship includes: Adaptively filter the electromagnetic field energy loss data, dynamically adjust the filter parameters according to the signal change amplitude, obtain the filtered electromagnetic field energy loss data, calculate the distribution probability of the filtered electromagnetic field energy loss data, determine the inflection point of the probability change as the segmentation point, and divide the range of the electromagnetic field energy loss into multiple sub-intervals; In each subinterval, the correlation coefficient between the electromagnetic field energy loss data and the temperature data is calculated, the correlation coefficient is used as a weight value, and a piecewise linear interpolation method is used to establish a subinterval mapping function for the weighted data; Calculate the slope difference of the mapping function of adjacent subintervals at the segmentation point, adjust the mapping function coefficient according to the slope difference, make the mapping function of adjacent subintervals continuous at the segmentation point, and obtain the piecewise linear mapping relationship between electromagnetic field energy loss and temperature; Real-time electromagnetic field energy loss data of the semiconductor device is obtained, the real-time electromagnetic field energy loss data is substituted into the piecewise linear mapping relationship to obtain a temperature prediction value, a corresponding material parameter correction factor is obtained according to the temperature prediction value, and the real-time electromagnetic field energy loss data is multiplied by the material parameter correction factor to obtain the heat source power density.

[0031] In one embodiment, a piecewise linear interpolation method is used to establish a mapping relationship between electromagnetic field energy loss and temperature. The method of converting electromagnetic field energy loss into heat source power density according to the mapping relationship can be specifically implemented as follows: When adaptively filtering electromagnetic field energy loss data, filter parameters can be dynamically adjusted during runtime to adapt to signal variations. Specifically, a bidirectional sliding window approach is used to process the raw electromagnetic field energy loss data. The window size is initially set to 5% of the total number of data points, and the standard deviation σ and mean μ are calculated within the window. When data point values ​​exceed the range of μ±2σ, it is considered a region of high variation. In this case, the window size is reduced to half to preserve signal detail. When data point values ​​are within the range of μ±0.5σ, it is considered a stable region. In this case, the window size is doubled, up to a maximum of 10% of the total number of data points, to enhance the filtering effect. This approach preserves features in regions of high variation, while applying stronger smoothing to stable regions. For example, for energy loss data from a semiconductor device, adjusting the window size to 3 points in the interface region, where data varies dramatically, retains over 80% of edge features after filtering. However, increasing the window size to 15 points in the uniform material region reduces noise by 92%.

[0032] After obtaining the filtered electromagnetic field energy loss data, calculate its distribution probability to determine the segmentation points. Sort the filtered loss data by size and divide it into 100 equally spaced intervals. Count the number of data points in each interval and divide it by the total number of data points to obtain a probability distribution curve. By calculating the first-order difference value of the probability curve, determine the position where the differential value changes significantly as the segmentation point. In practical applications, analysis of a certain type of power device shows that when the differential value exceeds 3 times the average differential value, it can be identified as a valid segmentation point. For example, for a set of energy loss distributions containing 10,000 data points, four key segmentation points were identified by this method, located at 0.05W / mm 3 , 0.35W / mm 3 , 1.2W / mm 3 and 2.8W / mm 3 The entire loss range is divided into 5 sub-intervals.

[0033] In each subinterval, it is necessary to establish a mapping relationship between electromagnetic field energy loss and temperature. First, the Pearson correlation coefficient r of the energy loss data and the corresponding temperature data in each subinterval is calculated. The correlation coefficient is calculated based on the data pairs in the subinterval, for example, in [0.05, 0.35] W / mm 3 In the subinterval, 500 sample points are taken to calculate the correlation coefficient r=0.87. The correlation coefficient is used as a weight value to adjust the contribution of the data points in the interval. For each subinterval, the linear interpolation method is used to fit the weighted data points to establish a linear mapping function of the form T=a×E+b, where T represents temperature, E represents electromagnetic field energy loss, and a and b are unknown coefficients. The coefficient of each subinterval is determined by the least squares method, for example, in [0.05, 0.35] W / mm 3 In the range, we get a=42.5℃·mm 3 / W, b=25.3℃.

[0034] To ensure the continuity of the overall mapping function, the coefficients of the mapping functions of adjacent subintervals need to be adjusted. The slope difference of the mapping functions of adjacent subintervals at the segmentation points is calculated. If the difference exceeds a preset threshold (e.g. 10%), the coefficient is adjusted. The adjustment method is to modify the slope of the linear function of the higher energy loss interval while keeping the function value at the segmentation point unchanged, so that the slope difference between it and the low energy loss interval is within the allowable range. For example, for the segmentation point 0.35W / mm 3At the first interval, the mapping function slope is 42.5, while the initial slope of the second interval is 50.8, resulting in a difference of 19.5%, exceeding the threshold. By adjusting the slope of the second interval to 46.2 and the intercept term, the two interval functions maintain value continuity at the segmentation point while keeping the slope difference below 8.7%, below the threshold. Similar processing is performed on all segmentation points to obtain a complete piecewise linear mapping relationship.

[0035] After obtaining the real-time electromagnetic field energy loss data of the semiconductor device, it is substituted into the established piecewise linear mapping relationship to predict the temperature. For example, the real-time energy loss of a test point is 0.78W / mm 3 , belongs to the third sub-interval [0.35, 1.2] W / mm 3 , the temperature prediction value is calculated to be 58.6℃ using the mapping function T=46.2×E+22.5 of this interval. According to the predicted temperature, the pre-established material parameter-temperature relationship database is queried to obtain the corresponding material parameter correction factor. This database is established based on the measured values ​​of material thermal conductivity, specific heat capacity and other parameters at different temperatures. For example, the thermal conductivity correction factor of a semiconductor material is 0.92 at 58.6℃. The real-time energy loss data is 0.78W / mm 3 Multiplying this by the correction factor of 0.92, we get the corrected heat source power density of 0.7176 W / mm 3 , as the input of the thermal analysis model, can more accurately reflect the influence of material parameters on heat conduction as the temperature changes.

[0036] In this embodiment, by introducing piecewise linear interpolation and adaptive filtering mechanisms, a more accurate and dynamically adjustable mapping relationship between electromagnetic field energy loss and temperature is achieved, making the estimated results of heat source power density more consistent with the energy transfer law under actual working conditions. By identifying the inflection point of the energy loss distribution and establishing a continuous and smooth sub-interval mapping function, the resolution and stability of temperature prediction are significantly improved. Combined with the introduction of material parameter correction factors, dynamic calibration of energy and heat coupling characteristics is achieved, enhancing the system's ability to adapt to the thermal response characteristics of different materials, thereby providing high-precision basic data support for subsequent thermal stability control and material optimization.

[0037] Figure 2This is a comparison diagram of the efficiency of the electromagnetic field energy loss data processing method according to an embodiment of the present invention, which shows the efficiency comparison of three electromagnetic field energy loss data processing methods in four different scenarios. The adaptive filtering method performed best in all scenarios, especially in high-energy scenarios, where it achieved a processing efficiency of 91.2%, which was 11.7 percentage points higher than the traditional parameter correction method. The piecewise linear mapping method was at an intermediate level and performed relatively stably in medium and high energy scenarios. The traditional parameter correction method had the lowest efficiency in multi-field interference scenarios, at only 69.2%. The results show that the method of adaptive filtering combined with piecewise linear mapping proposed in this paper can better adapt to different electromagnetic field environments, especially maintain a high processing efficiency in complex interference environments, and provide more reliable technical support for semiconductor device temperature prediction and heat source power density calculation.

[0038] In an optional embodiment, the heat source power density is input into a multi-scale decomposition network to obtain heat distribution characteristic data, and the location information of the hot spot area is determined based on the heat distribution characteristic data. Establishing the time series temperature data of the hot spot area includes: Inputting the heat source power density into a multi-scale decomposition network to perform signal decomposition to obtain a decomposition coefficient matrix group, and dividing the decomposition coefficient matrix group into multiple scale levels; Calculating local statistical features of decomposition coefficient matrices at multiple scale levels to obtain characteristic variance matrices and characteristic mean matrices, constructing a dynamic decision function based on the characteristic variance matrix and characteristic mean matrix, and performing a comparison operation between the decomposition coefficient matrix group and the dynamic decision function to obtain heat distribution characteristic data; The positional relationship of the hotspot area is calculated based on the heat distribution characteristic data, and the positional relationship is analyzed and processed using a density clustering method to obtain the position information of the hotspot area. Temperature data is collected based on the position information of the hotspot area, and the temperature data is sorted in chronological order to establish the time series temperature data of the hotspot area.

[0039] The process of inputting the heat source power density into the multi-scale decomposition network to obtain the heat distribution characteristic data, determining the location information of the hot spot area based on the heat distribution characteristic data, and establishing the time series temperature data of the hot spot area is as follows.

[0040] In this embodiment, heat source power density data of an electronic device is collected. This data can be obtained using a thermal imager or a temperature sensor array. The obtained heat source power density data is a two-dimensional matrix, where each element of the matrix represents the power density value at the corresponding location, in units of watts per square centimeter (W / cm²). For example, the obtained power density matrix size is 128×128, covering an area of ​​10cm×10cm, and the value range is between 0.01-2.5W / cm².

[0041] The obtained heat source power density matrix is ​​input into a pre-designed multi-scale decomposition network for processing. This multi-scale decomposition network consists of five convolutional and pooling layers, which decompose the input signal into different frequency components. The first layer of the network uses 64 3×3 convolution kernels with a stride of 1, followed by a 2×2 max pooling layer; the second layer uses 128 3×3 convolution kernels with a stride of 1, followed by a 2×2 max pooling layer; the third layer uses 256 3×3 convolution kernels with a stride of 1; the fourth layer uses 256 3×3 convolution kernels with a stride of 1; and the fifth layer uses 128 3×3 convolution kernels with a stride of 1. After processing through this network, five groups of decomposition coefficient matrices at different scale levels are obtained.

[0042] For the first scale level (highest frequency component), the decomposition coefficient matrix size is 64×64×64, representing the finest heat distribution features; for the second scale level, the decomposition coefficient matrix size is 32×32×128; for the third scale level, the decomposition coefficient matrix size is 32×32×256; for the fourth scale level, the decomposition coefficient matrix size is 32×32×256; for the fifth scale level (lowest frequency component), the decomposition coefficient matrix size is 32×32×128, representing the coarsest heat distribution features.

[0043] Local statistical features are calculated for the decomposition coefficient matrix at each scale level. Specifically, a 9×9 sliding window with a step size of 3 is used to calculate the variance and mean of the coefficients within the window, forming a feature variance matrix and a feature mean matrix, respectively. For example, for the first scale level, the generated feature variance matrix and feature mean matrix are both 19×19×64 in size; for the fifth scale level, the generated feature variance matrix and feature mean matrix are both 9×9×128 in size.

[0044] A dynamic decision function is constructed based on the feature variance matrix and the feature mean matrix. This decision function is calculated by applying a threshold to each element in the feature variance matrix and performing a weighted calculation based on the value of the corresponding position in the feature mean matrix. Specifically, for an element value in the feature variance matrix of the i-th scale level, if it is greater than 1.5 times the mean variance of that level, it is marked as a candidate hotspot and assigned a value of twice the mean feature value at that location; otherwise, it is assigned a value of 0. In this way, a binary decision matrix is ​​constructed for each scale level.

[0045] The decomposition coefficient matrix group is compared with the dynamic judgment function to obtain heat distribution characteristic data. Specifically, for each scale level, the decomposition coefficient matrix is ​​element-wise multiplied with the corresponding judgment matrix to obtain characteristic data highlighting areas of thermal anomalies. The characteristic data of the five scale levels are then weighted and fused, with weights of 0.1, 0.15, 0.2, 0.25, and 0.3, respectively, to obtain the final heat distribution characteristic data.

[0046] The hotspot locations are calculated based on the heat distribution feature data. The heat distribution feature data is normalized to a value between 0 and 1, and points with a value greater than 0.6 are selected as candidate hotspot locations. For the example 128×128 matrix, approximately 200–300 candidate hotspot locations may be identified.

[0047] Density-based spatial clustering applications with noise (DBSCAN) was used to perform density clustering analysis on candidate hotspot locations. Adjacent candidate hotspot locations were clustered into hotspot regions, with a clustering distance threshold of 5 pixels and a minimum sample size of 8. This approach identified three major hotspot regions within the 128×128 power density matrix, located near coordinates (32, 45), (78, 62), and (105, 30), with a radius of approximately 10-15 pixels.

[0048] Temperature sensors were set up based on the identified hotspot location information to collect temperature data. Specifically, five temperature sensors were placed inside each hotspot, sampling at a frequency of 2 Hz for 30 minutes, obtaining 3,600 temperature data points for each sensor. For example, in the hotspot area at position (32, 45), the initial temperatures of the five sensors were 42.5°C, 43.1°C, 41.8°C, 42.9°C, and 41.5°C, respectively. After 30 minutes, the temperatures rose to 56.2°C, 57.4°C, 54.9°C, 57.1°C, and 53.8°C, respectively.

[0049] The collected temperature data are sorted in chronological order to establish the time series temperature data of the hot spot area. For each hot spot area, the average temperature value of the five sensors at each time point is calculated to form a time series temperature sequence containing 3600 data points. At the same time, the temperature change rate, that is, the temperature difference between adjacent time points, is recorded to form a temperature change rate sequence. For example, in the hot spot area at position (78, 62), the recorded average temperature rises at a rate of 0.12°C / minute in the first 5 minutes, rises at a rate of 0.25°C / minute in the 10-15 minute interval, and stabilizes in the last 5 minutes, with a fluctuation range of less than 0.05°C / minute. These time series temperature data can be used for subsequent hot spot evolution prediction and cooling system optimization design.

[0050] In this embodiment, by inputting the heat source power density into the multi-scale decomposition network, the heat distribution characteristics at different scales can be accurately extracted, enhancing the ability to analyze complex heat conduction behaviors. With the help of the dynamic judgment function constructed by the characteristic variance and mean, the recognition accuracy of thermal anomaly areas is effectively improved. The density clustering method is used to analyze the heat distribution characteristics, and the spatial location of the hot spot area is accurately located. The time series temperature data of the hot spot area is further collected and constructed, providing a reliable basis for the subsequent temperature evolution trend prediction and thermal stability control, thereby significantly improving the thermal management efficiency and stability of semiconductor devices under high-frequency operating conditions.

[0051] In an optional embodiment, the time series temperature data is input into a recursive neural network to obtain the temperature change trend of the hot spot area, and the temperature gradient matrix is ​​generated according to the temperature change trend, including: Perform feature enhancement processing on time series temperature data, extract temperature value, temperature change, temperature change rate and hot spot area volume, and construct temperature feature vector; Calculating the temperature difference between the hotspot area and the adjacent area based on the temperature feature vector, obtaining the shape characteristics and position information of the hotspot area, and generating a spatial correlation feature; Input the temperature feature vector and spatial correlation feature into a two-layer recursive neural network, extract the short-term temperature change feature and temperature evolution law through the first layer network, input the short-term temperature change feature and temperature evolution law into the second layer network, establish the long-term temperature dependency, and obtain the temperature change trend of the hot spot area; The amplitude and direction of the spatial temperature gradient are calculated according to the temperature change trend, the gradient singular point distribution information is extracted, and the amplitude and direction of the spatial temperature gradient and the singular point distribution information are combined to generate a temperature gradient matrix.

[0052] This embodiment provides a method for inputting time series temperature data into a recursive neural network to obtain a temperature change trend of a hot spot area and generate a temperature gradient matrix according to the temperature change trend.

[0053] In practical applications, temperature data can be collected from conductor devices under different operating conditions, including temperature values ​​at multiple measurement points on the device surface. For example, 20 temperature sensors are placed on a 100W conductor device, collecting data every 5 seconds for 30 minutes, resulting in temperature data at 360 time points. This raw temperature data requires feature enhancement to extract more valuable feature information. Specifically, four types of features are extracted from the time-series temperature data: temperature value, temperature change, temperature change rate, and hotspot volume. The temperature value is directly obtained from the sensor reading; the temperature change is calculated by calculating the temperature difference between adjacent time points, such as the current temperature minus the previous temperature; the temperature change rate is the temperature change divided by the time interval, reflecting the speed of temperature change; and the hotspot volume is the area where the temperature exceeds a certain threshold (e.g., 85°C). This can be estimated by counting the number of measurement points exceeding the threshold and combining it with the distribution density of the measurement points. These four types of features are combined to form a temperature feature vector, one for each time point, with a vector dimension four times the number of measurement points.

[0054] Calculating the temperature difference between the hotspot and adjacent areas based on the temperature eigenvector is the key next step. A hotspot is typically defined as a continuous region with a temperature exceeding a certain threshold, such as above 85°C. Adjacent areas are those directly adjacent to the hotspot but with temperatures below the threshold. By calculating the temperature difference between each measurement point within the hotspot and those in the adjacent areas, the boundary information of the hotspot can be obtained. For example, if a hotspot contains five measurement points with an average temperature of 90°C, while its adjacent area contains eight measurement points with an average temperature of 80°C, the temperature difference between the hotspot and adjacent areas is 10°C. This temperature difference distribution can be used to analyze the shape characteristics of the hotspot, such as whether it is circular, elliptical, or irregular. The coordinates of the hotspot's location on the device are also recorded to obtain location information. Combining the shape characteristics and location information, spatial correlation features are generated. These features describe the spatial distribution characteristics of the hotspot and help understand the conduction and diffusion of heat in conductive devices.

[0055] Inputting temperature feature vectors and spatial correlation features into a two-layer recurrent neural network is the key to achieving temperature trend prediction. This two-layer recurrent neural network consists of two layers of long short-term memory (LSTM) or gated recurrent units (GRU). In practical implementation, a first-layer LSTM network with 64 hidden units and a second-layer LSTM network with 32 hidden units can be constructed. The first-layer network is primarily responsible for extracting short-term temperature variation characteristics and temperature evolution patterns. Short-term variation characteristics reflect temperature fluctuations over a period of several minutes, such as the temperature response to workload changes. Temperature evolution patterns describe temperature variations under stable operating conditions, such as heating rate and stabilization time. In specific implementation, the temperature feature vector and spatial correlation features at each time step are combined into an input vector and fed into the first-layer LSTM network. The network processes the sequence data and outputs the hidden state for each time step. These hidden states contain the short-term temperature variation characteristics and temperature evolution patterns. These hidden states serve as input to the second-layer LSTM network, which further establishes long-term temperature dependencies and captures patterns in temperature variation over longer timescales, such as hourly temperature trends. This two-layer structure allows the network to simultaneously consider both short-term temperature fluctuations and long-term temperature trends, improving prediction accuracy. During training, mean squared error (MSE) is used as the loss function, and the Adam optimizer is used for parameter updates. The learning rate is set to 0.001, and the training epochs are 200. After training, the network can output future temperature trends for hotspots, such as the predicted temperature for each measurement point within the next 30 minutes.

[0056] Calculating the spatial temperature gradient based on the temperature change trend is a key step in generating the temperature gradient matrix. The spatial temperature gradient describes the rate and direction of temperature change in space and is crucial for understanding heat flow and hotspot formation mechanisms. In practical implementation, the temperature gradient can be calculated for each spatial point based on the predicted temperature distribution. The magnitude of the temperature gradient indicates the severity of the temperature change, and the direction points to the direction of the fastest temperature increase. For example, on a two-dimensional plane, if the temperature at a point (x, y) is T(x, y), the temperature gradient in the x direction at that point can be calculated by dividing T(x + Δx, y) - T(x, y) by Δx. The y direction is calculated similarly. By combining the gradients in the x and y directions, the temperature gradient vector at that point can be obtained. For three-dimensional space, the gradient in the z direction also needs to be considered. During the calculation, Δx, Δy, and Δz can be set to the actual distance between adjacent measurement points, such as 5 mm. In addition, it is necessary to extract information on the distribution of gradient singular points. Singular points are locations where the temperature gradient changes significantly, such as points where the gradient amplitude approaches zero or where the direction suddenly changes. These singular points often correspond to heat sources or heat flow obstacles and are important for understanding the thermal characteristics of conductive devices. By analyzing the temperature gradient amplitude, direction, and distribution of singular points throughout the entire spatial domain, a temperature gradient matrix can be constructed. This matrix is ​​a multidimensional data structure, with each spatial point corresponding to an element whose value contains the gradient amplitude, direction, and whether it is a singular point. For example, for a 20×20×5 three-dimensional grid, the dimensions of the temperature gradient matrix are 20×20×5×4, where the first three dimensions represent spatial coordinates, and the fourth dimension contains the x-, y-, and z-components of the gradient and the singular point identifier.

[0057] The application of temperature gradient matrices is of great value in the coordinated optimization of the high-frequency electromagnetic properties and thermal stability of conductor devices. By analyzing the temperature gradient matrix, hotspots within a conductor device and their evolutionary trends can be identified, predicting potential thermal failure risks. For example, if the temperature gradient amplitude in a certain area continues to increase, it indicates that this area may become a new hotspot; if the direction of the temperature gradient in a hotspot points toward a critical component, that component may be at risk of overheating. Based on these analysis results, the structural design of the conductor device can be optimized, such as adjusting the conductor layout, adding heat dissipation channels, or changing the material combination to improve heat distribution. Furthermore, the temperature gradient matrix can be combined with electromagnetic field analysis to study the impact of temperature changes on the high-frequency electromagnetic properties of the conductor device, such as impedance changes caused by thermal expansion and increased losses due to temperature rise. By synergistically considering thermal and electromagnetic properties, the overall performance of the conductor device can be optimized, improving its reliability and efficiency in high-power, high-frequency applications.

[0058] In this embodiment, the temperature evolution process of the hotspot area is deeply modeled by a double-layer recursive neural network, which not only improves the response capability to short-term temperature fluctuations, but also effectively captures the dependency of long-term temperature changes, and realizes accurate prediction of hotspot temperature trends. By constructing a multi-dimensional feature vector that integrates temperature values, change rates, and spatial features, the model's ability to analyze complex heat conduction dynamics is enhanced. The temperature gradient matrix further generated accurately reflects the intensity direction and abnormal distribution of the spatial thermal field, which helps to identify abnormal thermal diffusion paths and potential thermal runaway risk points, and provides a key basis for subsequent reconstruction of heat conduction structures and adjustment of material parameters, significantly improving the active regulation capability of device thermal design.

[0059] In an optional embodiment, the temperature gradient matrix is ​​subjected to eigendecomposition to obtain heat conduction characteristics. Based on the heat conduction characteristics, the device regions are divided and heat conduction channels between the regions are established. The heat conduction channels are associated with the temperature gradient to construct a material parameter mapping relationship. The final dielectric constant distribution and conductivity distribution are generated through parameter sensitivity analysis and boundary continuity optimization, including: performing eigenvalue decomposition and eigenvector decomposition on the temperature gradient matrix to obtain heat conduction direction characteristic data, generating a gradient intensity distribution map based on the heat conduction direction characteristic data, dividing the device into regions according to the gradient intensity distribution map to obtain material parameter priority region data, and constructing heat conduction channels between adjacent regions based on the material parameter priority region data; Performing an associative operation on the heat conduction channel and the temperature gradient matrix to establish a mapping relationship between the temperature gradient and the dielectric constant and a mapping relationship between the temperature and the conductivity, and calculating the adaptive weight of the material parameters according to the mapping relationship; Calculating the parameter sensitivity of each region using the adaptive weight, comparing the parameter sensitivity with a preset threshold to obtain a parameter update interval, and generating local parameter adjustment data according to the parameter update interval; Performing boundary continuity analysis on the local parameter adjustment data, calculating parameter gradients and boundary features of adjacent regions, constructing boundary feature vectors, classifying boundary types based on the boundary feature vectors, calculating regional comprehensive weights based on the obtained boundary types, performing layered smoothing on the parameters, and generating a parameter smoothing transition scheme; Physical constraints of materials are imposed on the parameter smooth transition scheme, a mapping relationship between the temperature field and the material parameters is established, and the dielectric constant distribution and the conductivity distribution of each region of the device are calculated.

[0060] During implementation, the device's temperature gradient matrix is ​​first obtained. This matrix reflects the rate and direction of temperature change at each point on the device. For a semiconductor device with a 64×64 grid, temperature data at each grid point is collected using a thermal imager, and the temperature gradient matrix G is calculated. Eigenvalue decomposition of this temperature gradient matrix yields the principal eigenvalues ​​λ1=8.76, λ2=3.42, and λ3=0.88, with corresponding eigenvectors v1=[0.72, 0.65, 0.24], v2=[-0.54, 0.62, 0.57], and v3=[0.43, -0.44, 0.79]. These eigenvectors represent the primary direction of temperature conduction, while the eigenvalues ​​represent the conduction intensity in that direction. A gradient intensity distribution map is generated based on the eigenvalues ​​and eigenvectors. The gradient intensity is calculated as the inner product of the eigenvectors and the actual temperature gradient. In the gradient strength distribution diagram, high-intensity areas (such as strength values ​​greater than 7.5) are marked as high-priority areas, medium-intensity areas (intensity values ​​between 3.0 and 7.5) are marked as medium-priority areas, and low-intensity areas (intensity values ​​less than 3.0) are marked as low-priority areas, thereby obtaining material parameter priority area data.

[0061] Based on the priority area data, the heat conduction channel between adjacent areas is constructed. For each pair of adjacent areas A and B, the heat flux between them is calculated. The flux value is the product of the area boundary temperature difference and the boundary area. For example, the boundary temperature difference between area 1 and area 2 is 4.2℃ and the boundary area is 2.1mm. 2 The calculated heat flux is 8.82W / mm 2 Similarly, the heat flux between all adjacent pairs of regions is calculated to form a heat conduction channel network. In this network, the heat flux is greater than 5W / mm 2 The channel is defined as a strong heat conduction channel with a heat flux of 2W / mm 2 Up to 5W / mm 2 The medium heat conduction path is defined as less than 2W / mm 2 is defined as a weak heat conduction channel.

[0062] The heat conduction channel is associated with the temperature gradient matrix to establish the mapping relationship between the temperature gradient and the dielectric constant, and the mapping relationship between temperature and conductivity. Specifically, the mapping relationship between the dielectric constant ε and the temperature gradient amplitude |∇T| is ε=ε0+α|∇T|, where ε0 is the basic dielectric constant, with a value of 4.5, and α is the adjustment coefficient, with a value of 0.8; the mapping relationship between the conductivity σ and the temperature T is σ=σ0exp(βT), where σ0 is the basic conductivity, with a value of 0.02S / m, and β is the temperature coefficient, with a value of 0.03K. -1The adaptive weights of material parameters are calculated using these mapping relationships. For example, in an area with a temperature gradient of 5 K / mm, the dielectric constant weight is 4.5 + 0.8 × 5 = 8.5; in an area with a temperature of 85°C, the conductivity weight is 0.02 × exp(0.03 × 85) = 0.14 S / m.

[0063] Adaptive weights are used to calculate the parameter sensitivity S for each region. Sensitivity is defined as the degree to which a parameter change affects the temperature field. In practice, a small perturbation δp is applied to each region (e.g., a 1% change in the dielectric constant) and the temperature change δT is observed. Sensitivity S = |δT / δp|. For example, if the dielectric constant of region 3 changes from 5.0 to 5.05 (a 1% increase), the average temperature decreases by 0.35°C. Therefore, the sensitivity of this region to the dielectric constant is 0.35 / 0.05 = 7.0. The parameter sensitivity of each region is compared with a preset threshold (e.g., the sensitivity threshold is set to 5.0) to determine the parameter update range. In this example, the sensitivity of region 3 exceeds the threshold, so the dielectric constant update range is determined to be [4.8, 5.3], generating local parameter adjustment data.

[0064] Perform boundary continuity analysis on the local parameter adjustment data to calculate the parameter gradient and boundary characteristics of adjacent regions. For example, the dielectric constants of regions 4 and 5 are 6.2 and 4.8 respectively, and their boundary length is 1.5 mm. The parameter gradient is (6.2-4.8) / 1.5=0.93 mm. -1 Construct boundary feature vectors, including parameter gradient, boundary length, heat flux and other information. Classify boundary types based on boundary feature vectors, and parameter gradient is greater than 0.8mm. -1 The "steep change boundary" has a parameter gradient of 0.3mm. -1 to 0.8mm -1 The boundary between them is a "slowly changing boundary" with a parameter gradient of less than 0.3mm. -1 The area of ​​region 4 is 2.4 mm. 2 , the parameter sensitivity is 6.5, forming an "abrupt boundary" with area 5, and the calculated comprehensive weight W4=2.4×6.5×1.2=18.72 (where 1.2 is the weight coefficient of the abrupt boundary).

[0065] The parameters are smoothed in layers to generate a parameter smoothing transition scheme. Using the weighted average method, let the smoothed parameter p' of region i be i =p i ×(1-γ)+Σ(p j ×w j )×γ / Σw j , where p iis the original parameter value, p j is the parameter value of the adjacent region j, w j is the weight of region j, and γ is the smoothing factor, which is set to 0.3. For example, if the original dielectric constant of region 4 is 6.2, and the dielectric constants of its adjacent regions 5 and 6 are 4.8 and 5.5, respectively, with weights of 18.72 and 12.45, then the smoothed dielectric constant of region 4 is 6.2 × (1 - 0.3) + (4.8 × 18.72 + 5.5 × 12.45) / (18.72 + 12.45) × 0.3 = 5.83.

[0066] Finally, material physical constraints were imposed on the parameter smoothing scheme, such as limiting the dielectric constant to the range of [3.5, 8.0] and the conductivity to the range of [0.01 S / m, 0.2 S / m]. A mapping relationship between the temperature field and the material parameters was established, and the final dielectric constant and conductivity distributions for each device region were obtained through iterative calculations. In this example, the resulting dielectric constant distribution for each region of the semiconductor device ranged from 3.8 to 7.6, and the conductivity distribution ranged from 0.015 S / m to 0.18 S / m, achieving optimal configuration of the material parameters and meeting the design requirements for the device's thermal conductivity performance.

[0067] Figure 3 This is a comparison chart of the temperature gradient analysis and material parameter optimization effects of the embodiment of the present invention. Figure 3 The figure shows a visual comparison of the performance of the three methods for temperature gradient analysis and material parameter optimization. The adaptive weight method performs best in all five evaluation metrics, especially in computational convergence speed and thermal conductivity efficiency, achieving high efficiencies of 92.1% and 94.2%, respectively. The gradient feature method ranks in the middle, significantly outperforming the traditional method but slightly inferior to the adaptive weight method. The traditional method performs poorly in various metrics, especially with a material parameter sensitivity of only 58.9%. This demonstrates that the new method based on eigenvalue decomposition and adaptive weights can effectively improve the accuracy and efficiency of temperature field analysis and material parameter optimization, providing more reliable support for device design.

[0068] In the prior art, the material parameter distribution for semiconductor devices mostly relies on static thermal analysis or preset models, lacking the ability to dynamically respond to real-time temperature changes and heat conduction behavior, making it difficult to achieve coordinated optimization of the device's thermal field and electromagnetic performance. This application achieves precise division of device areas by characteristic decomposition of the temperature gradient matrix, extracting heat conduction direction information and gradient intensity distribution, and constructing a physically guided heat conduction channel, thereby enhancing the ability to respond to local thermal behavior. On this basis, by associating the heat conduction channel with the temperature gradient data, a dynamic mapping relationship between the dielectric constant and the temperature gradient, and between the conductivity and the temperature is established, effectively breaking through the limitations of traditional static parameter settings. Furthermore, based on the parameter sensitivity calculated based on adaptive weights and combined with the continuity analysis of local boundaries, a smooth transition strategy for parameter distribution is proposed to ensure the physical consistency of the material parameters of each region at the boundary. Compared with the thermal field discontinuity and electrical performance drift problems caused by the use of a unified material model or coarse-grained parameter adjustment method in the existing technology, this application achieves a spatially differentiated distribution of dielectric constant and conductivity while maintaining the clarity of the heat conduction path, significantly improving the thermal stability and electromagnetic consistency of the device under high-frequency conditions, and meeting the needs of high-performance semiconductor devices operating in complex environments.

[0069] In an optional embodiment, performing boundary continuity analysis on the local parameter adjustment data, calculating parameter gradients and boundary features of adjacent regions, constructing boundary feature vectors, classifying boundary types based on the boundary feature vectors, calculating regional comprehensive weights based on the obtained boundary types, performing layered smoothing on the parameters, and generating a parameter smoothing transition scheme include: Using a multi-directional differential template to analyze local parameter adjustment data, obtain parameter gradient information, calculate the normal gradient, tangential gradient and boundary curvature based on the parameter gradient information, and extract boundary features; Normalizing the boundary features, and combining the normalized parameter gradient amplitude, gradient direction, boundary curvature, and parameter jump variable to construct a boundary feature vector; Establishing a boundary classification matrix based on the boundary feature vector, determining a first boundary classification according to the parameter gradient amplitude, determining a second boundary classification in combination with the gradient direction continuity, and determining a third boundary classification by the boundary curvature feature; Constructing a weight association matrix based on the boundary classification, inputting boundary classification weights, regional area weights, parameter importance weights, and location correlation weights into the weight association matrix to calculate the regional comprehensive weight; A smoothing coefficient is selected based on the comprehensive weight of the area, a first smoothing coefficient is used for the first boundary classification area, a second smoothing coefficient is used for the second boundary classification area, and a third smoothing coefficient is used for the third boundary classification area. The boundary continuity of the smoothing result is evaluated, and when the boundary continuity meets the preset conditions, a parameter smoothing transition scheme is output.

[0070] The detailed implementation method of the parameter smooth transition scheme can be achieved by performing boundary continuity analysis on the local parameter adjustment data and classifying it based on the boundary feature vector, thereby realizing hierarchical smoothing of the parameters.

[0071] For example, using a multi-directional differential template to analyze local parameter adjustment data is the first step to achieve a smooth parameter transition. In the high-frequency electromagnetic and thermal stability optimization design of semiconductor devices, local parameter adjustment data usually includes the adjustment values ​​of key parameters such as the resistivity, thermal conductivity, and dielectric constant of semiconductor materials in different regions. The discontinuous distribution of these parameters in space will cause distortion of the electromagnetic and thermal fields, affecting device performance. In order to obtain parameter gradient information, a 3×3 or 5×5 multi-directional differential template can be used to scan and analyze the parameter distribution. Taking a 5×5 template as an example, on the two-dimensional plane of the semiconductor device, the center of the template is aligned with each node of the parameter grid, and the parameter differential values ​​in the horizontal, vertical, and diagonal directions are calculated respectively. For example, for the thermal conductivity parameters of semiconductor materials, a 5×5 grid is set in a certain area, and the size of each grid unit is 0.5 mm×0.5 mm. The thermal conductivity values ​​of adjacent grid points may suddenly change from 120W / (m·K) to 80W / (m·K). The differential template can be used to calculate that the parameter gradient here is 80W / (m·K) / 0.5 mm. Based on the parameter gradient information, the normal gradient, tangential gradient and boundary curvature can be further calculated. The normal gradient refers to the gradient component perpendicular to the parameter boundary, which indicates the degree of change of the parameter at the boundary; the tangential gradient refers to the gradient component along the parameter boundary, which indicates the change of the parameter in the boundary direction; the boundary curvature describes the degree of curvature of the parameter boundary, which can be obtained by calculating the rate of change of the tangent direction of adjacent points on the boundary. These features together constitute the basic description of the boundary characteristics.

[0072] Normalizing boundary features is a key step in constructing boundary feature vectors. Normalization ensures that different types of features are within the same scale range, which facilitates subsequent analysis and classification. The normalization of parameter gradient amplitudes can use the maximum-minimum normalization method to map the original gradient amplitude to a range of 0 to 1. For example, if the thermal conductivity gradient amplitude detected on a semiconductor device ranges from 0 to 100 W / (m·K) / mm, the gradient amplitude at each position can be normalized by dividing it by 100. The gradient direction is usually expressed as an angle, which can be normalized to a range of -1 to 1, indicating the degree of deviation from the reference direction. The boundary curvature also needs to be normalized, which can be achieved by dividing by the maximum curvature value. The parameter jump variable represents the difference in the parameter on both sides of the boundary, such as the resistivity from 1.68×10 -8 Ω·m jumps to 2.44×10 -8 Ω·m, the jump amount is 0.76×10 -8 Ω·m, when normalized, can be divided by the maximum jump value detected within the device. After normalization, the parameter gradient amplitude, gradient direction, boundary curvature, and parameter jump value are combined into a four-dimensional vector, the boundary feature vector. Each boundary point corresponds to a feature vector. These vectors form a mathematical representation of the boundary characteristics and provide the basis for subsequent boundary classification.

[0073] Establishing a boundary classification matrix based on boundary eigenvectors is the core of boundary classification. The boundary classification matrix is ​​a data structure used to classify boundary types and contains multi-level classification results. The first boundary classification is determined based on the parameter gradient magnitude. The gradient magnitude can be divided into multiple intervals corresponding to different boundary levels. For example, in electromagnetic-thermal coupling analysis of semiconductor devices, the normalized conductivity gradient magnitude can be divided into three intervals: 0-0.3 for low-gradient boundaries, 0.3-0.7 for medium-gradient boundaries, and 0.7-1.0 for high-gradient boundaries. The second boundary classification is determined based on gradient directional continuity. Directional continuity is determined by calculating the rate of change of the gradient direction between adjacent boundary points. If the gradient direction change between adjacent points is less than 15 degrees, the direction is considered highly continuous; if the change is between 15-45 degrees, the direction is considered moderately continuous; and if the change is greater than 45 degrees, the direction is considered lowly continuous. The third boundary classification is determined based on boundary curvature characteristics. Normalized curvature values ​​can be divided into three categories: low curvature (0-0.3), medium curvature (0.3-0.7), and high curvature (0.7-1.0). By combining the results of these three levels of classification, a complete boundary grading matrix can be established. For example, a boundary line on a semiconductor device may be classified as a high-gradient boundary at the first level, as medium-directional continuity at the second level, and as a low-curvature boundary at the third level, and finally marked as a boundary type that requires special treatment.

[0074] Constructing a weighted association matrix based on boundary classification is the basis for calculating the regional comprehensive weight. The weighted association matrix is ​​a data structure used to integrate multiple weighting factors, mapping different types of weights into regional comprehensive weights. Boundary classification weights are weighting factors determined based on the boundary classification results. For example, high-gradient boundaries can be assigned a weight of 0.8, medium-gradient boundaries a weight of 0.5, and low-gradient boundaries a weight of 0.2. Regional area weights reflect the impact of regional size on parameter smoothing and can be calculated by dividing the regional area by the total device area. For example, if a semiconductor region has an area of ​​100 square millimeters and the total device area is 1000 square millimeters, the area weight of this region is 0.1. Parameter importance weights indicate the degree of influence of a parameter on device performance. For example, in high-frequency semiconductor design, electrical conductivity may have a more significant impact on electromagnetic properties than thermal conductivity. Therefore, electrical conductivity may be assigned an importance weight of 0.7, and thermal conductivity an importance weight of 0.4. Positional relevance weights consider the importance of a region's location within the device. For example, regions near heat sources or critical signal paths may be assigned a higher location weight. These four weights are input into the weighted association matrix, and the regional comprehensive weight is calculated using weighted averaging or other combined methods. For example, if the boundary classification weight of a region is 0.8, the area weight is 0.3, the parameter importance weight is 0.6, and the location correlation weight is 0.7, then its comprehensive weight may be 0.65, indicating that the region has a higher priority in the parameter smoothing process.

[0075] The key to achieving hierarchical smoothing is to select a smoothing coefficient based on the regional comprehensive weight. Using different smoothing coefficients for different boundary classification areas can achieve a smooth transition of parameters in space. For the first boundary classification area (high gradient boundary), the first smoothing coefficient is used. For example, setting the smoothing coefficient to 0.2 means retaining 80% of the original parameter value and replacing 20% ​​with the weighted average of the parameters of the adjacent areas. In practical applications, for resistivity at the boundary from 1.68×10 -8 Ω·m jumps to 2.44×10 -8 For the case of Ω·m, a gradual transition can be implemented within five grid points near the boundary to avoid abrupt changes. For the second boundary classification region (medium-gradient boundary), a second smoothing coefficient, such as 0.5, is used, indicating that the original parameter value and the parameters of the adjacent region each account for 50%. For the third boundary classification region (low-gradient boundary), a third smoothing coefficient, such as 0.8, is used, indicating that the weighted average of the parameters of the adjacent regions is mostly used. After smoothing, the results need to be evaluated for boundary continuity. This can be assessed by calculating the rate of change of the parameter gradient before and after processing. For example, if the maximum parameter gradient after processing is reduced by 80%, the average gradient is reduced by 65%, and the first-order derivative of the parameter distribution is spatially continuous, the boundary continuity is considered to meet the preset conditions, and the parameter smoothing transition solution can be output.

[0076] Parameter smoothing schemes have important applications in the coordinated optimization of high-frequency electromagnetic properties and thermal stability of semiconductor devices. Smoothing can reduce discontinuities in parameter spatial distribution, minimize electromagnetic and thermal field distortion, and improve the accuracy of simulation analysis. For example, in the design of high-frequency power electronic devices, sudden changes in electrical and thermal conductivity at the interface of different materials can lead to localized current density concentration and the formation of hotspots. Parameter smoothing schemes can achieve a gradual transition in material properties, reducing current distortion and thermal stress concentration. Furthermore, parameter smoothing schemes can be used in the design of gradient materials for semiconductor devices, achieving coordinated optimization of electromagnetic properties and thermal stability by continuously adjusting material composition and performance parameters over space. For example, in the design of semiconductor heat dissipation structures for RF power amplifiers, parameter smoothing schemes can be used to generate spatial distributions of electrical and thermal conductivity, ensuring both high-frequency electromagnetic performance and rapid heat dissipation, extending device life. Parameter smoothing schemes are not only applicable to static parameter distribution optimization but can also be used to adjust parameters under dynamic operating conditions, such as by accounting for changes in material properties caused by rising temperature, to achieve optimized design across the entire operating range.

[0077] In the prior art, parameter smoothing often uses a unified coefficient or a simple weighted average method, which makes it difficult to take into account the differences in boundary continuity and regional physical characteristics, and easily leads to parameter jumps at the boundary, inconsistent thermal-electrical performance, and other problems, which limits the refined design and performance optimization of the device. This application constructs a boundary feature vector based on a multi-directional differential template, integrates key features such as parameter gradient, gradient direction, and boundary curvature, and for the first time realizes the fine classification and multi-level hierarchical expression of boundary types. By constructing a boundary grading matrix and combining it with a multi-dimensional weight factor, the regional comprehensive weight is accurately calculated, and then a differentiated smoothing coefficient is used to implement hierarchical processing for different boundary levels, effectively alleviating the local physical distortion caused by parameter mutations. At the same time, by introducing a boundary continuity evaluation mechanism, the smoothing results are dynamically verified and optimized to ensure a smooth transition of parameters and gradient continuity between regions, thereby improving the physical consistency of the overall temperature field and material parameter field. Compared with traditional methods, this solution is guided by the physical characteristics of the boundary for hierarchical regulation, has stronger adaptability and local sensitivity, and significantly enhances the modeling accuracy and engineering applicability of highly complex devices in thermal and electrical coupling scenarios.

[0078] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0079] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0080] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collaboratively optimizing the high-frequency electromagnetic characteristics and thermal stability of semiconductor devices, characterized in that: include: Arrange electromagnetic field detection points at the grid nodes of semiconductor devices, collect the electric field strength and magnetic field strength of the electromagnetic field detection points, calculate the electromagnetic field energy loss of the grid nodes based on the electric field strength and magnetic field strength, use piecewise linear interpolation method to establish a mapping relationship between electromagnetic field energy loss and temperature, and convert the electromagnetic field energy loss into heat source power density based on the mapping relationship; Input the heat source power density into the multi-scale decomposition network to obtain heat distribution characteristic data, determine the location information of the hot spot area based on the heat distribution characteristic data, establish the time series temperature data of the hot spot area, input the time series temperature data into the recursive neural network to obtain the temperature change trend of the hot spot area, and generate a temperature gradient matrix based on the temperature change trend; The temperature gradient matrix is ​​decomposed to obtain heat conduction characteristics. Based on the heat conduction characteristics, the device area is divided and heat conduction channels between the areas are established. The heat conduction channels are associated with the temperature gradient to construct a material parameter mapping relationship. The dielectric constant distribution and electrical conductivity distribution are generated through parameter sensitivity analysis and boundary continuity optimization. The dielectric constant distribution and conductivity distribution are used as initial parameters, and material parameters are configured for each area of ​​the semiconductor device. The new temperature distribution is calculated based on the configuration results, and the new temperature distribution is compared with the preset temperature threshold. When the temperature distribution meets the preset threshold requirements, the current dielectric constant distribution and conductivity distribution are determined as the optimized structural parameters of the semiconductor device.

2. The method according to claim 1, characterized in that Arrange electromagnetic field detection points at the grid nodes of semiconductor devices, collect the electric field strength and magnetic field strength of the electromagnetic field detection points, and calculate the electromagnetic field energy loss of the grid nodes based on the electric field strength and magnetic field strength. Arranging electromagnetic field detection points at the grid nodes, the electromagnetic field detection points including electric field detectors and magnetic field detectors, sampling output signals of the electric field detectors and the magnetic field detectors to obtain electromagnetic field sampling data; Filtering the electromagnetic field sampling data to obtain the electric field intensity and magnetic field intensity of the electromagnetic field detection point; Acquiring temperature data of the semiconductor device, determining material parameters based on the temperature data, calculating a dielectric loss value based on the electric field strength, calculating a magnetic loss value based on the magnetic field strength, and calculating a conduction loss value based on the electric field strength and the material parameters; The dielectric loss value, the magnetic loss value, and the conduction loss value are added together to obtain the electromagnetic field energy loss of the grid node.

3. The method according to claim 1, characterized in that The mapping relationship between electromagnetic field energy loss and temperature is established using the piecewise linear interpolation method. The electromagnetic field energy loss is converted into heat source power density according to the mapping relationship, including: Adaptively filter the electromagnetic field energy loss data, dynamically adjust the filter parameters according to the signal change amplitude, obtain the filtered electromagnetic field energy loss data, calculate the distribution probability of the filtered electromagnetic field energy loss data, determine the inflection point of the probability change as the segmentation point, and divide the range of the electromagnetic field energy loss into multiple sub-intervals; In each subinterval, the correlation coefficient between the electromagnetic field energy loss data and the temperature data is calculated, the correlation coefficient is used as a weight value, and a piecewise linear interpolation method is used to establish a subinterval mapping function for the weighted data; Calculate the slope difference of the mapping function of adjacent subintervals at the segmentation point, adjust the mapping function coefficient according to the slope difference, make the mapping function of adjacent subintervals continuous at the segmentation point, and obtain the piecewise linear mapping relationship between electromagnetic field energy loss and temperature; Real-time electromagnetic field energy loss data of the semiconductor device is obtained, the real-time electromagnetic field energy loss data is substituted into the piecewise linear mapping relationship to obtain a temperature prediction value, a corresponding material parameter correction factor is obtained according to the temperature prediction value, and the real-time electromagnetic field energy loss data is multiplied by the material parameter correction factor to obtain the heat source power density.

4. The method according to claim 1, wherein Input the heat source power density into the multi-scale decomposition network to obtain the heat distribution characteristic data. According to the heat distribution characteristic data, the location information of the hot spot area is determined. The time series temperature data of the hot spot area is established, including: Inputting the heat source power density into a multi-scale decomposition network to perform signal decomposition to obtain a decomposition coefficient matrix group, and dividing the decomposition coefficient matrix group into multiple scale levels; Calculating local statistical features of decomposition coefficient matrices at multiple scale levels to obtain characteristic variance matrices and characteristic mean matrices, constructing a dynamic decision function based on the characteristic variance matrix and characteristic mean matrix, and performing a comparison operation between the decomposition coefficient matrix group and the dynamic decision function to obtain heat distribution characteristic data; The positional relationship of the hotspot area is calculated based on the heat distribution characteristic data, and the positional relationship is analyzed and processed using a density clustering method to obtain the position information of the hotspot area. Temperature data is collected based on the position information of the hotspot area, and the temperature data is sorted in chronological order to establish the time series temperature data of the hotspot area.

5. The method according to claim 1, wherein Input the time series temperature data into the recurrent neural network to obtain the temperature change trend of the hot spot area. The temperature gradient matrix generated according to the temperature change trend includes: Perform feature enhancement processing on time series temperature data, extract temperature value, temperature change, temperature change rate and hot spot area volume, and construct temperature feature vector; Calculating the temperature difference between the hotspot area and the adjacent area based on the temperature feature vector, obtaining the shape characteristics and position information of the hotspot area, and generating a spatial correlation feature; Input the temperature feature vector and spatial correlation feature into a two-layer recursive neural network, extract the short-term temperature change feature and temperature evolution law through the first layer network, input the short-term temperature change feature and temperature evolution law into the second layer network, establish the long-term temperature dependency, and obtain the temperature change trend of the hot spot area; The amplitude and direction of the spatial temperature gradient are calculated according to the temperature change trend, the gradient singular point distribution information is extracted, and the amplitude and direction of the spatial temperature gradient and the singular point distribution information are combined to generate a temperature gradient matrix.

6. The method according to claim 1, wherein The temperature gradient matrix is ​​decomposed to obtain heat conduction characteristics. Based on the heat conduction characteristics, the device area is divided and heat conduction channels between the areas are established. The heat conduction channels are associated with the temperature gradient to construct a material parameter mapping relationship. The final dielectric constant distribution and conductivity distribution are generated through parameter sensitivity analysis and boundary continuity optimization, including: performing eigenvalue decomposition and eigenvector decomposition on the temperature gradient matrix to obtain heat conduction direction characteristic data, generating a gradient intensity distribution map based on the heat conduction direction characteristic data, dividing the device into regions according to the gradient intensity distribution map to obtain material parameter priority region data, and constructing heat conduction channels between adjacent regions based on the material parameter priority region data; Performing an associative operation on the heat conduction channel and the temperature gradient matrix to establish a mapping relationship between the temperature gradient and the dielectric constant and a mapping relationship between the temperature and the conductivity, and calculating the adaptive weight of the material parameters according to the mapping relationship; Calculating the parameter sensitivity of each region using the adaptive weight, comparing the parameter sensitivity with a preset threshold to obtain a parameter update interval, and generating local parameter adjustment data according to the parameter update interval; Performing boundary continuity analysis on the local parameter adjustment data, calculating parameter gradients and boundary features of adjacent regions, constructing boundary feature vectors, classifying boundary types based on the boundary feature vectors, calculating regional comprehensive weights based on the obtained boundary types, performing layered smoothing on the parameters, and generating a parameter smoothing transition scheme; Physical constraints of materials are imposed on the parameter smooth transition scheme, a mapping relationship between the temperature field and the material parameters is established, and the dielectric constant distribution and the conductivity distribution of each region of the device are calculated.

7. The method according to claim 6, characterized in that Performing boundary continuity analysis on the local parameter adjustment data, calculating the parameter gradient and boundary characteristics of adjacent regions, constructing boundary feature vectors, classifying boundary types based on the boundary feature vectors, calculating regional comprehensive weights based on the obtained boundary types, performing layered smoothing on the parameters, and generating a parameter smoothing transition scheme include: Using a multi-directional differential template to analyze local parameter adjustment data, obtain parameter gradient information, calculate the normal gradient, tangential gradient and boundary curvature based on the parameter gradient information, and extract boundary features; Normalizing the boundary features, and combining the normalized parameter gradient amplitude, gradient direction, boundary curvature, and parameter jump variable to construct a boundary feature vector; Establishing a boundary classification matrix based on the boundary feature vector, determining a first boundary classification according to the parameter gradient amplitude, determining a second boundary classification in combination with the gradient direction continuity, and determining a third boundary classification by the boundary curvature feature; Constructing a weight association matrix based on the boundary classification, inputting boundary classification weights, regional area weights, parameter importance weights, and location correlation weights into the weight association matrix to calculate the regional comprehensive weight; A smoothing coefficient is selected based on the comprehensive weight of the area, a first smoothing coefficient is used for the first boundary classification area, a second smoothing coefficient is used for the second boundary classification area, and a third smoothing coefficient is used for the third boundary classification area. The boundary continuity of the smoothing result is evaluated, and when the boundary continuity meets the preset conditions, a parameter smoothing transition scheme is output.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images