A method for analyzing temperature distribution of a radio frequency multi-finger active device

Abstract

The application relates to a radio frequency multi-finger active device temperature distribution analysis method, which comprises the following steps: using a genetic algorithm and a neural network to train a single-finger device temperature distribution model by using temperature distribution data of the radio frequency single-finger device along a center point of a heat source to an outer peripheral direction; given device information such as a multi-finger device index, a finger spacing, an environment temperature and power consumption, the coordinate system, the boundary position and the coordinate of each heat source center point are determined; a plurality of fitting data points are set; and an improved superposition algorithm is used to sufficiently consider the influence of thermal coupling on the thermal conductivity of the device material, so that the temperature distribution of the radio frequency multi-finger device is obtained. According to the method, the multi-finger device temperature distribution under any index, finger spacing, environment temperature and power consumption can be efficiently and accurately obtained through the single-finger device temperature distribution model, the problems of large computer resource occupation and long analysis time of the finite element analysis method are avoided, and the accuracy of the characterization result is higher.

Description

Technical Field

[0001] This invention belongs to the field of integrated circuit electrothermal analysis technology, specifically relating to a method for analyzing the temperature distribution of radio frequency multi-finger active devices. Background Technology

[0002] With the rapid development of microelectronics manufacturing processes, device sizes are constantly shrinking, and the power density of integrated circuits and systems is increasing exponentially. To provide high power density at high frequencies, transistors with large emitter areas and minimal emitter widths are required. For single-finger emitter devices, the current density on the fingers is mainly concentrated at the emitter edge, resulting in uneven distribution. Meanwhile, III-V type RF devices, heterojunction bipolar transistors (HBTs), have significant self-heating effects and are prone to thermal breakdown. Therefore, a comb-like arrangement of multiple emitter fingers in parallel is typically used to increase the emitter area ratio, thereby improving the device's self-heating effect, frequency response, and noise characteristics. However, this design also has a significant problem: severe thermal coupling exists between the multiple emitter fingers. This means that the thermal resistance of multi-finger devices does not decrease linearly with the increase in the number of fingers; some literature shows that the thermal resistance of an 8-finger device is only 5 times lower than that of a single-finger device.

[0003] To better optimize multi-finger HBT devices, it is essential to establish a model that accurately characterizes the relationship between the device's self-heating effect, the thermal coupling effect between the emitter fingers, and the device's structural parameters. Finite element method (FEM) modeling is the most commonly used method for predicting and analyzing device temperature distribution. While this method offers high accuracy, it is extremely time-consuming and resource-intensive in temperature analysis calculations. Infrared thermal imaging is also frequently used to measure device junction temperature; however, due to the micrometer-level dimensions of semiconductor dies, die testing requires a probe station, placing high demands on thermal imaging equipment, the testing environment, and the resolution of the infrared thermal imager. Even slight deviations can easily lead to large measurement errors. Existing temperature analysis techniques are highly unsuitable for analyzing the temperature distribution of multi-finger devices.

[0004] Neural networks are adaptive nonlinear dynamic systems composed of a large number of simple interconnected neurons, and they are highly advantageous in representing nonlinear relationships. Theoretically, with a sufficient number of hidden layers and neurons, any nonlinear mapping can be approximated. Furthermore, with a large enough training sample size, errors in a few samples do not affect modeling accuracy. Therefore, a well-trained neural network model can accurately describe complex nonlinear mappings between outputs without requiring explicit mathematical expressions. In addition, training directly on large datasets—that is, data-driven neural network methods—offer numerous advantages such as simplicity of implementation, good fitting results, and high robustness, and are therefore widely used in device modeling.

[0005] If the temperature distribution of multi-fingered devices is trained directly based on a neural network model, a large amount of training data will be required because the thermal coupling effect between devices is related to the device index, emitter size and finger spacing. This will also consume a lot of computer time and memory. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention provides a method for analyzing the temperature distribution of radio frequency multi-finger active devices. The technical problem to be solved by this invention is achieved through the following technical solution:

[0007] This invention provides a method for analyzing the temperature distribution of radio frequency multi-finger active devices, including the following steps:

[0008] The coordinate system, boundary positions, and coordinates of the center point of each heat source are determined based on the structural parameters of the RF multi-finger device.

[0009] Several fitting data points are set according to the accuracy of temperature distribution;

[0010] Combining the coordinates of each heat source center point, the first coupled temperature rise from all heat source center points to the first fitted data point is calculated using the single-finger device temperature distribution model under the first ambient temperature. The first ambient temperature and the first coupled temperature rise are then superimposed to calculate the heat distribution value of the first fitted data point. The single-finger device temperature distribution model is obtained by training a neural network on the temperature distribution data of the radio frequency single-finger device along the direction from the heat source center point to the periphery.

[0011] Using the heat distribution value as the second ambient temperature of the second fitted data point, the second coupled temperature rise from all heat source center points to the second fitted data point is calculated using the single-finger device temperature distribution model, and the heat distribution value and the second coupled temperature rise are superimposed to calculate the temperature value of the second fitted data point; wherein, along the boundary position toward the heat source center point, the first fitted data point is the previous neighboring point of the second fitted data point.

[0012] Calculate the temperature value for each fitted data point to obtain the temperature distribution of the RF multi-finger device.

[0013] In one embodiment of the present invention, the temperature distribution model of the single-finger device is obtained by training a neural network on the temperature distribution data of the radio frequency single-finger device along the direction from the center point of the heat source to the periphery, including:

[0014] A solid model of the electrothermal coupling of the RF single-finger device was constructed using finite element analysis software.

[0015] The electrothermal coupling solid model is subjected to power dissipation and boundary conditions, and then finite element meshing is performed.

[0016] Steady-state thermal analysis was performed on the electrothermal coupling solid model under different dissipation powers, different emitter lengths, and different ambient temperatures to obtain temperature distribution data along the direction from the center point of the heat source to the periphery.

[0017] The dissipated power, emitter length, ambient temperature, and distance from the center point of the heat source to the periphery in the temperature distribution data are used as input vectors. The temperature data in the temperature distribution data is used as the expected value. The weights and thresholds of the neural network are optimized using a genetic algorithm. The optimal weights and thresholds are then assigned to the neural network to obtain the temperature distribution model of the single-finger device.

[0018] In one embodiment of the present invention, the neural network includes a BP neural network;

[0019] The BP neural network includes an input layer, several hidden layers, and an output layer connected in sequence.

[0020] In one embodiment of the present invention, the plurality of hidden layers includes a first hidden layer and a second hidden layer, wherein,

[0021] The input layer, the first hidden layer, the second hidden layer, and the output layer are connected in sequence;

[0022] The first hidden layer comprises 6 neurons, and the second hidden layer comprises 3 neurons.

[0023] In one embodiment of the present invention, the transfer function of the hidden layer neurons in the BP neural network includes the tansig function, and the output layer function includes the purelin function.

[0024] In one embodiment of the present invention, determining the coordinate system, boundary position, and coordinates of the center point of each heat source based on the structural parameters of the radio frequency multi-finger device includes:

[0025] Obtain the index, feature size, and finger spacing of the radio frequency multi-finger device;

[0026] The center point of each finger of the radio frequency multi-finger device is taken as the center point of the heat source, and the origin and boundary position of the radio frequency multi-finger device are determined according to the index, the feature size and the finger spacing.

[0027] Based on the origin and the boundary position, the coordinates of the center point of each heat source are calculated according to the index, the feature size, and the finger spacing.

[0028] In one embodiment of the present invention, when the radio frequency multi-finger device is a uniformly distributed multi-finger device, the coordinates of the center point of each heat source are:

[0029] x center,k = k*x + (k-1)*We +1 / 2*W e (1≤k≤n, k∈Z)

[0030] Where, x center,k Let W be the coordinates of the center point of the k-th heat source, n be the exponent, x be the distance between exponents, and W be the distance between exponents. e This refers to the feature size.

[0031] In one embodiment of the present invention, the heat distribution value of the first fitted data point is:

[0032]

[0033] in, T represents the heat distribution value of the first fitted data point. a The first ambient temperature, Let n be the first coupled temperature rise of the first fitted data point, n be the exponent, k be the kth heat source, and ΔT be the temperature rise of the first coupled data point. k For the k-th heat source pair x i Thermal coupling temperature rise at the point, P diss For power dissipation, L e Let x be the emitter length, and |Δx| be x. i The distance A to the center point of the kth heat source ANN (T a ,P diss ,L e ,|Δx|) is a single-finger device temperature distribution model based on the first ambient temperature.

[0034] In one embodiment of the present invention, the temperature value of the second fitted data point is:

[0035]

[0036] in, The temperature value is the second fitted data point. The second coupled temperature rise for the second fitted data point. Let be the heat distribution value of the first fitted data point, n be the exponent, k be the kth heat source, and ΔT be the heat distribution value of the first fitted data point. k For the k-th heat source pair x i Thermal coupling temperature rise at the point, P diss For power dissipation, L e Let |Δx| be the emitter length, and |Δx| be the distance from the center of the heat source to the periphery. This is a temperature distribution model for a single-finger device based on thermal distribution values.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] The present invention provides a method for analyzing the temperature distribution of multi-finger active devices in radio frequency. First, it uses a single-finger device temperature distribution model to calculate the thermal distribution value of a first fitted data point at ambient temperature. Then, it uses the thermal distribution value of the first fitted point as the second ambient temperature for a second fitted data point. The single-finger device temperature distribution model is then used to calculate the temperature value of the second fitted data point, thus obtaining the temperature value for each fitted data point. By combining the single-finger device temperature distribution model with an improved superposition algorithm, the influence of thermal coupling on the thermal conductivity of the device material is fully considered. Only a neural network model training is needed for the single-finger device temperature distribution to obtain the device temperature distribution under arbitrary exponents, finger spacing, ambient temperature, and power consumption. While ensuring and improving analysis accuracy, this method greatly improves analysis efficiency and is applicable to the temperature analysis of multi-finger devices at any process node, avoiding the problems of high computer resource consumption and time-consuming analysis associated with finite element analysis methods. Attached Figure Description

[0039] Figure 1 A schematic flowchart illustrating the method for analyzing the temperature distribution of radio frequency multi-finger active devices provided in an embodiment of the present invention;

[0040] Figure 2 A flowchart illustrating the method for analyzing the temperature distribution of radio frequency multi-finger active devices provided in this embodiment of the invention;

[0041] Figure 3 This is a schematic diagram illustrating the coordinate system determination and heat source center point confirmation provided in an embodiment of the present invention.

[0042] Figures 4a-4c This is a schematic diagram of the temperature distribution and error distribution of a radio frequency multi-finger active device provided in an embodiment of the present invention. Detailed Implementation

[0043] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0044] Example 1

[0045] Please see Figure 1 and Figure 2 , Figure 1 This is a flowchart illustrating the method for analyzing the temperature distribution of radio frequency multi-finger active devices provided in an embodiment of the present invention. Figure 2 The flowchart illustrates the method for analyzing the temperature distribution of radio frequency multi-finger active devices provided in this embodiment of the invention.

[0046] Before analyzing the temperature distribution of multi-finger active RF devices, a temperature distribution model for single-finger devices is first obtained. This model is trained using a neural network on the temperature distribution data of the RF single-finger device along the direction from the center of the heat source outwards. The training method includes the following steps:

[0047] S11. Construct an electrothermal coupling solid model of the RF single-finger device using finite element analysis software.

[0048] First, based on the device fabrication information, the structural and physical parameters of the RF single-finger device are obtained. These RF single-finger devices include, but are not limited to, HBT devices. Structural parameters include the dimensions of the active region, substrate, and each layer of the active region, as well as the substrate dimensions. Physical parameters include device material properties and doping concentrations, with material properties referring to the material name, composition, and corresponding physical characteristics of each layer. Then, using finite element analysis software, an electrothermal coupling solid model is established based on the structural and physical parameters of the RF single-finger device.

[0049] S12. Load the dissipated power and boundary conditions onto the electrothermal coupled solid model, and perform finite element mesh generation.

[0050] Specifically, the power dissipation is the heat source. Boundary conditions can be set such that the back side of the substrate is at ambient temperature, and other chip surfaces in contact with the outside are thermally insulated. The electrothermal coupled solid model is meshed using the free-partition tetrahedron option.

[0051] S13. Perform steady-state thermal analysis on the electrothermal coupling solid model under different dissipation powers, different emitter lengths and different ambient temperatures to obtain temperature distribution data along the direction from the center point of the heat source to the periphery.

[0052] Specifically, steady-state thermal analysis was performed by changing the dissipation power, emitter length, and ambient temperature of the electrothermal coupled solid model. This yielded temperature distribution data of RF single-finger devices of different sizes at different dissipation power and ambient temperatures, along the direction from the center point of the heat source to the periphery. The data was then exported as a CSV file to generate the dataset required for neural network training.

[0053] Furthermore, a boundary equal-interval sampling method was adopted to sample 75% of the data points in the dataset as training data and import them into the neural network model for training and learning, while the remaining 25% of the data points were used as validation data to verify the accuracy of the constructed model.

[0054] S14. Using the dissipated power, emitter length, ambient temperature, and distance from the center point of the heat source to the periphery in the temperature distribution data as input vectors, and the temperature data in the temperature distribution data as expected values, a genetic algorithm is used to optimize the weights and thresholds of the neural network, and the optimal weights and thresholds are assigned to the neural network to obtain the temperature distribution model of the single-finger device.

[0055] First, determine the structure of the neural network.

[0056] Specifically, the neural network used is a backpropagation (BP) neural network. A BP neural network consists of an input layer, several hidden layers, and an output layer connected in sequence. The number of hidden layers and the number of neurons in each hidden layer can be selected according to actual needs. For example, a BP neural network includes an input layer, a first hidden layer, a second hidden layer, and an output layer connected in sequence, where the first hidden layer includes 6 neurons and the second hidden layer includes 3 neurons. The transfer function of the hidden layer neurons includes the tansig function, and the output layer function includes the purelin function.

[0057] The tansig function is used as the transfer function, which can effectively normalize the input data to another space, making the data easier to process. The Purelin function has the characteristics of linearity, unboundedness, and differentiability, and can map inputs of arbitrary size to outputs of arbitrary size. Therefore, the Purelin function is used in the output layer.

[0058] Then, using Matlab software, the BP neural network was trained and learned using the training data based on the genetic algorithm (GA algorithm) to obtain the temperature distribution model of the single-finger device. This temperature distribution model of the single-finger device is denoted as the GA-BP neural network.

[0059] Specifically, the power dissipation P in the temperature distribution data diss Emitter length L e Ambient temperature T a The distance |Δx| from the center point of the heat source to the periphery corresponds one-to-one with the temperature data, and the dissipation power P is included. diss Emitter length L e Ambient temperature T a The distance |Δx| from the center point of the heat source outwards is used as the input vector of the neural network. The temperature data within this vector is used as the expected value of the neural network. A genetic algorithm is used to optimize the weights and thresholds of the neural network. The optimal weights and thresholds obtained by the genetic algorithm are then assigned to the BP neural network. After training, the neural network is used to predict the output, thus obtaining a single-finger device temperature distribution model based on a neural network under a given process, denoted as A. ANN (T a ,P diss ,L e The model defines a single-finger device temperature distribution model, where |Δx| = 0, representing the location of the heat source center. This model accurately characterizes the temperature distribution of the single-finger device across the entire range emanating from the heat source center. For example, when optimizing using a genetic algorithm, the population size is 100, the number of winning subpopulations is 5, and the number of iterations is 20. The optimization is performed on a PC platform with an i7 9700 CPU and 8GB RAM, running in MATLAB.

[0060] Based on the temperature distribution model of a single-finger device, a temperature distribution analysis of a radio frequency multi-finger active device is performed. This method for analyzing the temperature distribution of radio frequency multi-finger active devices includes the following steps:

[0061] S21. Determine the coordinate system, boundary positions, and coordinates of the center point of each heat source based on the structural parameters of the RF multi-finger device. Specific steps include:

[0062] S211. Obtain the index, feature size, and finger spacing of the radio frequency multi-finger device.

[0063] Specifically, the structural parameters of a radio frequency multi-finger device include the multi-finger device exponent n, feature size We, finger pitch d, emitter length Le, and power consumption P. diss and the first ambient temperature T a This embodiment uses the multi-finger device index n, characteristic size We, and finger spacing d in the structural parameters to calculate the coordinate system and the coordinates of the center point of each heat source.

[0064] S212. Using the center point of each finger of the radio frequency multi-finger device as the center point of the heat source, determine the origin and boundary position of the radio frequency multi-finger device according to the index, the feature size and the finger spacing.

[0065] Specifically, each finger of the multi-finger device acts as a heat source center, with the center point of each finger designated as the heat source center point. Based on the finger index, the characteristic dimensions, and the finger spacing of the RF multi-finger device, the device dimensions and boundary coordinates are obtained, thereby determining the origin and boundaries of the multi-finger device's coordinate system. Coordinate axes are formed along the direction of the heat source center points from the origin, thus establishing a coordinate system. The boundaries of the RF multi-finger device include a left boundary and a right boundary, which are designated as the boundary positions.

[0066] Please see Figure 3 , Figure 3 This is a schematic diagram illustrating the coordinate system determination and heat source center point confirmation provided in an embodiment of the present invention. Figure 3 In the example of InP HBT multi-finger device, the origin of the coordinate system is the center point of the left boundary, the direction of the origin of the coordinate system along the center point of the heat source is taken as the x-axis, and the left and right boundaries of the device are taken as the boundary positions.

[0067] S213. Calculate the coordinates of the center point of each heat source in the coordinate system according to the index, the characteristic size and the finger spacing.

[0068] Specifically, when the radio frequency multi-finger device is a uniformly distributed multi-finger device, the coordinates of the center point of each heat source are:

[0069] x center,k = k*x + (k-1)*W e +1 / 2*W e(1≤k≤n, k∈Z)

[0070] Where, x center,k Let be the coordinates of the center point of the k-th heat source, n be the exponent, and x be the distance between the exponents, i.e., d, W. e This refers to the feature size.

[0071] When the radio frequency multi-finger device is a uniformly distributed multi-finger device, the coordinates of the center point of the heat source of the device are manually calculated.

[0072] S22. Set several fitting data points according to the accuracy of temperature distribution.

[0073] Specifically, the number of fitting points (m) is given, thereby determining all fitting data points x for the multi-finger device. i (1≤i≤m, i∈Z). The number of fitting points m can be arbitrarily given by the user according to their requirements for temperature distribution accuracy, without any restriction. Furthermore, the efficiency of temperature analysis based on neural networks will not decrease with an increase in the number of fitting points. The number of fitted data points is consistent with the total number of fitting points.

[0074] S23. Using the single-finger device temperature distribution model, and combining the coordinates of each heat source center point, calculate the first coupling temperature rise from all heat source center points to the first fitted data point under the first ambient temperature, and superimpose the first ambient temperature and the first coupling temperature rise to calculate the heat distribution value of the first fitted data point.

[0075] Specifically, the temperature at each first fitted data point consists of two parts: the first ambient temperature T. a and the first coupled temperature rise from the center point of all heat sources to a certain first fitted data point In this context, combining the coordinates of the center point of each heat source, the single-finger device temperature distribution model is used at the first ambient temperature T. a Emitter length Le of RF multi-finger device; Power consumption P of RF multi-finger device diss Under the condition of [condition], the first coupled temperature rise from the center point of all heat sources to a certain first fitted data point is calculated. Therefore, based on the temperature superposition algorithm, the first ambient temperature T is [calculated]. a and the first coupling temperature rise The temperature value of the first fitted data point is obtained by superimposing the data, which yields the heat distribution value of that first fitted data point. For the first fitted data point x... i Its heat distribution value is calculated by the following formula:

[0076]

[0077] in, T represents the heat distribution value of the first fitted data point. a The first ambient temperature, Let n be the first coupled temperature rise of the first fitted data point, n be the exponent, k be the kth heat source, and ΔT be the temperature rise of the first coupled data point. k For the k-th heat source pair x i Thermal coupling temperature rise at the point, P diss For power dissipation, L e Let x be the emitter length, and |Δx| be x. i The distance A to the center point of the kth heat source ANN (T a ,P diss ,L e ,|Δx|) is a single-finger device temperature distribution model based on the first ambient temperature.

[0078] Furthermore, the thermal distribution values ​​of m first fitted data points are calculated using the above method, and the temperature distribution map of the multi-finger device can be drawn based on the thermal distribution values ​​of the m fitted data points.

[0079] S24. Using the heat distribution value as the second ambient temperature of the second fitted data point, the second coupling temperature rise from the center point of all heat sources to the second fitted data point is calculated using the single-finger device temperature distribution model at the second ambient temperature, and the heat distribution value and the second coupling temperature rise are superimposed to calculate the temperature value of the second fitted data point; wherein, along the boundary position towards the center point of the heat source, the first fitted data point is the previous neighboring point of the second fitted data point.

[0080] Specifically, to consider the impact of local temperature rise on the thermal conductivity of the material, an iterative approach is incorporated into the superposition principle to form an improved superposition algorithm. Since the temperature at the device edge is lower, the thermal conductivity of the material is less affected by thermal coupling, and the temperature is relatively stable. However, closer to the center of the device's heat source, the thermal conductivity changes more rapidly with temperature, leading to a greater increase in the device junction temperature. Therefore, in this embodiment, when calculating from the boundary position towards the center of the device's heat source, the ambient temperature of a certain fitted data point is selected as the temperature value of the previous fitted data point. That is, when calculating the temperature of the next fitted data point, the temperature value of the previous adjacent fitted data point is used. As the local ambient temperature at that point, this approach can take into account the degradation of the material's thermal conductivity due to thermal coupling effects at the heat source location as much as possible, and it eliminates the need for the iterative process of previous algorithms, thus saving computer time.

[0081] Based on the improved superposition algorithm described above, this embodiment first uses the heat distribution value of the first fitted data point as the second ambient temperature of the second fitted data point. Combining the coordinates of each heat source center point, and utilizing the single-finger device temperature distribution model, the second ambient temperature, the emitter length Le of the RF multi-finger device, and the power consumption P of the RF multi-finger device are considered. dissUnder the given conditions, the second coupled temperature rise from the center point of all heat sources to the second fitted data point is calculated; then, the second ambient temperature and the second coupled temperature rise are superimposed to obtain the temperature value of the second fitted data point. It can be understood that, along the boundary position towards the center point of the heat source, the first fitted data point is the previous neighboring point of the second fitted data point.

[0082] Specifically, the degree value of a certain second fitted data point is:

[0083]

[0084] in, The temperature value is the second fitted data point. The second coupled temperature rise for the second fitted data point. Let be the heat distribution value of the first fitted data point, n be the exponent, k be the kth heat source, and ΔT be the heat distribution value of the first fitted data point. k For the k-th heat source pair x i Thermal coupling temperature rise at the point, P diss For power dissipation, L e Let |Δx| be the emitter length, and |Δx| be the distance from the center of the heat source to the periphery. This is a temperature distribution model for a single-finger device based on thermal distribution values.

[0085] S25. Calculate the temperature value of each fitted data point to obtain the temperature distribution of the RF multi-finger device.

[0086] Specifically, repeat steps S23 and S24. Using the superposition principle, first calculate the thermal coupling temperature rise of all finger-pair device coordinate origins to obtain the temperature distribution value at that point. Then, at the next fitted data point, based on the improved superposition algorithm (that is, setting the ambient temperature of a certain point as the thermal distribution value of the neighboring point, fully considering the influence of the ambient temperature rise caused by thermal coupling on the thermal conductivity of the device material), obtain the temperature distribution value of the next fitted data point. Repeat this process to obtain the temperature values ​​of all fitted data points, thus obtaining the temperature distribution of the RF multi-finger device and completing the model establishment of the temperature distribution of the multi-finger device.

[0087] Please see Figures 4a-4c , Figures 4a-4c This is a schematic diagram of the temperature distribution and error distribution of a radio frequency multi-finger active device provided in an embodiment of the present invention. Figure 4a The image shows a comparison between the temperature distribution of the four-finger device obtained using the GA-BP neural network and the improved superposition algorithm, and the temperature distribution of the four-finger device obtained using COMSOL simulation. Figure 4b The image shows a comparison between the temperature distribution of the six-finger device obtained using the GA-BP neural network and superposition algorithm and the temperature distribution of the six-finger device obtained using COMSOL simulation. Figure 4cThe temperature distribution error diagrams of 4-finger and 6-finger devices obtained using the method of the embodiments of the present invention are shown. Figures 4a-4c The temperature distribution of 4-finger and 6-finger devices was analyzed using the method of this embodiment, and compared with the temperature distribution results obtained by the finite element analysis method. The comparison showed that the prediction error of the two different size devices was within 1.8%, and the prediction accuracy was high.

[0088] This embodiment of the method for analyzing the temperature distribution of RF multi-finger active devices leverages the powerful nonlinear self-learning capability of neural network models and combines an improved superposition algorithm to quickly and accurately characterize the temperature distribution of RF multi-finger active devices. The method first uses a single-finger device temperature distribution model to calculate the thermal distribution value of the first fitted data point at ambient temperature. Then, the thermal distribution value of the first fitted point is used as the second ambient temperature for the second fitted data point. The single-finger device temperature distribution model is then used to calculate the temperature value of the second fitted data point, thus obtaining the temperature value for each fitted data point. By combining the single-finger device temperature distribution model with the improved superposition algorithm, the influence of thermal coupling on the thermal conductivity of the device material is fully considered. Only the single-finger device temperature distribution needs to be trained using a neural network model to efficiently and accurately obtain the temperature distribution of RF multi-finger devices under arbitrary exponents, finger spacing, ambient temperature, and power consumption. While ensuring and improving analysis accuracy, this method greatly improves analysis efficiency and is applicable to the temperature analysis of multi-finger devices at any process node. This method not only avoids the problems of high computer resource consumption and time-consuming analysis associated with finite element analysis methods but also achieves high accuracy in the characterization results.

[0089] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing the temperature distribution of radio frequency multi-finger active devices, characterized in that, Including the following steps: Determining the coordinate system, boundary position, and coordinates of each heat source center point based on the structural parameters of the RF multi-finger device includes: obtaining the index, feature size, and finger spacing of the RF multi-finger device; using the center point of each finger of the RF multi-finger device as the heat source center point, determining the coordinate origin and boundary position of the RF multi-finger device based on the index, feature size, and finger spacing; and calculating the coordinates of each heat source center point based on the coordinate origin and boundary position, according to the index, feature size, and finger spacing. Several fitting data points are set according to the accuracy of temperature distribution; Combining the coordinates of each heat source center point, the first coupling temperature rise from all heat source center points to the first fitted data point is calculated using the single-finger device temperature distribution model under the first ambient temperature. The first ambient temperature and the first coupling temperature rise are then superimposed to calculate the heat distribution value of the first fitted data point. The single-finger device temperature distribution model is obtained by training a neural network on the temperature distribution data of the radio frequency single-finger device along the direction from the heat source center point to the periphery. This includes: constructing an electrothermal coupling solid model of the radio frequency single-finger device using finite element analysis software; loading dissipation power and boundary conditions onto the electrothermal coupling solid model and performing finite element mesh generation; performing steady-state thermal analysis on the electrothermal coupling solid model under different dissipation powers, different emitter lengths, and different ambient temperatures to obtain temperature distribution data along the direction from the heat source center point to the periphery; using the dissipation power, emitter length, ambient temperature, and distance from the heat source center point to the periphery in the temperature distribution data as input vectors, and the temperature data in the temperature distribution data as expected values, a genetic algorithm is used to optimize the weights and thresholds of the neural network, and the optimal weights and thresholds are assigned to the neural network to obtain the single-finger device temperature distribution model. Using the heat distribution value as the second ambient temperature of the second fitted data point, the second coupled temperature rise from all heat source center points to the second fitted data point is calculated using the single-finger device temperature distribution model, and the heat distribution value and the second coupled temperature rise are superimposed to calculate the temperature value of the second fitted data point; wherein, along the boundary position toward the heat source center point, the first fitted data point is the previous neighboring point of the second fitted data point. Calculate the temperature value for each fitted data point to obtain the temperature distribution of the RF multi-finger device.

2. The method for analyzing the temperature distribution of radio frequency multi-finger active devices according to claim 1, characterized in that, The neural network includes a BP neural network; The BP neural network includes an input layer, several hidden layers, and an output layer connected in sequence.

3. The method for analyzing the temperature distribution of radio frequency multi-finger active devices according to claim 2, characterized in that, The plurality of hidden layers includes a first hidden layer and a second hidden layer, wherein, The input layer, the first hidden layer, the second hidden layer, and the output layer are connected in sequence; The first hidden layer comprises 6 neurons, and the second hidden layer comprises 3 neurons.

4. The method for analyzing the temperature distribution of radio frequency multi-finger active devices according to claim 2, characterized in that, The transfer function of the hidden layer neurons in the BP neural network includes the tansig function, and the output layer function includes the purelin function.

5. The method for analyzing the temperature distribution of radio frequency multi-finger active devices according to claim 1, characterized in that, When the radio frequency multi-finger device is a uniformly distributed multi-finger device, the coordinates of the center point of each heat source are: in, For the first Coordinates of the center point of each heat source For index, The distance between the fingers is the distance between the fingers. Feature size.

6. The method for analyzing the temperature distribution of radio frequency multi-finger active devices according to claim 1, characterized in that, The heat distribution value of the first fitted data point is: in, The heat distribution value of the first fitted data point. The first ambient temperature, The first coupling temperature rise for the first fitted data point. For index, For the first One heat source, For the first Each heat source Thermal coupling temperature rise at the location, For power dissipation, For emitter length, for To the The distance between the center points of the heat sources This is a temperature distribution model for a single-finger device based on the first ambient temperature.

7. The method for analyzing the temperature distribution of radio frequency multi-finger active devices according to claim 1, characterized in that, The temperature values ​​of the second fitted data point are: in, The temperature value is the second fitted data point. The second coupled temperature rise for the second fitted data point. The heat distribution value of the first fitted data point. For index, For the first One heat source, For the first Each heat source Thermal coupling temperature rise at the location, For power dissipation, For emitter length, The distance from the center point of the heat source to the outer perimeter. This is a temperature distribution model for a single-finger device based on thermal distribution values.

Images