A large-signal reliability modeling method for multi-module degradation unit collaborative correction
The large-signal reliability modeling method with multi-module degradation unit collaborative correction solves the problem of rapid and accurate modeling of RF devices under multi-stress environments, realizes the reliability design and simulation of RF circuits and systems, and is applicable to various reliability factors of microwave devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-16
AI Technical Summary
Existing reliability research on radio frequency devices and circuits focuses on the physical level of the devices. The research and development cycle is long and the modeling speed is slow. It is impossible to quickly and accurately model the reliability of devices under multiple stress environments, such as electrical degradation, thermal degradation and aging degradation, which makes it difficult to meet the requirements of high reliability design.
A large-signal reliability modeling method with multi-module degradation unit collaborative correction is established. Initial parameters are extracted through compact model topology, the degradation behavior of key device characteristics is analyzed, a correction network is constructed to predict sensitive parameters, and multiple degradation correction modules are constructed for collaborative correction to form a large-signal reliability model.
It enables rapid and accurate simulation and prediction of device degradation under multi-stress environments, is suitable for the reliability design of microwave devices, simplifies RF circuit and system-level simulation, has strong applicability, and can take into account various factors such as high and low temperatures and electrothermal aging.
Smart Images
Figure CN122221641A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of microelectronics and integrated circuit analysis technology, specifically relating to a large-signal reliability modeling method for collaborative correction of multi-module degradation units. Background Technology
[0002] In recent years, with the rapid development of information technology, people have increasingly higher requirements for data transmission speed, making the demand for wide bandwidth, high speed, and high power in wireless communication systems more and more urgent. Among these, radio frequency semiconductor chips are a crucial component of wireless communication systems. Microwave and millimeter-wave integrated circuits are often used under various complex operating conditions, making it increasingly important to ensure their high reliability under multi-stress scenarios.
[0003] As the performance requirements for key components in RF integrated chips increase, the current density and internal electric field strength during device operation also increase, leading to serious electrical, thermal, and electrothermal coupling reliability issues, thus revealing potential reliability challenges. Device models serve as a crucial bridge between device fabrication and circuit design. Reliability models, built upon these models and incorporating reliability degradation considerations for different scenarios, are key to achieving high-reliability design and verification. Furthermore, in the RF circuit design flow, from selecting the static operating point to simulating power harmonic characteristics and intermodulation features, and finally to layout simulation, accurate large-signal models are indispensable.
[0004] However, existing research on the reliability of RF devices and circuits still focuses on the device physics level. Its long development cycle, slow modeling speed, and difficulty in compatibility with circuit design software all pose challenges to the design of high-reliability products. Furthermore, the development of existing large-signal reliability models for RF active devices has been slow, failing to provide assistance for efficient high-reliability circuit design. From a practical application perspective, large-signal reliability modeling of RF active devices needs to consider the comprehensiveness of various effects and parameter degradation. It must consider not only the degradation information of small-signal peripheral resistances but also the degradation behavior of the intrinsic current and charge characteristics of the device under multiple biases and wide frequency bands in different reliability application scenarios. For the complex degradation process caused by increased power density and improved performance of the device itself, the degradation parameters of the corresponding DC, AC, and thermal characteristics of the device exhibit different sensitivities and drift trends under complex reliability environments, making it difficult to unify the parameter degradation modeling methods for device large-signal models related to reliability parameters. In addition, the introduction of multi-dimensional reliability parameters also makes parameter extraction more time-consuming and difficult.
[0005] In summary, how to quickly and accurately perform large-signal reliability modeling for devices in multi-stress environments, including electrical degradation, thermal degradation, and aging degradation, and how to achieve reliability-aware design of microwave devices in different engineering applications and reliability assessment of integrated circuits are urgent technical problems that need to be solved. Summary of the Invention
[0006] To address the aforementioned problems in the existing technology, this invention provides a large-signal reliability modeling method with multi-module degradation unit collaborative correction. The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, this invention proposes a large-signal reliability modeling method for collaborative correction of multi-module degradation units, including: S1: Establish a corresponding compact model topology for RF active devices and extract the undegraded initial parameters of each component in the compact model topology; S2: Based on the initial parameters, analyze the degradation behavior of key device characteristics, and use the compact model topology to simulate the degradation behavior of key device characteristics, and determine the sensitive parameters that affect the characterization of device degradation characteristics. S3: Construct a correction network and train the correction network to quickly predict different levels of sensitivity and drift trends of sensitive parameters, thereby mapping multi-dimensional input vectors to sensitive parameter correction amounts; S4: Based on the number and type of sensitive parameters, construct multiple different degradation correction modules. Each degradation correction module includes multiple parallel distributed correction networks to collaboratively correct different types of sensitive parameters. Embed the output sensitive parameter correction amount into the compact model topology to form a large signal reliability model.
[0007] In a second aspect, the present invention proposes an electronic device comprising a large-signal reliability model established using the method provided in the first aspect of the present invention. This large-signal reliability model is used for reliability-aware design of radio frequency circuits, aging lifetime prediction of integrated circuits, and optimization design of high-reliability electronic systems.
[0008] The beneficial effects of this invention are: This invention provides a large-signal reliability modeling method with multi-module degradation unit collaborative correction. First, a corresponding compact model topology is established for RF active devices, and the initial parameters of each component in the compact model topology are extracted to prevent degradation. Then, based on the initial parameters, the degradation behavior of key device characteristics is analyzed, and the degradation behavior of key RF active device characteristics is simulated using the compact model topology to determine the sensitive parameters affecting the characterization of device degradation characteristics. Next, a correction network is constructed, and by training the correction network, the different sensitivity levels and drift trends of the sensitive parameters are rapidly predicted, thereby mapping the multi-dimensional input vector to the sensitive parameter correction amount. Finally, based on the number and type of sensitive parameters, multiple different degradation correction modules are constructed. Each degradation correction module includes multiple parallel distributed correction networks to collaboratively correct different types of sensitive parameters, and the output sensitive parameter correction amount is embedded into the compact model topology to form a large-signal reliability model. This method has a simple implementation process, is easy to use for RF circuit or system-level simulation, and has strong applicability. The reliability model established based on this method can simultaneously consider multiple reliability factors such as high and low temperatures and electrothermal aging, and can quickly and accurately simulate and predict the degradation of large signal characteristics such as radio frequency scattering parameters, DC, AC characteristics and power characteristics of microwave devices. At the same time, it can avoid the influence caused by the inconsistency of the dimensions of DC, AC and thermal parameter degradation, and is widely applicable to the reliability design of microwave devices in different engineering applications.
[0009] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0010] Figure 1 A flowchart illustrating a large-signal reliability modeling method for collaborative correction of multi-module degradation units provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the correction network provided in an embodiment of the present invention; Figure 3 A framework diagram of the large-signal reliability model provided in the embodiments of the present invention; Figure 4 A flowchart illustrating the classification of sensitive regions under electrothermal stress in an HBT device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of multiple degradation correction modules constructed for HBT devices according to an embodiment of the present invention; Figure 6 This is a comparison chart of simulation results and actual degradation of the electrothermal aging sensitive parameters of HBT devices according to an embodiment of the present invention; Figure 7 This invention presents a graph showing the degradation process of the forward Gummel curve of an HBT device under electrothermal aging stress and the corresponding degradation curve of the forward active current gain under bias voltage. Figure 8 This is a schematic diagram illustrating the degradation process of the HBT device's reverse Gummel curve, i.e., the input characteristics of the device's BC junction, under electrothermal aging stress, according to an embodiment of the present invention. Figure 9 This is a schematic diagram illustrating the degradation process of the IV output characteristic curve of an HBT device under electrothermal aging stress according to an embodiment of the present invention. Detailed Implementation
[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0012] The first aspect of the present invention provides a large-signal reliability modeling method for collaborative correction of multi-module degradation units.
[0013] Please see Figure 1 , Figure 1 This is a flowchart illustrating a large-signal reliability modeling method for collaborative correction of multi-module degradation units provided in an embodiment of the present invention. The method includes: S1: Establish a corresponding compact model topology for RF active devices and extract the undegraded initial parameters of each component in the compact model topology.
[0014] Specifically, in this embodiment, a compact model topology can be established by directly selecting the model of the corresponding device. Then, large-signal parameter extraction technology is used to extract the undegraded initial parameters of each component in the compact model topology.
[0015] S2: Based on the initial parameters, analyze the degradation behavior of key characteristics of the device, use the compact model topology to simulate the degradation behavior of key characteristics of RF active devices, and determine the sensitive parameters that affect the characterization of device degradation characteristics.
[0016] Specifically, firstly, it is necessary to clarify the main degradation behaviors of the device under reliability scenarios (such as electrical stress, thermal stress, and aging) in order to identify the degradation behavior of key characteristics.
[0017] Then, based on the physical characteristics of the device, we analyze which parameters are most susceptible to degradation driving forces (stress), thereby identifying the sensitive parameters that affect the characterization of device degradation characteristics.
[0018] Finally, the sensitive parameters are simulated using simulation tools to determine the sensitive parameters.
[0019] It is understandable that the sensitive parameters are a subset of the initial parameters.
[0020] S3: Construct a correction network and train the correction network to quickly predict different levels of sensitivity and drift trends of sensitive parameters, thereby mapping multi-dimensional input vectors to sensitive parameter correction amounts.
[0021] Alternatively, as one implementation method, please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of the correction network provided in an embodiment of the present invention. The correction network designed in this embodiment adopts a single-hidden-layer extreme learning machine architecture, including an input hidden layer and an output layer; wherein, The input layer is used to receive multidimensional input vectors; Hidden layers consist of multiple neurons used to compute an output matrix from a multidimensional input vector; The output layer is used to generate independent sensitivity parameter corrections for each sensitivity parameter based on the output matrix of the hidden layer.
[0022] Specifically, in this embodiment, the correction network is also a feedforward neural network used for the degradation parameters of the subsequent large signal model, and the amount of sensitivity parameter correction output by it is the degradation parameter of the large signal model.
[0023] Optionally, in this embodiment, the multidimensional input vector may include reliability input parameters, device geometry, and operating bias, expressed by the formula: ; In the formula, Input parameters for reliability. For device geometry information, For work bias, T The number of training set samples for the proposed network.
[0024] It can be understood that the dimensions of the multidimensional input vector can be adjusted according to the actual reliability application scenario.
[0025] Furthermore, in the hidden layer, the number of neurons can be determined through empirical estimation followed by experimental tuning; each neuron uses the Sigmoid function as its transfer function, the expression of which is: ; In the formula, This is a weighted input term for each hidden layer neuron.
[0026] Furthermore, the bias vector of the hidden layer neurons can be set as follows: ; In the formula, This represents the bias vector of a hidden layer neuron. Indicates the first h Bias of hidden layer neuronsH This represents the number of neurons in the hidden layer of the network.
[0027] Furthermore, for the modified network constructed in this embodiment, the connection weight vector between its input layer and hidden layer can be set as follows: ; In the formula, This is the connection weight vector between the input layer and the hidden layer. , , These are the reliability input parameters, the device geometry, and the connection weights between the input layer and the hidden layer for the operating bias, respectively. H This represents the number of neurons in the hidden layer of the network.
[0028] It should be noted that the connection weight vectors between the input layer and the hidden layer, as well as the bias vectors of the hidden layer neurons, need to be randomly assigned before training and remain unchanged during training.
[0029] After defining all the structural variables in the modified network as described above, the hidden layer output matrix can be obtained as shown in the following equation: ; Accordingly, the training objective matrix of the corrected network is denoted as... T， Its expression is: ; In the formula, For parameters sensitive to large signals, n This represents the corresponding number of sensitive parameters.
[0030] The training objective function of the modified network is then expressed as: ; In the formula, The predicted value of the target matrix. M This is the output matrix of the hidden layer. β To correct the connection weight matrix between the hidden and output layers that needs to be solved during network training, This is the norm symbol.
[0031] Furthermore, in this embodiment, the connection weight matrix between the hidden layer and the output layer can be solved using the mathematical principle that solving for the least squares solution is equivalent to minimizing the training objective function. Since directly solving for the least squares solution of the above equation is equivalent to minimizing the training objective function, its least squares solution can be directly written out: ; In the formula, This represents the connection weight matrix between the hidden layer and the output layer obtained by solving, which is also the predicted connection weight matrix between the hidden layer and the output layer. Representation matrix M The generalized inverse matrix.
[0032] The final prediction target matrix for: .
[0033] The correction network constructed in this invention, through training, can establish a sensitive parameter mapping network between the sensitive parameter degradation amount and actual reliability parameters, size, bias and other variables, thereby obtaining the sensitive parameter correction amount to correct the model topology.
[0034] S4: Based on the number and type of sensitive parameters, construct multiple different degradation correction modules. Each degradation correction module includes multiple parallel distributed correction networks to collaboratively correct different types of sensitive parameters. Embed the output sensitive parameter correction amount into the compact model topology to form a large signal reliability model.
[0035] For details, please see Figure 3 , Figure 3 This is a framework diagram of the large-signal reliability model provided in an embodiment of the present invention. The input vector of this large-signal reliability model includes three types of vectors: reliability input parameters, device geometry, and operating bias. Furthermore, different mapping modules, i.e., degradation correction modules, are established by determining the device's DC, AC, thermal characteristics, and peripheral degradation parameters. Each degradation correction module is trained using multiple parallel-distributed correction network structures established in S3 to achieve rapid prediction of the different sensitivities and drift trends of each degradation parameter.
[0036] Optionally, in this embodiment, the various degradation correction modules include an external resistance correction module, a DC characteristic correction module, an AC characteristic correction module, and a thermal characteristic correction module.
[0037] It is understandable that the number of degradation correction modules and their network count can be adjusted using the actual number of sensitive parameters determined in S2.
[0038] Thus, modeling of large-signal reliability through collaborative correction of multi-module degradation units has been achieved.
[0039] Furthermore, after obtaining the large-signal reliability model, it also includes: S5: Establish an interface between the large-signal reliability model and variables to embed it into microwave RF device circuit design software, thereby enabling the simulation and prediction of the degradation of the large-signal characteristics of the device.
[0040] Specifically, an interface can be established between the established large-signal reliability model and actual reliability parameters, dimensions, bias and other variables, and embedded into the circuit design software to realize the degradation simulation and prediction of large-signal characteristics such as RF scattering parameters, DC and AC characteristics and power characteristics of microwave RF devices.
[0041] In summary, this invention provides a large-signal reliability modeling method for collaborative correction of multi-module degradation units. First, a corresponding compact model topology is established for RF active devices, and the initial parameters of each component in the compact model topology are extracted to prevent degradation. Then, based on the initial parameters, the degradation behavior of key device characteristics is analyzed, and the degradation behavior of key RF active device characteristics is simulated using the compact model topology to determine the sensitive parameters affecting the characterization of device degradation characteristics. Next, a correction network is constructed, and by training the correction network, the different sensitivity levels and drift trends of the sensitive parameters are rapidly predicted, thereby mapping the multi-dimensional input vector to the sensitive parameter correction amount. Finally, based on the number and type of sensitive parameters, multiple different degradation correction modules are constructed. Each degradation correction module includes multiple parallel distributed correction networks to collaboratively correct different types of sensitive parameters, and the output sensitive parameter correction amount is embedded into the compact model topology to form a large-signal reliability model. This method has a simple implementation process, is easy to use for RF circuit or system-level simulation, and has strong applicability. The reliability model established based on this method can simultaneously consider multiple reliability factors such as high and low temperatures and electrothermal aging, and can quickly and accurately simulate and predict the degradation of large signal characteristics such as radio frequency scattering parameters, DC, AC characteristics and power characteristics of microwave devices. At the same time, it can avoid the influence caused by the inconsistency of the dimensions of DC, AC and thermal parameter degradation, and is widely applicable to the reliability design of microwave devices in different engineering applications.
[0042] The following section uses a heterojunction bipolar transistor (HBT) as an example to demonstrate and verify the method of the present invention by using the large-signal reliability modeling method of multi-module degradation unit collaborative correction.
[0043] 1. Experimental setup: Apply electrical-thermal-aging stress to the HBT devices used.
[0044] Specifically, this experiment used a Keysight B1500A semiconductor device parameter analyzer to apply high current stress to the device via a DC probe. The B1500A is equipped with a highly precise DC probe cycle control module for accurate stress application. This module can also monitor changes in the device's junction characteristics and whether breakdown occurs during the test. The device's base was set to current input mode, and the collector and emitter were set to voltage input mode, with a control stress bias of V. CE =2V, J C =360kA / cm 2 Furthermore, the electrothermal stress lasts for more than 2000 minutes, with a junction temperature of 130.8℃. The aforementioned cycle control module is used to control the stress application and measure the electrical characteristics of the device before and after the stress application.
[0045] 2. Sensitive parameter analysis: During the electro-thermal-aging stress process, the characteristic changes of the corresponding devices are analyzed. Under electro-thermal stress, the heat of the HBT device mainly accumulates at the BC junction. After a long period of stress, a large amount of heat accumulates at this junction and diffuses to the surrounding area, forming a temperature gradient. The high current density applied causes atoms in the metal electrodes to migrate, forming high-resistivity intermetallic compounds at the metal-semiconductor contact. Since the base and collector contacts are closer to the heat source, they are more susceptible to the influence, thus the base and collector resistance parameters will degrade accordingly. Secondly, the degradation of the resistance parameters will also change the current characteristics of the device, resulting in a decrease in the actual voltage dropped across the junction, a corresponding weakening of the voltage used to reduce the potential barrier height, and a decrease in the number of conductive charge carriers inside the device. The degradation of the corresponding current characteristics is simulated by adjusting the model current parameters in the corresponding regions.
[0046] Based on the above degradation analysis and actual device testing, a flowchart for analyzing the electrothermal aging-sensitive parameters of HBT devices using the Keysight AHBT compact model is presented. Please refer to [link / reference]. Figure 4 , Figure 4 This is a flowchart illustrating the classification of sensitive regions under electrothermal stress in an HBT device according to an embodiment of the present invention. The sensitive parameters are extracted from the sensitive regions using the reliability modeling method proposed in this invention. First, the sensitive peripheral resistance parameters are obtained using the cutoff state method. R B and R C These parameters primarily affect the linear region of the device's current gain characteristics and IV output characteristics. Secondly, the AHBT large-signal model is used to extract the sensitive parameters that mainly influence the forward and reverse Gummel characteristics and IV characteristics. I SH , N H , ISC , R TH The detailed physical correlation descriptions of the above sensitive parameters are shown in Table 1.
[0047] Table 1 Description of the physical correlation of sensitive parameters
[0048] 3. Establishment of the correction network: Once the sensitive parameters are determined, the degradation amount of each sensitive parameter is captured during the electrothermal degradation process of the device. This degradation amount is a complex variable related to electrical stress, thermal stress, and aging time. A sensitive parameter degradation network, also known as a degradation correction module, is constructed based on this. Its structure is as follows: Figure 5 As shown.
[0049] For each training sample, the identified sensitive parameters are divided into three modules: resistance, DC, and self-heating, corresponding to three degradation correction modules. The number of small units in each module is the same as the number of output parameters of the corresponding module. All modules involve a correction network with six parallel-distributed parameter small units. The output of these six units is... R B Age , R C Age , I SH Age , N H Age , I SC Age , R TH Age These are the electrothermal aging correction terms for the corresponding original model parameters, and then each small unit is... Figure 2 The training shown has a network input layer with four input neurons that describe the degradation of device parameters, representing the electrical stress vector. E stress thermal stress vector Tj stress and aging time vector t stress ,in, E stress Based on the devices used, it is classified into collector voltage stress. V CE,stress and collector current stress J C,stress The established Extreme Learning Machine (ELM) network has a single-layer hidden layer structure, with each unit containing 12 hidden neurons. After training, the network was used to simulate six degradation parameters in degradation-sensitive regions. Please refer to [link / reference]. Figure 6 , Figure 6 This is a comparison chart of simulation results and actual degradation of the electrothermal aging sensitive parameters of HBT devices according to an embodiment of the present invention. Figure 6 Figures (a), (b), (c), (d), and (e) show the comparison curves between the simulated degradation results and the actual parameter degradation for the peripheral base, collector resistance sensitive parameter, forward Gummel sensitive parameter, reverse Gummel sensitive parameter, and thermal parameter, respectively. It can be seen that the model structure established using the method of this invention provides a very good simulation effect on different degradation trends and degrees of the sensitive parameters, closely matching the actual degradation process. The overall average relative error range is approximately 2.45% to 3.59%, enabling the rapid and accurate establishment of parameter correction models.
[0050] At this point, the above parameter correction terms have been constructed. In ADS, the device is defined using symbols and embedded into the AHBT large-signal model, so that the corrected large-signal model can accurately characterize the degradation process of device characteristics in electrothermal application scenarios.
[0051] Based on existing tests, the proposed modeling method is further validated by comparing and analyzing the simulation results of the device's DC characteristics under different conditions.
[0052] 4. Verification and Analysis: Please see Figure 7 , Figure 7 This invention presents a degradation process diagram of the forward Gummel curve of an HBT device under electrothermal aging stress and a corresponding degradation curve of the forward active current gain under bias voltage; wherein, Figure 7 Figure (a) shows the degradation process of the forward Gummel curve, and Figure (b) shows the degradation curve of the forward active current gain under the corresponding bias voltage. Figure 7 As can be seen, with the increase of stress application time, the main effect is on the characteristics of the high bias region of the BE junction of the device, and the base current. I b With collector current I c All decreased, but the magnitudes differed, as reflected in... Figure 7 (b) The decrease in the ratio of collector current to base current in the high-bias region of the BE junction, i.e., the decrease in device current gain. Correspondingly, I b , I c The reduction and β The reduction is affected by the device's sensitive parameters. Figure 6 The degradation of the series resistance in figures (a) and (b) and the saturation current coefficient in the high bias region of the base in figure (c) are shown. I SH and ideal factor NH Joint regulation.
[0053] Therefore, the modeling method proposed in this invention effectively captures the electrothermal aging degradation trend of the base-emitter junction current of the device during forward active amplification.
[0054] Furthermore, in addition to focusing on the characteristics of the BE junction, Figure 8 It also reflects the degradation process of the HBT device's reverse Gummel curve, i.e., the input characteristics of the device's BC junction, under electrothermal aging stress. From Figure 8 As can be seen, with the increase of stress duration, the BC junction of the device also exhibits the same base current under the same junction bias in the high bias region due to the degradation of the series resistance. I b and emitter current I e The base current decreases. Furthermore, in the low bias region, the base current... I b The leakage current gradually decreases during the stress process, and the change in this leakage region is reflected in... Figure 6 The corresponding saturation current coefficient in Figure (d) I SC On the sensitivity of.
[0055] Furthermore, after verification analysis of the junction characteristic degradation of the device, Figure 9 The figure shows the degradation process of the IV output characteristic curve of an HBT device under electrothermal aging stress. It can be seen that the slope of the linear region of the device's output characteristic decreases significantly. This is due to the degradation process during electrothermal aging, such as... Figure 6 The collector resistor shown in Figure (b) R C This is caused by degradation. As the collector voltage increases, the non-ideal interface traps in the base plate (BC) lead to charge accumulation and recombination effects, creating a local electric field anomaly in the collector region. This suppresses carrier transitions, causing the device to reach saturation only at higher collector voltages, resulting in a soft knee effect. During electrothermal aging, the concentration of interface traps increases with stress time, gradually strengthening the suppression of carrier injection in the base region. The figure shows a more severe soft knee phenomenon, with the linear region widening and the knee voltage shifting backward. The knee voltage refers to the collector-emitter voltage corresponding to the collector current reaching saturation, which is the boundary between the linear and saturation regions of the transistor. Furthermore, electrothermal aging stress also has a certain impact on the self-heating effect in the high-power region, manifested in… Figure 6 The change in thermal resistance of the device is shown in Figure (e). Figure 9 As can be seen, with the increase of the applied stress time, the given collector voltage... V CEThe current shows an upward trend because the collector junction resistance increases, which increases the voltage drop across the device, and the actual voltage drop across the device decreases as the stress increases.
[0056] Thus, based on the large-signal reliability modeling framework and AHBT model topology proposed in this invention, an electrothermal aging current model considering the combined effects of thermal stress, electrical stress, and time dimension has been established, and the effectiveness of this invention has been verified.
[0057] Based on the same inventive concept, a second aspect of the present invention also provides an electronic device comprising a large-signal reliability model established using the method provided in the first aspect of the present invention. This large-signal reliability model is used for reliability-aware design of radio frequency circuits, aging lifetime prediction of integrated circuits, and optimization design of high-reliability electronic systems.
[0058] 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 large-signal reliability modeling method with collaborative correction of multi-module degradation units, characterized in that, include: S1: Establish a corresponding compact model topology for RF active devices, and extract the undegraded initial parameters of each component in the compact model topology; S2: Based on the initial parameters, analyze the degradation behavior of key device characteristics, and use the compact model topology to simulate the degradation behavior of key device characteristics, and determine the sensitive parameters that affect the characterization of device degradation characteristics. S3: Construct a correction network and train the correction network to quickly predict the different sensitivity levels and drift trends of the sensitive parameters, thereby mapping the multi-dimensional input vector into the sensitive parameter correction amount; S4: Based on the number and type of the sensitive parameters, construct multiple different degradation correction modules. Each degradation correction module includes multiple parallel distributed correction networks to collaboratively correct the sensitive parameters of different categories. The output sensitive parameter correction amount is embedded in the compact model topology to form a large signal reliability model.
2. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 1, characterized in that, In S3, the corrected network adopts a single-hidden-layer extreme learning machine architecture, including an input hidden layer and an output layer; wherein, The input layer is used to receive multidimensional input vectors; The hidden layer includes multiple neurons, which are used to calculate the output matrix from the input multidimensional input vector; The output layer is used to generate an independent sensitivity parameter correction amount for each sensitivity parameter based on the output matrix of the hidden layer.
3. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 2, characterized in that, The multidimensional input vector includes reliability input parameters, device geometry, and operating bias.
4. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 2, characterized in that, Each neuron in the hidden layer uses the Sigmoid function as its transfer function, and its expression is: ； In the formula, This is a weighted input term for each hidden layer neuron.
5. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 2, characterized in that, The connection weight vector between the input layer and the hidden layer is: ； In the formula, This is the connection weight vector between the input layer and the hidden layer. , , These are the reliability input parameters, the device geometry, and the connection weights between the input layer and the hidden layer for the operating bias, respectively. H This represents the number of neurons in the hidden layer of the network. The bias vector of the hidden layer neurons is represented as follows: ； In the formula, This represents the bias vector of a hidden layer neuron. Indicates the first h Bias of hidden layer neurons.
6. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 2, characterized in that, The training target matrix of the corrected network is denoted as... T Its expression is: ； In the formula, For parameters sensitive to large signals, n The corresponding number of sensitive parameters; The training objective function of the modified network is then expressed as: ； In the formula, The predicted value of the target matrix. M This is the output matrix of the hidden layer. β To correct the connection weight matrix between the hidden and output layers that needs to be solved during network training, This is the norm symbol.
7. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 6, characterized in that, The connection weight matrix between the hidden layer and the output layer is solved using the mathematical principle that solving for the least squares solution is equivalent to minimizing the training objective function, and thus the following can be obtained: ； In the formula, This represents the connection weight matrix between the hidden layer and the output layer obtained by solving. Representation matrix M The generalized inverse matrix.
8. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 1, characterized in that, In S4, the various degradation correction modules include an external resistance correction module, a DC characteristic correction module, an AC characteristic correction module, and a thermal characteristic correction module.
9. The large-signal reliability modeling method for collaborative correction of multi-module degradation units according to claim 1, characterized in that, Also includes: S5: Establish an interface between the large-signal reliability model and the variables to embed it into the microwave RF device circuit design software to realize the simulation and prediction of the degradation of the large-signal characteristics of the device.
10. An electronic device, characterized in that, The method includes a large-signal reliability model established according to any one of claims 1-9, wherein the large-signal reliability model is used for reliability-aware design of radio frequency circuits, aging lifetime prediction of integrated circuits, and optimization design of high-reliability electronic systems.