A three-dimensional micro-assembly method for microwave hybrid integrated circuits

Abstract

The application discloses a three-dimensional micro-assembly method of a microwave hybrid integrated circuit, relates to the field of three-dimensional assembly control of circuits, and adopts a deep learning technology to perform visual feedback on microwave device preparation, so that the trained deep learning model can accurately predict the position and orientation of the device, thereby improving the accuracy in the three-dimensional assembly process and optimizing the assembly quality and efficiency of the microwave device.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional assembly control of circuits, and more particularly to a three-dimensional micro-assembly method for microwave hybrid integrated circuits. Background Technology

[0002] Microwave hybrid integrated circuits typically consist of multiple circuit components, including power amplifiers, mixers, filters, and so on. These components can be integrated onto a single chip to form a complete circuit. The design of microwave hybrid integrated circuits requires consideration of the characteristics and performance of each component to ensure the overall circuit performance and reliability. The design and manufacture of microwave hybrid integrated circuits utilize microwave integrated circuit technology. This technology integrates multiple circuit components onto a single chip, thereby reducing the size and complexity of the circuit. Microwave integrated circuit technology also improves circuit performance and reliability because it reduces interconnections and interference between components within the circuit.

[0003] 3D circuit assembly is an advanced electronic manufacturing technology that aims to vertically stack and connect multiple chip layers within a single integrated circuit (IC) package to achieve higher integration and performance. Traditional integrated circuits typically arrange all components and interconnects on a single plane, limiting further improvements in performance and size. 3D circuit assembly, by stacking multiple chips vertically, enables more compact designs, shorter signal transmission distances, and higher performance. The core idea of ​​this technology is to stack chips with different functions together and connect them using fine interconnects. These interconnects can be implemented in various ways, such as through inter-chip metal traces or through-silicon vias (TSVs). 3D circuit assembly enables higher data transmission rates, lower power consumption, smaller package sizes, and more powerful computing capabilities.

[0004] In the current microwave hybrid integrated circuits, the three-dimensional assembly process often suffers from inaccurate positioning of microwave devices, requiring manual verification. Therefore, a high-precision, high-performance method is needed to manage the assembly of microwave hybrid integrated circuits. Summary of the Invention

[0005] The purpose of this invention is to provide a three-dimensional micro-assembly method for microwave hybrid integrated circuits to solve the above-mentioned problems.

[0006] This invention is achieved through the following technical solution:

[0007] A three-dimensional micro-assembly method for microwave hybrid integrated circuits includes the following steps:

[0008] S1. Using a silicon nitride substrate as the base, the substrate is placed on an assembly template, and the geometric features of the assembly template are used for auxiliary positioning and alignment, followed by micro-welding.

[0009] S2. Microwave devices required for fabricating microwave integrated circuits, wherein the microwave devices include at least a quarter-wavelength microstrip line, dual parallel PIN diodes, a DC bias circuit, and a control switch circuit;

[0010] S3. Fabricate three-dimensional assembled components and packaged components, wherein the three-dimensional assembled components include at least stacked radio frequency components and integrated antenna arrays, and the packaged components include at least QFN packages and BGA packages;

[0011] S4. Precisely assemble the prepared microwave devices, three-dimensional assembled components, and packaging components;

[0012] Specifically, step S4, which involves the precise assembly of microwave devices, includes the following sub-steps:

[0013] During the assembly process, S401 uses a high-resolution camera or microscope to capture images of microwave devices and adds labels to the captured images. The labels are added to the image data according to the location and orientation of the microwave devices, and the labels indicate the bounding box and category of the microwave devices.

[0014] S403. Input the image into the trained YOLO model. The model detects and locates devices in the image through multi-scale prediction and grid segmentation, and generates predicted bounding boxes and category information.

[0015] Furthermore, it also includes a pre-step S0, which trains the YOLO model through sampling, specifically including the following sub-steps:

[0016] S001. Collect image samples of microwave devices, extract sample labels from the images through feature extraction, and generate a training set using the sample labels;

[0017] S002. Generate a YOLO model, train the YOLO model, calculate the parameter gradients using the object detection loss, class classification loss, and bounding box coordinate loss functions, and adjust the YOLO model based on the calculation results.

[0018] Furthermore, in step S002, calculating the parameter gradient using the object detection loss function, the category classification loss function, and the bounding box coordinate loss function specifically includes the following sub-steps:

[0019] S0021. Calculate the target loss, specifically as follows:

[0020]

[0021] Wherein, the L objThe loss represents the bounding box coordinates, where λ is the bounding box coordinate loss. obj The weights represent the object detection loss, where s represents the number of grid cells, B represents the number of predicted bounding boxes in each grid cell, i represents the index of the grid cell, j represents the index of the bounding box, and IOU represents the weights of the object detection loss. ij The intersection-union ratio (IoU) between the predicted bounding box j and the ground truth bounding box is represented by bx. i The term "by" represents the horizontal offset of the predicted bounding box center relative to the i-th grid cell. i The value represents the longitudinal offset of the predicted bounding box center relative to the i-th grid cell. This indicates an indicator function, which is 1 if the j-th bounding box of the i-th grid cell contains the target, and 0 otherwise.

[0022] S0022. Calculate the category classification loss, specifically as follows:

[0023]

[0024] Wherein, the L cls λ represents the category classification loss. cls The weights represent the class classification loss, where the class i p represents the predicted probability of the category corresponding to the predicted bounding box j. i The probability of the true label representing the category to which the true bounding box belongs;

[0025] S0023. Calculate the frame coordinate loss, specifically:

[0026]

[0027] Wherein, the L coord The loss represents the bounding box coordinates, where λ is the bounding box coordinate loss. coord The weights representing the loss of the bounding box coordinates, w i The square root of the width of the predicted bounding box j is represented by the... h represents the square root of the width of the actual bounding box. i The square root of the height of the predicted bounding box j is represented by the... Represents the square root of the actual bounding box height;

[0028] S0024. Calculate the total loss, specifically as follows:

[0029] L = L obj +L cls +L coord

[0030] Wherein, L represents the total loss.

[0031] Furthermore, the weights of the object detection loss function, category classification loss function, and bounding box coordinate loss function during training are determined by the hyperparameter λ. obj , λ cls , λ coord Adjustments are made to balance the impact of different loss terms.

[0032] Furthermore, in step S002, adjusting the YOLO model based on the calculation results specifically involves adjusting the model parameters based on the difference between the predicted and the true labels during backpropagation to minimize the value of the loss function.

[0033] Furthermore, step S5 is included, which specifically involves performing electrical tests on the three-dimensionally assembled microwave hybrid integrated circuit after assembly.

[0034] Furthermore, the substrate data of the silicon nitride substrate are as follows: thickness = 0.02 mm, diameter = 50 mm, dielectric constant = 7.5, loss tangent = 0.002, conductivity = 1 x 10^-6 S / m, and coefficient of thermal expansion = 3.0 x 10^-6 / ℃.

[0035] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0036] This invention uses the YOLO model for visual feedback in microwave device fabrication, ensuring that the trained deep learning model can accurately predict the position and orientation of the device, thereby improving the accuracy of the three-dimensional assembly process and optimizing the assembly quality and efficiency of microwave devices. Attached Figure Description

[0037] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0038] Figure 1 This is a flowchart of a three-dimensional micro-assembly method for a microwave hybrid integrated circuit according to the present invention. Detailed Implementation

[0039] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the invention, and not all of them. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0041] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0042] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0043] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0044] like Figure 1 As shown, a three-dimensional micro-assembly method for microwave hybrid integrated circuits includes the following steps:

[0045] S1. Using a silicon nitride substrate as the base, the substrate is placed on an assembly template, and the geometric features of the assembly template are used for auxiliary positioning and alignment, followed by micro-welding.

[0046] S2. Microwave devices required for fabricating microwave integrated circuits, wherein the microwave devices include at least a quarter-wavelength microstrip line, dual parallel PIN diodes, a DC bias circuit, and a control switch circuit;

[0047] S3. Fabricate three-dimensional assembled components and packaged components, wherein the three-dimensional assembled components include at least stacked radio frequency components and integrated antenna arrays, and the packaged components include at least QFN packages and BGA packages;

[0048] S4. Precisely assemble the prepared microwave devices, three-dimensional assembled components, and packaging components;

[0049] Specifically, step S4, which involves the precise assembly of microwave devices, includes the following sub-steps:

[0050] S401. During the assembly process, images of the microwave device are captured using a high-resolution camera or microscope, and the captured images are labeled, wherein the image data is labeled according to the location and orientation of the microwave device, and the labels indicate the bounding box and category of the microwave device;

[0051] S403. Input the image into the trained YOLO model. The model detects and locates devices in the image through multi-scale prediction and grid segmentation, and generates predicted bounding boxes and category information.

[0052] Specifically, regarding the use of silicon nitride as the substrate proposed in this embodiment, silicon nitride is a high-performance substrate material with low dielectric constant and loss, making it suitable for high-frequency microwave circuits. Preferably, high-frequency FR-4 boards, high-frequency ceramics, etc., can also be selected as substrate materials. High-frequency FR-4 is a common substrate material suitable for mid-to-low frequency microwave circuits, while high-frequency ceramic materials, such as aluminum oxide (Al2O3) or aluminum alumina nitride (AlN), have low loss and excellent conductivity, making them suitable for high-frequency microwave circuits. The assembly template proposed in this embodiment is an auxiliary structure used to assist in positioning and aligning devices. It can be matched with the specific structure of the substrate or device, thereby helping to achieve precise positioning and assembly at the microscale.

[0053] Furthermore, the micro-welding proposed in this embodiment is a process for connecting tiny devices or components together, typically using tiny solder joints. In the fabrication of microwave hybrid integrated circuits, micro-welding is used to connect different circuit components, ensuring electrical connection and stability between them. For example, one implementation involves using a tiny heat source (e.g., a laser, a soldering pen, etc.) or a micro-area heat source (e.g., focused laser) to heat the metal solder, causing it to melt. In the molten state, the solder establishes a connection between the devices. Placing the molten solder or solder wire at the connection point, due to the surface tension and capillary action of the solder, the solder automatically forms tiny solder joints. Once the solder joints are formed, heating is stopped, and the solder is allowed to cool and solidify.

[0054] Furthermore, regarding the three-dimensional assembled components proposed in this embodiment, stacked radio frequency (RF) components involve stacking multiple microwave devices together to reduce circuit size and optimize performance; integrated antenna arrays integrate multiple microwave antenna components to achieve functions such as beam control; QFN (Quad Flat No-Lead) packages are used for packaging microwave devices, featuring compact size and good heat dissipation; BGA (Ball Grid Array) packages offer high pin density and connectivity, suitable for highly integrated microwave devices. It should be noted that QFN and BGA packages have different advantages in different applications. QFN packages typically have a smaller size, enabling highly integrated and compact layouts of microwave circuits; BGA packages offer high pin density, suitable for complex microwave devices such as RF and microwave communication chips.

[0055] Furthermore, it also includes a pre-step S0, which trains the YOLO model through sampling, specifically including the following sub-steps:

[0056] S001 acquires image samples of microwave devices, extracts sample labels from the images through feature extraction, and generates a training set based on the sample labels;

[0057] S002. Generate a YOLO model, train the YOLO model, calculate the parameter gradients using the object detection loss, class classification loss, and bounding box coordinate loss functions, and adjust the YOLO model based on the calculation results.

[0058] Furthermore, in step S002, calculating the parameter gradient using the object detection loss function, the category classification loss function, and the bounding box coordinate loss function specifically includes the following sub-steps:

[0059] S0021. Calculate the target loss, specifically as follows:

[0060]

[0061] Wherein, the L obj The loss represents the bounding box coordinates, where λ is the bounding box coordinate loss. obj The weights represent the object detection loss, where s represents the number of grid cells, B represents the number of predicted bounding boxes in each grid cell, i represents the index of the grid cell, j represents the index of the bounding box, and IOU represents the weights of the object detection loss. ij The intersection-union ratio (IoU) between the predicted bounding box j and the ground truth bounding box is represented by bx. i The term "by" represents the horizontal offset of the predicted bounding box center relative to the i-th grid cell. i The value represents the longitudinal offset of the predicted bounding box center relative to the i-th grid cell. This indicates an indicator function, which is 1 if the j-th bounding box of the i-th grid cell contains the target, and 0 otherwise.

[0062] S0022. Calculate the category classification loss, specifically as follows:

[0063]

[0064] Wherein, the L cls λ represents the category classification loss. cls The weights represent the class classification loss, where the class i p represents the predicted probability of the category corresponding to the predicted bounding box j. i The probability of the true label representing the category to which the true bounding box belongs;

[0065] S0023. Calculate the frame coordinate loss, specifically:

[0066]

[0067] Wherein, the L coord The loss represents the bounding box coordinates, where λ is the loss. coord The weights representing the loss of the bounding box coordinates, w i The square root of the width of the predicted bounding box j is represented by the... h represents the square root of the width of the actual bounding box. i The square root of the height of the predicted bounding box j is represented by the... Represents the square root of the actual bounding box height;

[0068] S0024. Calculate the total loss, specifically as follows:

[0069] L = L obj +L cls +L coord

[0070] Here, L represents the total loss. Specifically, the object detection loss is used to adjust the model's bounding box predictions to make them closer to the true bounding boxes. The object detection loss function plays a role in the bounding box position prediction, optimizing the model by calculating the positional difference between the predicted bounding box and the true bounding box. The class classification loss is used to adjust the model's classification predictions for different classes. The class classification loss function plays a role in the class prediction part, optimizing the model by calculating the difference between the predicted class probability and the true class label. The box coordinate loss is used to adjust the model's predicted bounding box size and position to make them closer to the true bounding boxes. The box coordinate loss function plays a role in the bounding box size prediction, optimizing the model by calculating the difference between the predicted bounding box size and the true bounding box. By calculating the loss function in each training iteration and optimizing the model parameters through backpropagation based on the total loss, the YOLO model gradually adjusts itself to more accurately predict the position and class of devices, achieving precise assembly of microwave devices.

[0071] Furthermore, as a preferred embodiment of this practice, images are captured in real time during the microwave device assembly process and input into a trained YOLOv3 model. The model detects and locates the devices in the images, generating predicted bounding boxes and category information. This information is then transformed into visual feedback to guide the operator in fine-tuning and assembly.

[0072] Furthermore, the weights of the object detection loss function, category classification loss function, and bounding box coordinate loss function during training are determined by the hyperparameter λ. obj , λ cls , λ coord Adjustments are made to balance the impact of different loss terms.

[0073] Furthermore, in step S002, adjusting the YOLO model based on the calculation results specifically involves adjusting the model parameters based on the difference between the predicted and the true labels during backpropagation to minimize the value of the loss function.

[0074] Furthermore, step S5 is included, which specifically involves performing electrical tests on the three-dimensionally assembled microwave hybrid integrated circuit after assembly.

[0075] Furthermore, the silicon nitride substrate has the following specifications: thickness = 0.02 mm, diameter = 50 mm, dielectric constant = 7.5, loss tangent = 0.002, conductivity = 1 x 10^-6 S / m, and coefficient of thermal expansion = 3.0 x 10^-6 / ℃. Specifically, the substrate thickness affects the characteristics of microwave circuits, such as resonant frequency and impedance; the substrate size determines the overall physical dimensions of the circuit and also affects the characteristic impedance of microwave transmission lines; in high-frequency microwave circuits, the dielectric constant and loss tangent of the substrate directly affect signal propagation speed and coupling efficiency; conductivity: the conductivity of the substrate needs to be considered in aspects such as grounding or RF shielding in microwave circuits; microwave circuits need to operate in environments with temperature variations, and the coefficient of thermal expansion of the substrate ensures the stability of circuit performance.

[0076] Furthermore, as a preferred embodiment of this practice, a cross-validation localization method is proposed to assist the YOLO model in localization. Specifically, the predicted positions of the device are obtained from different viewpoints, and cross-validation is used to ensure the consistency of the predictions. The predicted positions from different viewpoints are compared, and the differences between them are checked. If the predicted positions from multiple viewpoints are spatially consistent, then these predictions can be considered reliable. By eliminating possible erroneous predictions, the accuracy of cross-validation is improved, ensuring the accuracy and stability of the final device assembly process. The specific process principle is as follows: In each iteration of cross-validation, the difference index of the predicted positions from different viewpoints (such as mean square error) is recorded. The difference index of all iterations is statistically analyzed to obtain the overall performance index of consistency from different viewpoints. The difference in predicted positions from different viewpoints is analyzed to observe whether there are consistency problems. If the predicted positions from different viewpoints are relatively consistent, it indicates that the position confirmation during the assembly process is relatively reliable. If there are large differences, the assembly process and model parameters need to be adjusted to improve the accuracy of position confirmation.

[0077] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A three-dimensional micro-assembly method for microwave hybrid integrated circuits, characterized by, Comprise the following steps: S1. Adopting silicon nitride substrate as the base, placing the substrate on the assembly template, assisting positioning and alignment through the geometric features of the assembly template, and micro welding; S2. Preparing microwave devices required by microwave integrated circuits, the microwave devices at least including quarter wavelength microstrip line, double parallel PIN diode, DC bias circuit, control switch circuit; S3. Preparing three-dimensional assembly elements and packaging elements, the three-dimensional assembly elements at least including stacked radio frequency components and integrated antenna array, the packaging elements at least including QFN packaging and BGA packaging; S4. Precise assembly of the prepared microwave devices, three-dimensional assembly elements and packaging elements; Wherein, the step S4, the precise assembly of the microwave devices specifically comprises the following sub-steps: S401. In the assembly process, using a high-resolution camera or a microscope to capture the image of the microwave device, and adding a label to the captured image, wherein the image data is labeled according to the position and orientation of the microwave device, and the label indicates the bounding box and category of the microwave device; S402. Input the image into the trained YOLO model, and the model detects and locates the device in the image through multi-scale prediction and grid segmentation, and generates the predicted bounding box and category information.

2. A three-dimensional micro-assembly method of a microwave hybrid integrated circuit according to claim 1, wherein Also includes pre-step S0, training YOLO model by sampling, specifically comprising the following sub-steps: S001. Collecting image samples of microwave devices, extracting sample labels in the image through feature extraction, and generating a training set through sample labels; S002. Generating YOLO model, training YOLO model, calculating parameter gradient through target detection loss, category classification loss and frame coordinate loss function, and adjusting YOLO model according to the calculation result.

3. A three-dimensional micro-assembly method of a microwave hybrid integrated circuit according to claim 2, wherein In the step S002, the calculation of the parameter gradient through the target detection loss function, the category classification loss function and the frame coordinate loss function specifically comprises the following sub-steps: S0021. Calculate the target loss, specifically: ； wherein the represents a target loss, the represents a weight of a target detection loss, the represents a number of grid cells, the represents a number of predicted bounding boxes in each grid cell, the represents an index of a grid cell, the represents an index of a bounding box, the represents a predicted bounding box represents an intersection over union with a real bounding box, the represents a horizontal direction offset of a center of a predicted bounding box relative to a th grid cell, the represents a vertical direction offset of a center of a predicted bounding box relative to a th grid cell, the represents an indicator function, which is 1 when a th bounding box of a th grid cell contains a target, and 0 otherwise. S0022. Calculate the category classification loss, specifically: ； wherein the denotes a class classification loss, the denotes a weight of the class classification loss, the denotes a predicted bounding box a predicted probability of a corresponding class, the denotes a probability of a real label of a class to which a real bounding box belongs; S0023. Calculate the frame coordinate loss, specifically: ； wherein the represents a bounding box coordinate loss, the represents a weight of the bounding box coordinate loss, the represents a square root of a width of a predicted bounding box , the represents a square root of a width of a true bounding box, the represents a square root of a height of a predicted bounding box, the represents a square root of a height of a true bounding box; S0024. Calculate the total loss, specifically: ； wherein the represents the total loss.

4. A three-dimensional micro-assembly method of a microwave hybrid integrated circuit according to claim 3, wherein The weights of the target detection loss function, the class classification loss function and the box coordinate loss function in the training process are adjusted by hyperparameters 、 、 for balancing the influence of different loss terms.

5. A three-dimensional micro-assembly method of a microwave hybrid integrated circuit according to claim 2, wherein In the step S002, adjusting the YOLO model according to the calculation result specifically refers to adjusting the parameters of the model according to the difference between the prediction and the true label in the back propagation process, and minimizing the value of the loss function.

6. A three-dimensional micro-assembly method for microwave hybrid integrated circuits as claimed in claim 1, characterized in that, Also includes step S5, specifically: after the assembly is completed, the three-dimensional assembled microwave hybrid integrated circuit is electrically tested.

7. A three-dimensional micro-assembly method for microwave hybrid integrated circuits as claimed in claim 1, characterized in that, The base data of the silicon nitride substrate is thickness = 0.02 mm, diameter = 50 mm, dielectric constant = 7.5, loss tangent = 0.002, electrical conductivity = 1 x 10 -6 S / m, thermal expansion coefficient = 3.0 x 10 -6 / °C.

Images