A cloud-fog-edge-end collaborative intelligent management system and method for a wire bonding machine

Abstract

The application discloses a kind of intelligent management systems of cloud-fog-edge-end fusion wire bonding machine, and intelligent management system includes: the equipment layer comprising the multi-source sensor array integrated with wire bonding machine or carried, physical execution unit and microsecond level closed-loop controller, multi-source sensor array is used to collect the real-time multi-physical field data of bonding process, physical execution unit is used to execute bonding operation, microsecond level closed-loop controller is used to respond control instruction and drive physical execution unit;Edge computing layer is deployed in wire bonding machine, and microsecond level communication is carried out with equipment layer;Fog computing layer is deployed in local network, and is used to carry out factory area level cooperation to the data and model of at least two edge computing layers;Cloud computing layer is deployed in remote data center, and is used to carry out global aggregation and analysis to the data and model of at least one fog computing layer, and provide cross-domain knowledge service and whole life cycle management, it is conducive to providing high reliability, adaptive bonding environment for semiconductor package.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial internet, and in particular to a lead bonding machine intelligent management system and method fusing cloud-fog-edge-end cooperation. BACKGROUND

[0002] With the development of the semiconductor industry towards higher integration and smaller size, lead bonding, as the core process of chip packaging, directly affects the performance and reliability of electronic products in terms of quality and efficiency. Compared with traditional general industrial control systems, lead bonding process requires micron-level positioning accuracy, millisecond-level action response, and simultaneous control of mechanical pressure, ultrasonic power, temperature field and other multi-physical field parameters, with precision requirements far exceeding conventional industrial scenarios.

[0003] However, the current lead bonding industry faces many technical challenges. First, quality control is difficult. Due to bonding defects, global packaging plants have reported annual scrap chip values exceeding hundreds of millions of dollars. The lead bonding process involves the coupling of mechanical pressure, ultrasonic vibration, temperature field and other multi-physical fields. Traditional monitoring systems only analyze single physical quantities, cannot effectively process multi-physical field coupling data, and cannot accurately predict process results and equipment status. Thousands of process data (such as solder deformation, wire arc curvature, and gold ball diameter) are generated by the equipment every second, but a large amount of data is not effectively utilized.

[0004] Second, real-time control is insufficient. Traditional industrial cloud platforms have a data sampling frequency of seconds or more, which cannot meet the real-time data acquisition needs of millisecond-level bonding actions in the lead bonding process, resulting in the inability to accurately and timely capture and online warning of key process parameters such as solder deformation and split knife micro-cracks. Most cloud platforms use centralized data processing mode, with device data transmitted to the cloud computing layer for analysis and feedback, resulting in large delays and inability to meet real-time control needs.

[0005] Third, the data island problem is serious. Existing lead bonding systems use mutually incompatible private communication protocols, forming a barrier to cross-platform data interoperation. Centralized data processing architecture requires raw process parameters to be uploaded to the cloud for analysis, resulting in a risk of leakage of enterprise core process details. Due to privacy concerns, production bases cannot share bonding experience data, process optimization strategies are isolated, and global collaborative optimization is difficult to achieve.

[0006] Fourth, the intelligent level is low. Process parameter optimization highly depends on on-site operation of engineers, and key constraint conditions such as material thermal expansion coefficient and equipment positioning accuracy are not established in a digital correlation model. The optimal parameter combination of similar equipment under similar working conditions cannot be reused across production lines, resulting in a significant extension of the debugging period. Personnel flow causes a fault in the process of experience inheritance, and new engineers highly depend on trial and error when debugging equipment, which is low in efficiency. The same type of bonder produces parameter drift due to assembly tolerance and component wear, and the traditional scheme cannot perform adaptive compensation.

[0007] Fifth, equipment maintenance is lagging. Equipment maintenance relies on fixed period maintenance or after-the-fact maintenance, and there is no early warning mechanism for gradual failure such as micro-crack propagation of a cleaving knife and performance degradation of a transducer. The traditional monitoring system is limited by data processing capacity and cannot build a failure prediction model based on multi-source sensor time series data. Sudden failure leads to unplanned shutdown of the production line, and the time-consuming maintenance significantly affects the delivery cycle.

[0008] There are two typical technical routes in the prior art: one is based on a general industrial Internet of Things architecture, which aggregates device data through an edge gateway and uploads it to a cloud computing layer for processing. Although this can achieve basic state monitoring, it cannot meet the real-time response requirements of motion control and energy regulation in the bonding process due to network transmission delay between the cloud computing layer and the device. The second is to use a local embedded computing unit in the device, which ensures the real-time performance of the core control loop, but is limited by hardware resources and cannot support intelligent algorithms that require high computing power. Both solutions have structural defects: the cloud computing layer solution causes real-time control functions to fail due to a long control link, and the local solution weakens intelligent decision-making capabilities due to computing power bottlenecks. Therefore, there is an urgent need to develop an intelligent management system and method for wire bonding machines to increase data processing accuracy and bonding quality during wire bonding processing. SUMMARY

[0009] One object of the present application is to provide an intelligent management system and method for wire bonding machines that integrates cloud-fog-edge-end collaboration to solve the intelligent problem under the triple technical constraints of micro-scale, multi-physical field coupling and high reliability specific to the field of semiconductor wire bonding.

[0010] One object of the present application is to provide an intelligent management system and method for wire bonding machines that integrates cloud-fog-edge-end collaboration to solve the intelligent problem under the triple technical constraints of micro-scale, multi-physical field coupling and high reliability specific to the field of semiconductor wire bonding.

[0011] To achieve the above object, the application provides a cloud-fog-edge-end collaborative intelligent management system for a wire bonding machine, comprising: a device layer comprising a multi-source sensor array integrated with or mounted on the wire bonding machine, a physical execution unit, and a microsecond-level closed-loop controller, the multi-source sensor array being configured to collect real-time multi-physical field data of a bonding process, the physical execution unit being configured to perform a bonding operation, and the microsecond-level closed-loop controller being configured to respond to a control instruction and drive the physical execution unit; an edge computing layer deployed on the wire bonding machine and in microsecond-level communication with the device layer, the edge computing layer comprising: an edge data processing and modeling module configured to process the real-time multi-physical field data and build a local model representing the bonding process; a real-time control and decision module configured to make a millisecond-level intelligent decision based on the local model and generate a real-time control instruction to be sent to the microsecond-level closed-loop controller; and an edge learning module configured to train a model locally using classified process data and upload only model parameter updates; a fog computing layer deployed on a local network and configured to perform factory-level collaboration on data and models of at least two edge computing layers; and a cloud computing layer deployed on a remote data center and configured to perform global aggregation and analysis on data and models of at least one fog computing layer and provide cross-domain knowledge services and full-life-cycle management.

[0012] In some embodiments, the real-time control and decision module of the edge computing layer further comprises a bonding process emergency control sub-module configured to monitor abnormal conditions in a bonding process in real time, perform microsecond-level emergency shutdown at the device layer when a preset abnormal condition is detected, and automatically backtrack and store multi-source sensor data before the abnormal condition occurs for traceability analysis.

[0013] In some embodiments, the edge data processing and modeling module of the edge computing layer comprises a multi-physical field coupling engine sub-module configured to fuse the multi-physical field data through a high-precision time synchronization mechanism and dynamically calculate real-time prediction values of a key indicator of a bonding point quality based on a physical information neural network model to form a closed-loop control for adaptive adjustment of process parameters.

[0014] In some embodiments, the real-time control and decision module of the edge computing layer comprises a vision-operations linkage submodule, which, based on a solder feature-operations parameter correlation model, predicts the quality risk of the next solder according to the image feature change trend of the current and historical solders, and generates an operations parameter pre-adjustment instruction before defects occur to achieve active prevention of bonding quality problems; the edge computing layer is also preconfigured with a set of baseline parameters for quick start when a new device is connected; the fog computing layer maintains a debugging parameter pool that aggregates optimized process parameters of devices that have been stably running in the factory; when a new device is connected, the system automatically migrates the optimized process parameters that best match the initial operating conditions of the new device from the debugging parameter pool and compensates for fine tuning based on the unique bias of the new device to achieve quick and high-precision debugging.

[0015] In some embodiments, the fog computing layer comprises a factory-level intelligent application module, which comprises a rapid process design engine submodule that automatically generates a recommended process parameter set by inputting chip design parameters, combining a process rule library and a machine learning model; and uses digital twin technology to pre-visualize the bonding process in a virtual environment to detect and avoid physical risks, thereby optimizing the recommended process parameter set.

[0016] In some embodiments, the system further comprises a flexible activation mechanism configured to dynamically adjust the system architecture according to the number of wire bonding machines or the process complexity; when a first predetermined condition is met, the edge computing layer is directly connected to the cloud computing layer; when a second predetermined condition is met, the fog computing layer is activated and deployed to perform factory-level collaboration between the edge computing layer and the cloud computing layer; the data flow between the levels of the system uses a layered optimized communication protocol stack: the industrial real-time Ethernet or high-speed field bus protocol is used between the device layer and the edge computing layer; the lightweight message queue protocol or OPC UA PubSub mode is used between the edge computing layer and the fog computing layer; the enterprise-level message protocol or HTTPS RESTful API is used between the fog computing layer and the cloud computing layer.

[0017] In some embodiments, the cloud computing layer comprises a global process knowledge and optimization module configured to: build and maintain a material-equipment-process three-dimensional dynamic knowledge graph; aggregate model parameters uploaded by each of the fog computing layers to continuously evolve the knowledge graph without accessing original classified process data through a federated learning architecture; and achieve rapid adaptation of process parameters in new material or new equipment scenarios through transfer learning based on the knowledge graph; the cloud computing layer further comprises an equipment life cycle management module configured to: build a device health and process stability joint prediction model, dynamically associate the predicted health status of key components of the device with process parameters, and issue adaptive process parameter compensation instructions; and build a performance degradation analysis and design feedback model to feed back anonymized failure and process data of multiple factories to the equipment research and development design end to form an optimization closed loop of "use-design".

[0018] To achieve the above purpose, the application provides a lead bonding machine intelligent management method integrating cloud-fog-edge-end cooperation, comprising the following steps: at the device layer, real-time acquisition of multi-physical field data of the bonding process is performed by using a multi-source sensor array, and control instructions from the edge computing layer are executed by a microsecond-level closed-loop controller; at the edge computing layer, real-time processing and modeling of the multi-physical field data are performed, millisecond-level intelligent decision making is performed based on a local model to generate real-time control instructions, and only model parameter updates are uploaded after the model is trained using local classified process data; at the fog computing layer, model parameter updates from at least two of the edge computing layers are aggregated and cooperatively optimized at the plant level; at the cloud computing layer, the aggregated model from at least one of the fog computing layers is globally trained and analyzed, and cross-domain knowledge services and full life cycle management are provided.

[0019] In some embodiments, the step of making intelligent decisions at the edge computing layer further comprises: real-time monitoring of abnormal conditions during the bonding process, triggering microsecond-level emergency shutdown of the device layer when a preset serious abnormality is detected, and automatically backtracking and storing multi-source sensor data before the abnormality occurs; the method further comprises a zero-sample debugging step: when a new device is connected, a set of baseline parameters preloaded in the edge computing layer is first loaded to achieve rapid startup; then, optimized process parameters matching the initial working condition of the new device are intelligently retrieved and migrated from a debugging parameter pool deployed in the fog computing layer; finally, the migrated parameters are fine-tuned in combination with the unique bias of the new device to complete high-precision debugging.

[0020] In some embodiments, the step of collaborative optimization at the fog computing layer comprises a fast process design step: receiving chip design parameters as input, automatically generating a preliminary recommended process parameter set; using digital twin technology to pre-visualize the bonding process using the recommended process parameter set in a virtual environment; detecting and correcting potential physical risks according to the pre-visualization results to output an optimized process parameter set; the step of global training and analysis at the cloud computing layer comprises the step of building and applying a knowledge graph: through a federated learning architecture, aggregating model parameters uploaded by each fog computing layer to continuously evolve a material-equipment-process three-dimensional dynamic knowledge graph without touching the original confidential process data; when new materials or new equipment are introduced, relationship reasoning and similarity matching are performed based on the knowledge graph, and initial process parameters for rapid adaptation are generated for the edge computing layer based on transfer learning.

[0021] Compared with the prior art, the beneficial effects of the present application are:

[0022] (1) The present application fuses a cloud-fog-edge-end collaborative wire bonding machine intelligent management system, sets different levels for the characteristics of wire bonding in the four levels, such as setting a special sensor for wire bonding in the equipment level, improving the sampling period of the sensor, and increasing the sensor data measured by the sensor. The edge computing layer with multiple modules is used to process the real-time multi-physical field data and make millisecond-level intelligent decisions based on the local model. The fog computing layer integrates the data and models of multiple edge computing layers for factory-level collaboration, the cloud computing layer performs global aggregation and analysis, and provides cross-domain knowledge services and full life cycle management, thereby improving the data processing accuracy and processing speed of the data processing of the wire bonding machine intelligent management system, and further improving the security of data processing. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 An architecture diagram of an exemplary cloud-fog-edge-end collaborative wire bonding machine intelligent management system in the present application.

[0024] Figure 2 An architecture diagram of an exemplary equipment layer in the present application.

[0025] Figure 3 A workflow diagram of an exemplary edge computing layer in the present application.

[0026] Figure 4 An architecture diagram of an exemplary edge computing layer in the present application.

[0027] Figure 5 A workflow diagram of an exemplary elastic activation mechanism of a fog computing layer in the present application.

[0028] Figure 6Process driven layered architecture diagram of an exemplary wire bonding machine intelligent management system in the present application.

[0029] Figure 7 Architecture diagram of an exemplary fog computing layer in the present application.

[0030] Figure 8 Multi-protocol compatible architecture diagram of an exemplary wire bonding machine intelligent management system with three-layer decoupling design in the present application.

[0031] Figure 9A Construction process of an exemplary material-equipment-process three-dimensional dynamic knowledge graph analysis model in the present application.

[0032] Figure 9B An exemplary material-equipment-process three-dimensional dynamic knowledge graph analysis model in the present application.

[0033] Figure 10 Workflow diagram of an exemplary process parameter transfer learning sub-module in the present application.

[0034] Figure 11 Workflow diagram of an exemplary global process parameter intelligent optimization sub-module in the present application.

[0035] Figure 12 Workflow diagram of an exemplary health degree and stability prediction sub-module in the present application.

[0036] Figure 13 Workflow diagram of an exemplary performance degradation analysis and design feedback sub-module in the present application.

[0037] Figure 14 Workflow diagram of an exemplary after-sales resource and cost optimization sub-module in the present application.

[0038] Figure 15 Workflow diagram of an exemplary equipment health degree-process stability joint prediction model in the present application.

[0039] Figure 16 Workflow diagram of an exemplary global quality and process control module in the present application.

[0040] Figure 17 Workflow diagram of an exemplary cross-facility production and energy consumption management module in the present application.

[0041] Figure 18 Framework diagram of an exemplary wire bonding machine intelligent management system in the present application.

[0042] Figure 19 Workflow diagram of an exemplary real-time data acquisition module in the present application.

[0043] Figure 20 Workflow diagram for an exemplary device status monitoring module in the present application.

[0044] Figure 21 Workflow diagram for an exemplary remote control module in the present application.

[0045] Figure 22 Workflow diagram for an exemplary data analysis module in the present application.

[0046] Figure 23 Workflow diagram for an exemplary alarm system module in the present application.

[0047] Figure 24 Workflow diagram for an exemplary process parameter management module in the present application.

[0048] Figure 25 Workflow diagram for an exemplary historical data query module in the present application.

[0049] Figure 26 Workflow diagram for an exemplary user management module in the present application.

[0050] Figure 27 Architecture diagram for an exemplary cloud computing layer in the present application.

[0051] In the figure: 1, device layer; 11, multi-source sensor array; 111, mechanical motion sensing unit; 112, ultrasonic energy sensing unit; 113, temperature field sensing unit; 114, visual sensing unit; 115, environment detection unit; 12, microsecond-level closed-loop controller; 13, physical execution unit; 131, XY platform driving mechanism; 132, Z-axis actuator; 133, ultrasonic transducer; 134, heating mechanism; 2, edge computing layer; 21, edge data processing and modeling module; 211, multi-source data real-time processing submodule; 212, multi-physical field coupling engine submodule; 213, lightweight AI quality inspection submodule; 22, real-time control and decision module; 221, visual-operation linkage submodule; 222, bonding process emergency control submodule; 23, edge learning module; 3, fog computing layer; 31, fog layer data and communication management module; 311, plant-level data aggregation and processing submodule; 312, multi-protocol gateway submodule; 313, security agent submodule; 32, plant-level collaborative learning and model management module; 33, factory-level intelligent application module; 331, rapid process design engine submodule; 332, production line-level quality and process control submodule; 333, production line-level production and energy consumption management submodule; 4, cloud computing layer; 41, global federated learning aggregation module; 42, global process knowledge and optimization module; 421, knowledge graph construction and management submodule; 422, process parameter transfer learning submodule; 423, global process parameter intelligent optimization submodule; 43, equipment life cycle management module; 431, health degree and stability prediction submodule; 432, performance degradation analysis and design feedback submodule; 433, after-sales resource and cost optimization submodule; 44, global quality and process control module; 45, cross-plant production and energy consumption management module. DETAILED DESCRIPTION

[0052] Hereinafter, the present application will be further described in conjunction with specific embodiments, and it should be noted that the embodiments described below or technical features between the embodiments can be combined in any manner to form new embodiments without conflict.

[0053] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence.

[0054] The terms "include" and "have" and any variations thereof in the specification and claims of the present application are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but includes other steps or units not clearly listed or inherent to the process, method, product or device.

[0055] The application provides a lead bonding machine intelligent management system fusing cloud-fog-edge-end cooperation, which is used for solving the contradiction between real-time control and global optimization in a lead bonding process, and constructing a cooperative architecture with layered time scales, as shown in Figure 1 The cooperative architecture includes a device layer 1 with a response time of microsecond level, an edge computing layer 2 with a response time of millisecond level, a fog computing layer 3 with a response time of hundred-millisecond level and a cloud computing layer 4 with a response time of second level, and forms a whole life cycle closed loop from perception and control to optimization and redesign.

[0056]

Device layer 1

[0057] As shown in Figure 2 It can be known that the device layer 1 is the physical basis and execution terminal of the system, that is, the lead bonding machine body and its accessory units. The accessory units mainly include a multi-source sensor array 11, a microsecond-level closed-loop controller 12 and a physical execution unit 13.

[0058] In a specific embodiment, since the lead bonding machine generates thousands of high-frequency data per second, at this time, the multi-source sensor array 11 is used to comprehensively and frequently collect real-time multi-physical field data in the bonding process, specifically including a mechanical motion sensing unit 111, an ultrasonic energy sensing unit 112, a temperature field sensing unit 113, a visual sensing unit 114 and an environment detection unit 115, so as to monitor the original physical signals in the whole lead bonding process and realize real-time perception and data collection of the running state of the device layer 1.

[0059] The mechanical motion sensing unit 111 can be a high-precision grating ruler or encoder installed on the XY platform and Z axis, which is used to collect real-time displacement, speed and acceleration information of micron level and monitor the motion trajectory of the wedge.

[0060] The ultrasonic energy sensing unit 112 is used to monitor the dynamic waveform of the power, frequency and amplitude output by the ultrasonic transducer in real time, and the sampling frequency can reach more than 10 kHz.

[0061] The temperature field sensing unit 113 can be a non-contact infrared thermal imager or a thermocouple, which is used to collect the transient temperature distribution of the bonding area.

[0062] The visual sensing unit 114 can be a high-resolution and high-frame-rate (for example, ≥30 fps) industrial camera equipped with a high-power lens and a professional imaging system, which is used to collect pad positioning images, solder ball appearance images and solder joint quality images after bonding.

[0063] The environment detection unit 115 is used to monitor the environmental parameters such as temperature and humidity of the workshop where the device is located, and provide a basis for process compensation.

[0064] In a specific embodiment, the physical execution unit 13 is a mechanism that implements the bonding action, responsible for executing instructions from the controller, including specifically an XY platform driving mechanism 131, a Z-axis actuator 132, an ultrasonic transducer 133, and a heating mechanism 134.

[0065] The XY platform driving mechanism 131 can be driven by a servo motor or a linear motor, used to carry the chip or substrate for fast and accurate positioning in the horizontal plane.

[0066] The Z-axis actuator 132 is usually driven by a voice coil motor, used to control the up-and-down movement of the wedge in the vertical direction and apply precise bonding pressure.

[0067] The ultrasonic transducer 133 is used to apply controllable ultrasonic energy at the bonding interface to promote the diffusion and bonding of metal atoms.

[0068] The heating mechanism 134 is used to preheat the substrate to achieve the appropriate bonding process temperature.

[0069] It should be understood that the microsecond-level closed-loop controller 12 is the core of the device layer 1 control, usually composed of an FPGA (Field Programmable Gate Array) or DSP (Digital Signal Processor) with a main frequency not less than 500MHz. It is directly connected with the multi-source sensor array 11 and the physical execution unit 13, forming an ultra-low delay hardware control loop with a response time ≤100μs. Its main functions are: receiving and analyzing control instructions (such as motion trajectory correction, process parameter fine-tuning) from the upper layer (edge computing layer 2), combining real-time feedback from multiple sensing units in the multi-source sensor array 11, accurately driving the physical execution unit 13 to complete the action, achieving motion trajectory tracking, multi-physical field coordination, and basic safety protection. Not only does it form a two-way data interaction channel between the physical execution unit 13 and the microsecond-level closed-loop controller 12, but it also provides good data credibility for data optimization for the fog computing layer 3 and the cloud computing layer 4.

[0070]

Edge Computing Layer 2

[0071] The edge computing layer 2 is deployed in the embedded industrial computer or edge artificial intelligence box of each wire bonding machine, and is the core of realizing real-time intelligent decision-making and local data processing. It communicates with the device layer 1 at the microsecond level to ensure deterministic transmission of control instructions.

[0072] As Figure 3As shown, the edge computing layer 2 can implement: (1) real-time preprocessing of data collected by the device layer 1, including data denoising, format normalization, feature extraction; (2) automatic correction of pad offset, system motion control, etc. based on the feedback of the visual sensor, realizing real-time monitoring of the wire bonding machine; (3) during the force / position switching process, when there is no force sensor to measure the bonding pressure, the lightweight artificial intelligence constructed by the edge computing layer 2 can estimate the bonding contact force size through a large amount of data and laboratory calibration data, and determine the force / position switching condition; (4) set up an edge federated learning node, classify the data based on the data leakage risk level (such as parameters related to product yield are sensitive data, and device running time is non-sensitive data) or user-defined rules, train the model locally for sensitive data, and only upload parameter update information to the fog computing layer 3, and non-sensitive data can be directly uploaded to the fog computing layer 3; (5) synchronize the construction of multi-sensor fusion for time alignment, construct a multi-physical field coupling engine, so as to realize the bonding process dynamic optimization module, develop a microsecond-level abnormal detection algorithm to realize 10ms process parameter adaptive adjustment, for example, temperature compensation value when using different material bonding wires, and run a lightweight YOLOvx model for real-time quality inspection of solder ball deformation; (5) generate real-time control decisions and issue them to the device layer 1 for execution.

[0073] It is worth mentioning that the main reason for using federated learning is that chip packaging process parameters are enterprise core secrets, and federated learning can avoid the leakage of raw data. In addition, the data distribution of different manufacturer devices (such as copper wire bonding machines and gold wire bonding machines) is quite different, and the local training of federated learning can be compatible with heterogeneous data.

[0074] As shown in FIG. 2, the edge computing layer 2 is mainly composed of an edge data processing and modeling module 21, a real-time control and decision module 22, and an edge learning module 23. Figure 4 As can be seen, the edge computing layer 2 mainly includes an edge data processing and modeling module 21, a real-time control and decision module 22, and an edge learning module 23. The edge data processing and modeling module 21, as the core of the edge computing layer 2, is mainly used to convert the massive raw sensor data of the device layer 1 into structured and information-rich feature data streams, and on this basis, to construct a localized model for real-time analysis and prediction. This module provides high-quality input for the real-time control and decision module 22 and is the prerequisite for achieving millisecond-level intelligent decision-making; the real-time control and decision module 22 is the "decision-making brain" of the edge computing layer 2, responsible for making millisecond-level intelligent decisions based on the local model generated by the edge data processing and modeling module 21 and real-time data, and generating real-time control instructions to issue to the microsecond-level closed-loop controller 12 of the device layer 1. This module ensures that the wire bonding process is fast and accurate.

[0075] The edge data processing and modeling module 21 includes a multi-source data real-time processing sub-module 211, a multi-physical field coupling engine sub-module 212, and a lightweight AI quality inspection sub-module 213.

[0076] The multi-source data real-time processing submodule 211 is responsible for receiving massive raw data from the device layer 1, performing preliminary filtering, noise reduction, dimension normalization, format conversion, and microsecond-level time stamp alignment, and other preprocessing, to provide high-quality data foundation for the upper layer analysis.

[0077] The multi-physical field coupling engine submodule 212 receives the time-aligned, feature-rich multi-dimensional data stream processed by the multi-source data real-time processing submodule 211, aiming to reveal the complex coupling relationship between various physical fields in the wire bonding process. Specifically, to solve the problem that traditional technologies cannot analyze the multi-physical field coupling relationship, the multi-physical field coupling engine submodule 212 fuses feature data from multiple physical fields such as mechanics, sound, heat, and vision in real time through a high-precision time synchronization mechanism. In a preferred embodiment, the multi-physical field coupling engine submodule 212 is based on a lightweight physical information neural network (PINN) model. This model not only learns the correlation from the data, but also embeds the physical equations of the bonding process (such as the heat conduction equation and the vibration equation) as regularization terms or constraints. This enables the engine to dynamically and accurately calculate and output real-time prediction values of key indicators of the solder joint quality (such as solder joint deformation, ball diameter, and interface bonding strength) within ≤5 ms. Based on the real-time comparison of this prediction value with the process threshold, the system can immediately generate fine-tuning instructions for ultrasonic power or bonding pressure, forming a millisecond-level response process parameter self-adaptive adjustment closed loop, and realizing online quality intervention.

[0078] The lightweight AI quality inspection submodule 213 performs real-time online quality detection using machine vision. To achieve efficient inference on edge devices with limited computing power, this module runs a lightweight AI model that has been deeply optimized (e.g., YOLOvx, or lightweight Vision Transformer, or MobileNetV3-SSD). Specifically, the lightweight AI quality inspection submodule 213 performs real-time quality detection on the solder joint images collected by the visual sensing unit 114, determines whether the solder joint has morphological defects (such as cracks, offsets, and size abnormalities) within 10 ms, and realizes 100% online detection of data.

[0079] The real-time control and decision module 22 includes a visual-operations control linkage submodule 221 and a bonding process emergency control submodule 222.

[0080] To solve the pain point of yield loss caused by the passive adjustment of operation control parameters after the occurrence of solder joint quality defects in existing technologies, and to address the risk of insufficient bonding accuracy due to the lack of collaborative control when the visual, operation control, and structural subsystems work independently, the visual-operations control linkage submodule 221 aims to improve bonding quality by building a deep collaboration between visual perception and motion control to intervene before defects occur.

[0081] Specifically, the visual-operation linkage submodule 221 embeds a "solder joint feature-operation parameter association model" in the edge computing layer 2. The model is trained based on historical process data of wire bonding, and its input items include but are not limited to: (1) solder joint core feature parameters, collected by the visual sensing unit 114 and extracted by the edge data processing and modeling module 21, such as ball diameter, wire arc height, bonding interface flatness, etc.; (2) current operation control parameters, such as bonding dwell time, operation control moving speed, ultrasonic power output value, etc.; (3) material property parameters, such as wire material (gold wire / copper wire), hardness, ductility, etc. The output item is the dynamic mapping relationship between "solder joint feature change trend-operation parameter adjustment threshold".

[0082] During the training process, the model continuously corrects the mapping logic through iterative optimization algorithms such as gradient descent, ensuring that when a certain solder joint feature shows a trend close to the process threshold boundary, the corresponding operation control parameter adjustment direction and value range can be accurately output. For example, when the ball diameter feature data shows a "continuous trend towards the lower limit", the model can output an adjustment suggestion of "extending the bonding dwell time by 0.1ms-0.3ms". The model supports dynamic updating of parameter weights according to different bonding scenarios (gold wire / copper wire bonding, different chip package specifications).

[0083] In the production process, the operation control parameter pre-adjustment works according to the following process:

[0084] (1) High-frequency acquisition: the visual sensing unit 114 acquires feature data of each completed solder joint in real time through high frame rate (for example, sampling frequency ≥100fps);

[0085] (2) Trend analysis and risk prediction: after completing a solder joint bonding, the feature data of the solder joint is immediately input into the "solder joint feature-operation parameter association model". The model not only analyzes the state of a single solder joint, but more importantly, analyzes the historical feature change law of multiple consecutive solder joints. For example, the model monitors that the ball diameters of three consecutive solder joints are 25μm, 24.5μm, and 24μm, showing a clear trend of gradually approaching the 23μm process lower limit;

[0086] (3) Generate pre-adjustment instructions for early intervention: based on the above trend, the model predicts the quality risk of the next solder joint. For example, when the model monitors that the ball diameter shows a clear trend of gradually approaching the 23μm process lower limit, it predicts that the ball diameter of the next solder joint is likely to be below the process lower limit, and there is a risk of false bonding, and immediately generates corresponding operation control parameter pre-adjustment instructions;

[0087] (4) Early intervention: The instruction is directly transmitted to the vision- control linkage submodule 221, so that the relevant parameters are adjusted in advance before the next solder joint bonding is performed. For example, when it is predicted that the small ball diameter is caused by too fast control speed and insufficient bonding time, the vision-control linkage submodule 221 automatically increases the bonding dwell time from 1.0 ms to 1.2 ms while keeping other parameters such as ultrasonic power, bonding pressure, etc. stable, so as to realize “intervention before defect occurs”, which is fundamentally different from the passive mode of traditional schemes that wait for defects to occur before adjusting.

[0088] For complex bonding scenarios such as ball grid array packaging (BGA packaging) and stress-sensitive pads, edge computing layer 2 can link device layer 1 to form a “vision-control-structure” three-in-one control, and the specific mechanism is as follows:

[0089] (1) Vision guidance: The vision sensing unit 114 of device layer 1 collects the pad array data of BGA packaging through 3D vision scanning technology (such as laser triangulation), generates a 3D pad distribution map containing pad position, pitch and edge stress concentration area, and marks the priority bonding area and the avoidance area on the 3D pad distribution map. The edge stress concentration area, also known as the avoidance area, is an area within 5 μm from the edge of the pad with a stress value 30% higher than that of the central area of the pad.

[0090] (2) Control execution: After receiving the 3D map, the vision-control linkage submodule 221 generates an optimal bonding path through path planning algorithm (such as A algorithm), which actively avoids the avoidance area located at the edge of the pad and optimizes the bonding sequence between the priority bonding areas in adjacent pads (such as progressing from the center of the array to the edge), to reduce the impact of device vibration on the bonded pads.

[0091] (3) Structure stability: The vision-control linkage submodule 221 synchronously transmits path variation information (such as the number of path inflection points or the moving distance) to the physical execution unit 13 (such as the clamp control mechanism) of device layer 1 while issuing the path planning instruction. The clamp control mechanism dynamically adjusts the clamping force of the clamp based on the information (for example, when there are many path inflection points, the clamp clamping force is increased from 50 N to 60 N to reduce the risk of large amplitude vibration caused by device movement), to suppress the risk of device vibration caused by control path changes reducing the precision of the pads, and ultimately achieve double improvement of bonding quality and efficiency in complex scenarios.

[0092] For the high-risk characteristics of “broken bond, virtual soldering leads to chip scrap” in the wire bonding process, the bonding process emergency control submodule 222 builds a set of emergency and fault tolerance mechanisms that are different from general edge control and are exclusive to the bonding process, to ensure high reliability of the production process.

[0093] Firstly, the bonding process emergency control submodule 222 can realize process abnormality emergency interruption and backtracking, so as to cope with sudden abnormalities that seriously threaten product quality or equipment safety. The specific work flow includes:

[0094] (1) Real-time monitoring of key abnormal conditions in the bonding process, such as sudden drop in ultrasonic power, solder joint deformation exceeding process threshold, abnormal Z-axis motion of the wedge, etc.

[0095] (2) When the above-mentioned preset serious abnormality is detected, the bonding process emergency control submodule 222 first performs instantaneous evaluation of the abnormality risk level through the built-in “process risk priority determination logic”. Based on the evaluation result, if it is determined as a high-risk event, the microsecond-level emergency shutdown operation of the device layer 1 physical execution unit 13 is immediately triggered to prevent defect expansion or equipment damage.

[0096] (3) At the same time of triggering the shutdown, the bonding process emergency control submodule 222 automatically backtracks and extracts, stores various data transmitted by the multi-source sensor array 11 within 10 ms before the abnormality occurs, including but not limited to ultrasonic energy change trend, bonding area temperature curve, wedge motion trajectory, etc., to generate a process abnormality traceability report containing the changes of key parameters before and after the abnormal condition occurs. This provides key basis for engineers to quickly and accurately locate the problem source, which is much better than the traditional solution of only performing a single alarm action.

[0097] Secondly, the bonding process emergency control submodule 222 can allow bonding process fault-tolerant algorithm, which aims to intelligently handle some typical slight process deviations, avoid unnecessary shutdown, and improve production continuity. The work flow includes:

[0098] (1) The bonding process fault-tolerant algorithm can identify typical slight abnormal conditions generated in the bonding process.

[0099] (2) On the basis of conventional position correction, such as slight offset phenomenon of the pad during gold wire bonding, the bonding process fault-tolerant algorithm can combine the characteristics of the bonding material (such as the ductility of gold wire) and the current process parameters (such as ultrasonic power), and through the built-in model, it can calculate and output a multi-parameter coordinated adjustment compensation strategy in real time, for example, for a slight pad offset phenomenon, the bonding process fault-tolerant algorithm can output an accurate pad position offset compensation amount and an accurate ultrasonic power fine tuning value at the same time.

[0100] It is worth mentioning that this bonding process fault-tolerant algorithm using multi-parameter coordination adjustment reduces the risk of over-adjustment caused by a single parameter (such as only adjusting the position) that may trigger new process defects (such as excessive wire stress), enhances the fault tolerance of the general control scheme in the edge computing layer 2 to the bonding process, effectively reduces the chip rejection rate under high-risk working conditions, and improves the robustness of the overall wire bonding system.

[0101] The edge learning module 23 is a core component of the distributed federated learning architecture, aiming to realize the collaborative training and evolution of the model under the premise of ensuring the absolute security of the core process data. Its functions are mainly realized through the following steps:

[0102] (a) Data sensitivity classification: The edge learning module 23 classifies the data processed by the edge data processing and modeling module 21 according to pre-set rules or user-defined strategies. For example, process parameters directly related to product yield, such as specific material ultrasonic power curve, bonding pressure, bonding time, and high-resolution solder joint images, are classified as classified process data (i.e. sensitive data), while device runtime, environmental temperature and humidity, and non-critical component vibration data are classified as non-classified data;

[0103] (b) Local model training and fine-tuning: The edge learning module 23 uses the classified process data stored locally to train or periodically fine-tune various intelligent models deployed in the edge computing layer 2. For example, the model of the lightweight AI quality inspection sub-module 213 can be optimized using local solder joint images and process parameter data, or the associated model in the vision-operations linkage sub-module 221 can be iterated. This localized training method not only ensures that sensitive data does not leave the device, but also enables the model to be highly adaptable to the unique working conditions of the current device (such as parameter drift due to component wear), effectively addressing the problem of large data distribution differences between different devices (such as copper wire bonding machines and gold wire bonding machines);

[0104] (c) Secure parameter upload: After completing a round of local training, the edge learning module 23 only extracts the updated model parameters (such as neural network weights and biases) or gradient information. These parameter information is encrypted before uploading, and then sent to the factory-level collaborative learning and model management module 32 in the fog computing layer 3 in the form of encrypted gradients. This "parameter-only, non-data" upload mechanism fundamentally eliminates the risk of original process data leakage, while significantly reducing network transmission bandwidth occupancy;

[0105] (d) Model receiving and updating: the edge learning module 23 is configured to receive the aggregated and optimized model parameters issued from the upper layer (fog computing layer 3 or cloud computing layer 4). After receiving the new global or factory-level model parameters, the edge learning module 23 updates the local model with the new model parameters, so that the local model is fused with the model parameters of multiple devices or even multiple factories, and the collaborative intelligence is realized under the premise of protecting privacy.

[0106]

Fog computing layer 3

[0107] The fog computing layer 3 is usually deployed in the local server or network device of the factory, and serves as a bridge between the edge computing layer 2 and the cloud computing layer 4, and undertakes the task of regional data aggregation and collaborative optimization at the factory level, and reduces the cross-device collaborative decision delay from seconds to 100 milliseconds.

[0108] The fog computing layer 3 can achieve: (1) regional collaboration and access, serving as a bridge between the edge computing layer 2 and the cloud computing layer 4, realizing cross-device data and model collaboration of multiple heterogeneous devices in the factory, and breaking the data island phenomenon through a multi-protocol gateway; (2) local federal aggregation, aggregating model parameter updates from multiple edge computing layers 2, generating a more optimal factory-level model under the premise of protecting data privacy; (3) carrying high-order intelligent applications, including a rapid process design engine for new product import, a production line-level quality process control and production energy consumption management; (4) dynamically deploying according to the production scale through an elastic activation mechanism, and providing key parameter set support for "zero sample debugging" of new devices.

[0109] In some embodiments, the edge computing layer 2 preloads a baseline parameter set for rapid startup when a new device is connected; the fog computing layer 3 maintains a debugging parameter pool that aggregates optimized process parameters of devices that have been stably running in the factory; when a new device is connected, the baseline parameter set is first loaded to achieve rapid startup within ten minutes. Subsequently, the system automatically collects the initial bonding data of the new device, automatically migrates the optimized process parameters most matched with the initial working condition of the new device from the debugging parameter pool through transfer learning and similarity matching algorithm, and performs compensation fine-tuning combined with the unique bias of the new device, to achieve rapid and high-precision debugging.

[0110] In one embodiment, the system includes an elastic activation mechanism, as shown in Figure 5 The mechanism dynamically adjusts the system architecture according to the number of wire bonding machines in the factory, daily output or process complexity, etc. Specifically, when the first preset condition is met, the edge computing layer 2 is directly connected to the cloud computing layer 4; when the second preset condition is met, the fog computing layer 3 is activated and deployed to perform factory-level collaboration between the edge computing layer 2 and the cloud computing layer 4.

[0111] Furthermore, this elastic activation mechanism is configured to activate only a single fog node to handle plant-level collaborative optimization when the second preset condition is met but the third preset condition is not. In other words, for small-scale production line applications, the fog computing layer 3 collaborative function is deployed to high-performance edge nodes (such as NVIDIA Jetson AGX); for ultra-large-scale deployments, regional cloud nodes are added above the fog computing layer 3 to achieve hierarchical federated aggregation. When both the second and third preset conditions are met, a distributed fog cluster is enabled to achieve broader regional collaborative control.

[0112] In some embodiments, the first preset condition may be that the number of devices N ≤ 3, or that the device is in a research and development laboratory or a small-batch production scenario. In this case, fog computing layer 3 is disabled, and edge computing layer 2 is directly connected to cloud computing layer 4, which is to say, an "edge-cloud" two-level architecture is adopted. The computing power allocation of edge nodes is dynamically adjusted, such as by adding edge cluster load balancing to achieve small-scale device collaboration. However, its response speed and collaboration efficiency are weaker than the solution when fog computing layer 3 is enabled. The second preset condition may be that the number of devices is N+1 to M, such as 4 to 10, and the daily output is ≥ 100,000 solder joints or mixed production ≥ 2 types of packaging. In this case, a single fog computing node is activated. The third preset condition may be that the number of devices is > M, such as more than 10, or there is a need for cross-factory collaboration. In this case, a distributed fog cluster is enabled to achieve regional collaborative control over a larger area. Among these, parameters such as the number of devices, daily output, and mixed production packaging types can be collaboratively determined based on production costs, preparation efficiency requirements, and specific user requirements.

[0113] Among them, such as Figure 6 As can be seen, this application proposes a process-driven layered architecture. In the equipment layer 1, microsecond-level real-time control is achieved, primarily handling the control of bonding parameters such as temperature, motion control algorithms, and ultrasonic power. In the edge computing layer 2, millisecond-level optimization is used, mainly constructing equipment health assessments such as solder joint quality detection models. In the fog computing layer 3, 100ms-level collaboration is implemented, achieving multi-machine parameter synchronization, spare parts early warning, and parameter drift compensation. Furthermore, the fog computing layer 3 employs a dynamic activation mechanism based on the number of workstations and the amount of bonding. In the cloud computing layer 4, second-level analysis is performed, conducting long-term process analysis to provide process optimization suggestions and lifetime predictions, ultimately achieving layered quality control.

[0114] like Figure 7 It can be seen that the fog computing layer 3 includes a fog layer data and communication management module 31, a factory-level collaborative learning and model management module 32, and a factory-level intelligent application module 33.

[0115] The fog layer data and communication management module 31 is the data and communication hub of the fog computing layer 3, and is the key infrastructure for realizing plant-level device collaboration and breaking data silos. The core function is to solve the access problem of multi-source heterogeneous devices, realize effective aggregation and safe flow of data in the plant, and provide a unified, safe and reliable data foundation for upper-layer collaborative learning and intelligent application. This module ensures that the data stream transmitted from the edge computing layer 2 is standardized, and at the same time guarantees the stability and security of the communication link between the edge computing layer 2 and the cloud computing layer 4.

[0116] The plant-level collaborative learning and model management module 32 receives the model parameter updates uploaded by the multiple edge learning modules 23, and uses the federated average (FedAvg) algorithm for local model aggregation. To improve the model convergence under non-independent and identically distributed data, more advanced aggregation algorithms such as FedProx (adding proximal term constraint) or SCAFFOLD (gradient correction mechanism) can also be used. The aggregated plant-level model can be distributed to each edge computing layer 2 for updating, or uploaded to the cloud computing layer 4 for global aggregation.

[0117] The plant-level intelligent application module 33 is the core value realization unit of the fog computing layer 3, which provides advanced intelligent services directly facing production efficiency, quality, cost and flexibility for the factory by using the data and models aggregated, processed and enhanced by the fog layer data and communication management module 31 and the plant-level collaborative learning and model management module 32.

[0118] The fog layer data and communication management module 31 includes a plant-level data aggregation and processing submodule 311, a multi-protocol gateway submodule 312, and a security agent submodule 313.

[0119] The plant-level data aggregation and processing submodule 311 is used to collect non-classified data and model parameter updates from multiple edge computing layers 2 in the plant, and perform preliminary feature extraction and analysis.

[0120] The multi-protocol gateway submodule 312 is used to solve the problem of incompatible communication protocols of devices from different manufacturers, such as Figure 8As shown, the sub-module is built-in with multiple mainstream industrial protocol analysis and conversion capabilities (such as Hesse HBN protocol based on CANopen, KS Modbus RTU, and domestic EtherCAT), and through OPC UA or MTConnect standard information model, it unifies heterogeneous device data into a standard format, or through a general OPC UA gateway to extend support for Profinet, EtherNet / IP, and other protocols, or deploy protocol conversion middleware (such as Apache PLC4X) to achieve plug-and-play data access of multi-vendor devices. If OPC UA / MTConnect protocol is not used, a customized API interface can be used to adapt to the bonding machine private protocol of a specific vendor.

[0121] The security agent sub-module 313 uses technologies such as integrated zero-knowledge proof verification mechanisms to perform security verification and threat isolation on internal and external communications, and builds a dual protection system of privacy protection and external threat isolation.

[0122] The factory-level intelligent application module 33 includes a rapid process design engine sub-module 331, a line-level quality and process control sub-module 332, and a line-level production and energy consumption management sub-module 333.

[0123] To solve the industry pain points of high dependence on manual experience, long trial-and-error period, and large material waste during new product introduction, the rapid process design engine sub-module 331 provides an automated, model-based process design solution, and its workflow is as follows: first, input quantifiable chip design parameters (such as chip size, pad layout, lead material type, wire diameter, etc.) into the rapid process design engine sub-module 331; then, the rapid process design engine sub-module 331 automatically generates a complete set of recommended process parameters that can be directly issued through the embedded process rule base and machine learning model (which can be a local optimization model aggregated by the factory-level collaborative learning and model management module 32, or a global model issued by the cloud computing layer 4) built from historical data and expert knowledge, including not only basic process parameters (such as bonding pressure, ultrasonic power, bonding time), but also precise knife splitting motion trajectories and time-varying ultrasonic power curves generated for specific chip layouts.

[0124] A key feature of the rapid process design engine submodule 331 is its deep integration of digital twin technology. After generating the recommended process parameters, instead of directly issuing the recommended process parameters to the physical equipment, a digital twin that is highly synchronized with the physical wire bonding machine is first constructed in a virtual environment. The rapid process design engine submodule 331 uses the digital twin to pre-play the entire wire bonding process with the recommended parameters. Through the built-in physical simulation engine, the system can calculate and detect various potential risks in real time during the virtual bonding process, such as collision between wires, interference between the blade and the chip surface, collapse of the loop, and potential risks such as broken wires, and accordingly automatically correct and optimize the recommended process parameters until the pre-play process safely meets the preset quality indicators.

[0125] Through this "design-simulation-optimization" closed-loop process, the rapid process design engine submodule 331 significantly shortens the process development and verification process that relies on experienced engineers for several hours or even several days in traditional technology to a minute-level automated intelligent decision-making process, thereby improving the rapid response capability of the production line to new product and new packaging demands, achieving optimal first launch of new processes, and reducing physical risks in the wire bonding process.

[0126] That is, compared to the real-time quality inspection of individual solder joints in the edge computing layer 2, the production line level quality and process control submodule 332 focuses on macro quality management and process control at the production line level. The production line level quality and process control submodule 332 uses the massive data aggregated from multiple edge nodes (including solder joint images, sensor time series data, process parameters, etc.) to construct more complex and accurate quality prediction and diagnosis models than edge layer models, thereby improving the accuracy and reliability of the recommended process parameters provided.

[0127] It should be understood that the production line level quality and process control submodule 332 can analyze the quality trend of the entire production line in real time, for example, by analyzing the bonding point image features (such as gold ball shape, bonding interface flatness) in real time through the visual sensing unit 114, and data such as pressure fluctuations and temperature curves from various sensors to predict product quality and detect potential defects such as fake welding and cracks in advance. That is, the production line level quality and process control submodule 332 can identify systematic quality fluctuations caused by specific batch materials, specific environmental changes, or parameter drifts of multiple devices. Potential systemic defects such as fake welding and cracks can be detected in advance, and an early warning can be sent to engineers, or even an automatic trigger of production line level process parameter adjustment or equipment calibration instructions can be triggered to achieve closed-loop control of the production process and ensure long-term stability of product quality.

[0128] In one specific embodiment, the wire bonding machine intelligent management system further comprises a material batch-optimal process parameter dynamic matching library, which is used to store the mapping relationship between different batches of wire bonding materials (gold wire, copper wire) and corresponding optimal process parameters, and its working mechanism is as follows:

[0129] (1) When the factory replaces a new batch of wire materials (such as gold wire, copper wire), the edge computing layer 2 collects the basic physical parameters of the new batch of materials through the sensors (such as hardness sensors, ductility testers) carried by the device layer 1 or manually input by the operator, and uploads the parameters to the fog computing layer 3 after encryption;

[0130] (2) After the fog computing layer 3 receives the material parameters, it searches for historical batch materials with high similarity (for example, ≥90%) to the new batch of materials in the dynamic matching library through the built-in similarity matching algorithm (such as K-nearest neighbor algorithm);

[0131] (3) Once the matching is successful, the system will extract the verified optimal process parameters corresponding to the historical batch of materials, such as the copper wire batch with similar hardness requiring 85W of ultrasonic power and 12N of bonding pressure, and will issue the optimal process parameters to the edge computing layer 2;

[0132] (4) The edge computing layer 2 performs trial bonding verification based on the issued parameters, and if the quality meets the requirements (for example, the pass rate of solder joints ≥99.5%), the parameters are directly applied; if not, fine-tuning is performed, and the final verified optimal parameters are fed back to the fog computing layer 3 for updating the dynamic matching library.

[0133] The production line level production and energy consumption management submodule 333 is responsible for intelligent scheduling of production tasks and fine management of energy consumption for the entire cluster of wire bonding machines in the factory. That is, in terms of production scheduling, the production line level production and energy consumption management submodule 333 considers order demand (such as product type, quantity, delivery date), real-time state of equipment (such as health degree, whether in maintenance period), process parameters, and material inventory, etc. information, establishes a production scheduling optimization model, through which the system can reasonably arrange the order of production tasks and their allocation among different devices, aiming to achieve multiple optimization goals, such as minimizing changeover time, maximizing equipment overall efficiency (OEE), ensuring on-time delivery of orders, etc.

[0134] In terms of energy consumption management, the production line level production and energy consumption management submodule 333 combines the energy consumption monitoring data collected from each device to analyze the energy consumption characteristics of the device under different working conditions (such as different bonding speeds and different process parameters). Based on these analyses, the system can optimize the production process and device operating parameters to reduce the energy consumption cost per product. For example, by dynamically adjusting the start-stop sequence and operating speed of multiple bonding machines, the system can achieve "peak load shifting" under the premise of meeting the delivery cycle, thereby reducing the peak power demand and overall energy consumption of the factory.

[0135]

Cloud computing layer 4

[0136] Cloud computing layer 4 uses a cloud server cluster as the hardware carrier and is deployed in a remote data center of a device manufacturer or a third-party cloud service provider. It is the global intelligent hub of the system and provides global computing, storage, and analysis capabilities for wire bonding systems.

[0137] As Figure 27 As can be seen, cloud computing layer 4 includes a global federated learning aggregation module 41, a global process knowledge and optimization module 42, a device lifecycle management module 43, a global quality and process control module 44, and a cross-factory production and energy consumption management module 45.

[0138] The global federated learning aggregation module 41 is the global intelligent decision-making core and the overall command center of distributed learning of the entire cloud-fog-edge-end collaborative intelligent management system. It is responsible for coordinating, managing, and executing global model training tasks across multiple physical locations (e.g., factories in different cities). That is, the global federated learning aggregation module 41 is used to receive model parameters uploaded from one or more fog computing layers 3 that have been updated through local aggregation, perform higher-level global model aggregation and training, and ultimately generate a globally optimized model with industry-level universality and high precision.

[0139] The global process knowledge and optimization module 42 is the "process brain" of the system. It aims to overcome the core difficulties of data silos and the difficulty of digitizing and scaling expert experience due to the complex interweaving of materials, equipment, and process parameters in the wire bonding field. This module converts massive and scattered process data into a structured, inferable, and continuously evolving global knowledge system, and provides intelligent process optimization and rapid adaptation capabilities based on this.

[0140] In some embodiments, the global federated learning aggregation module 41 is implemented as follows:

[0141] I. Coordination and scheduling of training tasks

[0142] The global federated learning aggregation module 41 initiatively initiates and manages the life cycle of the entire federated learning. In a typical training process, it performs the following operations: selecting or generating an initial global model (for example, a pre-trained model based on historical data or a randomly initialized model); broadcasting the current global model and the configuration of the current training task (for example, training rounds, learning rate, and other hyperparameters) to all participating fog computing layers 3; setting a time window or waiting condition to receive local aggregated model parameter updates uploaded from each fog computing layer 3.

[0143] II. Secure global parameter aggregation

[0144] The global federated learning aggregation module 41 aggregates the “intelligent crystals” from multiple factories (represented by fog computing layers 3) while strictly protecting data privacy. In a preferred embodiment, the global federated learning aggregation module 41 uses the federated averaging algorithm and performs weighted averaging on the received model parameter updates according to the number of devices or the amount of data represented by each fog computing layer 3 to calculate the global model parameters of the new generation. To achieve the highest level of security, this module can rely on homomorphic encryption, differential privacy, and other key technologies.

[0145] III. Update and distribution of global model

[0146] After aggregation, the global federated learning aggregation module 41 generates a global model with better performance, and then safely distributes the updated global model back to all participating fog computing layers 3, so that they can be further distributed to their respective edge computing layers 2 for guiding the next round of local training or directly applied to production.

[0147] IV. Model version management and performance monitoring

[0148] To ensure the stability and traceability of the system, the global federated learning aggregation module 41 also has the following functions: (1) The system automatically records and manages each version of the global model, including its training configuration, participants, performance indicators, and other metadata, which allows the system to easily roll back to a previous stable version or perform A / B testing on different versions of the model; the global federated learning aggregation module 41 also continuously monitors the performance indicators (such as accuracy on the validation set, loss function value, etc.) of the global model to determine whether the model has converged, which provides a basis for automatically adjusting the learning rate and deciding when to stop training, achieving automation and intelligence in the training process.

[0149] In some embodiments, as Figure 27 It can be seen that the global process knowledge and optimization module 42 includes a knowledge graph construction and management submodule 421, a process parameter transfer learning submodule 422, and a global process parameter intelligent optimization submodule 423.

[0150] The core function of the knowledge graph construction and management submodule 421 is to construct and continuously evolve a "material-equipment-process" three-dimensional dynamic knowledge graph. The knowledge graph is not a simple database, but a semantic network that depicts deep and implicit relationships between multiple elements. As shown in Figure 9A and Figure 9B The nodes of the knowledge graph include material properties (such as lead type, wire diameter, melting point, hardness, ductility, resistivity, etc.), equipment attributes (such as equipment model, static positioning accuracy, dynamic aging curve, frequency response characteristics of ultrasonic transducer, wedge model, etc.), process parameter set (such as bonding pressure, ultrasonic power, bonding time, temperature, wire arc control parameters, etc.), and production process parameters (such as solder joint image features including gold ball diameter and interface flatness, control trajectory data, and corresponding quality results). Through the "material-equipment-process" three-dimensional dynamic knowledge graph, the knowledge graph is continuously evolved and enriched at the global level without aggregating raw sensitive data from each factory. Based on this system, intelligent parameter transfer learning across materials and equipment is achieved. When dealing with new materials (such as switching from gold wire to copper wire) or new equipment models, the system automatically maps the reference process parameters through similarity matching and relationship reasoning in the knowledge graph, and superimposes compensation curves based on physical laws (such as temperature compensation for higher oxidation of copper wire), or transfers the calibration parameters of high-precision new equipment to old equipment after reverse calculation by the aging model to compensate for its precision drift. Finally, the debugging cycle of new processes is shortened from hours of traditional trial and error to minutes.

[0151] The secure construction and continuous evolution of the knowledge graph are achieved through a federated learning architecture. The edge computing layer 2 and the fog computing layer 3 of each factory train on local data, extract feature vectors or model parameters of the knowledge graph (i.e., digital representation of knowledge), and upload them after encryption. The knowledge graph construction and management submodule 421 only aggregates these encrypted parameters to update the global knowledge graph, thereby aggregating the wisdom of the entire industry without touching any raw sensitive data, making the knowledge graph continuously rich and perfect.

[0152] The core function of the process parameter transfer learning submodule 422 is to apply the knowledge graph to achieve rapid adaptation of process parameters in new scenarios, solving the problem of time-consuming and material-consuming traditional trial and error debugging. The workflow is as follows Figure 10The shown (1) scene recognition and knowledge retrieval, when the factory introduces new materials (for example, switching from mature gold wire process to copper wire process) or new models of equipment, the system first identifies the key features of the new scene, and then the sub-module performs similarity matching and relationship reasoning in the knowledge graph to find the closest historical successful case to the new scene; (2) reference parameter generation, the system extracts the verified process parameter set from the matched knowledge graph node as the reference parameter or initial value of the new scene; (3) intelligent compensation and correction, the sub-module will use the deep relationship in the knowledge graph for intelligent compensation, for example, when switching from gold wire to copper wire, the relationship of the knowledge graph points out that "copper wire is more prone to oxidation than gold wire", the system will automatically superimpose a compensation curve based on physical laws, such as appropriately increasing the bonding temperature or adjusting the protective atmosphere parameters, another example is that the calibration parameters of a new high-precision device are transferred to an old device after being inversely calculated by the aging model in the knowledge graph, to compensate for the precision drift caused by long-term use; (4) issuing and local fine-tuning, the generated optimized initial parameters are issued to the edge computing layer 2, which uses a small amount of sample data collected locally to fine-tune, for example, freezing the bottom feature extraction layer of the pre-trained model and only training the top classification layer, so as to quickly converge and realize the combination of "cloud global knowledge" and "local personalized needs".

[0153] On the other hand, since the process parameter transfer learning sub-module 422 uses a transfer learning scheduling strategy, the optimal process parameters of similar devices are mapped to each other, dynamic production scheduling reconstruction is achieved, and based on the federated learning framework, the production data and process experience of different enterprises after desensitization are fused to train the process parameter optimization model. Among them, a general model is pre-trained in the process knowledge graph constructed in the cloud computing layer 4, and a feature-based transfer method is used to extract common features of process parameters across devices. When new equipment or materials are added, the model parameters of similar working conditions in the graph are used as initial values, and local fine-tuning is performed in the edge computing layer 2 using a small amount of samples to quickly adapt the process parameters in the new scene. The time-consuming problem of traditional trial and error debugging is solved, the debugging cycle is shortened, the material waste phenomenon is reduced, the cloud federated aggregation generates a general process model as the transfer source model; when new equipment is connected, the model is loaded and fine-tuned (the bottom feature layer is frozen) through multiple local samples to achieve cross-device zero-sample rapid adaptation.

[0154] The function of the global process parameter intelligent optimization sub-module 423 is to actively explore and find the globally optimal process parameter combination. For example, Figure 11 With Figure 27As shown, the sub-module adopts advanced algorithms such as deep reinforcement learning to model the wire bonding process as a reinforcement learning environment. Deep reinforcement learning learns the optimal "strategy" (i.e. the rule for selecting the best process parameters under certain conditions) by "exploring" (trying different combinations of process parameters) in the environment and learning from the "rewards" (such as product yield, production efficiency, energy consumption, etc. comprehensive indicators) obtained. This optimization process is driven by the massive, multi-dimensional global production data aggregated by cloud computing layer 4 after federated learning aggregation, ensuring the universality and robustness of the optimization results. The optimization results are not a fixed set of parameters, but a dynamic recommendation model that can generate optimal process parameter combination suggestions under current conditions in real time according to the input specific conditions (such as chip type, wire batch, substrate material). The better process parameter combination discovered by the sub-module will be verified and added back to the knowledge graph as a new high-quality node, forming a virtuous cycle of "knowledge application-discovery of new knowledge-knowledge sedimentation", so that the intelligence level of the entire system continues to improve.

[0155] The device lifecycle management module 43 realizes the data closed loop of the wire bonding machine from research and development design, production and use to maintenance and scrap, adopts a "production manufacturer-user customer" two-way, "research and development-production-use-scrap" full-stage data collaboration mechanism. In other words, the research and development stage optimizes the device design (such as transducer frequency response) through data feedback; the production stage realizes adaptive optimization of process parameters; the use stage realizes predictive maintenance and remote diagnosis, the maintenance stage realizes spare parts prediction and maintenance knowledge base construction. In order to solve the technical problems that the existing wire bonding machine research and development process relies on laboratory single environment test, or lacks actual multi-scene working condition data support, resulting in insufficient adaptability of device design to actual application scene, such as Figure 12 、 Figure 13 and Figure 14 As shown, the device lifecycle management module 43 further includes a health and stability prediction sub-module 431, a performance degradation analysis and design feedback sub-module 432, and an after-sales resource and cost optimization sub-module 433.

[0156] The health and stability prediction sub-module 431 constructs a "device health-process stability" joint prediction model, such as Figure 12 and Figure 15As shown, this model uses a temporal deep learning network (such as LSTM or Transformer) as its core, comprehensively analyzing historical equipment operation data and real-time sensor readings to accurately predict the remaining lifespan and health status of key components (such as blade wear and ultrasonic transducer performance degradation). Furthermore, the health and stability prediction submodule 431 dynamically correlates the predicted equipment health status indicators with real-time acquired process parameters (such as bonding pressure and ultrasonic energy transfer efficiency), thereby quantifying and diagnosing the drift patterns of process parameters caused by equipment performance degradation. Based on this mapping relationship, the system proactively sends adaptive process parameter compensation instructions to the edge computing layer 2 (e.g., increasing the ultrasonic power setting value proportionally based on the transducer efficiency degradation curve), thereby maintaining the quality stability of the bonding process even under the premise of component performance degradation, effectively extending the effective service life of the equipment, and realizing a shift from passive maintenance to proactive health management, thus achieving accurate prediction and proactive management of key components.

[0157] Performance degradation analysis and design feedback submodule 432, such as Figure 13 and Figure 27 As shown, the performance degradation analysis and design feedback submodule 432 integrates operating data from different models and service years of equipment to establish a performance degradation curve model. By analyzing the differences between actual operating data and design standards, key factors affecting the long-term stability of the equipment are identified. These insights are stored in an anonymized and aggregated form in an anonymized fault database that integrates operating data of wire bonding machines from different users. Specifically, this includes, but is not limited to, lifespan data of core components (such as cutters and ultrasonic transducers) in multiple factory scenarios (e.g., the average lifespan and failure frequency of a certain model of cutter in 10 factories with different production environments), and correlation data between core component failures and operating environment parameters (temperature, humidity, production load) (e.g., the correspondence between the number of ultrasonic transducer failures and the workshop ambient temperature). All data is anonymized before being uploaded to the cloud computing layer 4, removing sensitive information such as customer identification and factory location, and is fed back to the R&D department of the equipment manufacturer to form an optimization closed loop of "use-design". Equipment manufacturers' R&D departments can access this anonymized fault database to obtain real-world operating data across multiple scenarios, replacing traditional single-laboratory environment test data. This data can be used for design optimization of next-generation wire bonding machines. For example, if the database shows that "a certain type of bonding machine experiences an ultrasonic power drift rate of 5% in a workshop with an ambient temperature >30℃, significantly higher than the 1% drift rate at room temperature," the R&D department can specifically improve the heat dissipation structure of the ultrasonic transducer for that model of bonding machine (such as adding micro-heat sinks or optimizing the heat dissipation duct), avoiding design defects caused by laboratory room temperature tests not covering high-temperature operating conditions. For instance, if data shows that a certain type of bonding machine experiences a significantly increased ultrasonic power drift rate in a high-temperature workshop, the R&D department can improve its heat dissipation structure accordingly.

[0158] Meanwhile, a process parameter-equipment structure mapping model can be built in the cloud computing layer 4, which inputs the process optimization data of multiple factories, specifically including the adjustment of parameter pairs by different factories for different bonding processes (such as copper wire bonding and gold wire bonding) (for example, the ultrasonic power needs to be increased by 15% to meet the bonding quality requirement when copper wire bonding), the loss influence data of the equipment structure components (such as clamps and guide rails) caused by the parameter adjustment (for example, the clamp wear rate is accelerated by 30% after the ultrasonic power is increased), and the correlation between “process parameter requirement-equipment structure loss” is established through the model algorithm; the R&D department can adjust the equipment structure design based on the model, for example, for the process requirement of higher ultrasonic power for copper wire bonding leading to faster clamp wear, the original clamp material is optimized from stainless steel to ceramic material, realizing “process requirement driven equipment structure design”, and improving the adaptability and durability of the next generation bonder to different bonding processes.

[0159] The after-sales resource and cost optimization submodule 433, such as Figure 14 and Figure 27 As shown, combined with the failure prediction results and the geographical distribution information of the equipment, the operational optimization algorithm (such as genetic algorithm and simulated annealing algorithm) is used to intelligently schedule maintenance personnel and spare parts inventory, reduce after-sales service cost, and improve response efficiency.

[0160] As shown in Figure 27 It can be seen that the cloud computing layer 4 also includes a global quality and process control module 44 and a cross-factory production and energy consumption management module 45, which are global versions of the corresponding modules of the fog computing layer, such as Figure 16 and Figure 17 As shown, the global quality and process control module 44 uses machine vision and sensor data to build a quality prediction model, analyzes bonding point features and process parameters in real time, predicts quality defects, triggers parameter adjustment or equipment calibration, and realizes production closed-loop control. The cross-factory production and energy consumption management module 45 can establish a production scheduling optimization model by comprehensively considering order demand, equipment state and process parameters, and reasonably arrange production tasks. Combined with energy consumption monitoring data, the equipment operation parameters are optimized to reduce energy consumption cost. That is, the global quality and process control module 44 and the cross-factory production and energy consumption management module 45 use macro data across factories to perform higher-dimensional quality analysis, trend prediction, and group-level production resource collaborative scheduling.

[0161] To solve the pain point of long deployment and debugging period of new equipment, the intelligent management system of the application provides a zero-shot debugging mechanism based on hierarchical parameter reuse, which includes:

[0162] Edge layer fast start: Edge computing layer 2 presets a set of baseline parameters for each new device based on the factory calibration data of the same batch of devices. The baseline parameter set includes the minimum parameter combination to achieve the basic bonding function. When the new device is connected, the parameter set is directly called to complete the initialization, and the "power-on operation" can be realized within 10 minutes.

[0163] Fog computing layer 3 high-precision debugging: Fog computing layer 3 maintains a "debugging parameter pool" that aggregates all optimized parameters of the same type of device in the factory. After the new device completes the preliminary debugging, edge computing layer 2 uploads a small amount of test bonding data to fog computing layer 3. Fog computing layer 3 calls the optimized parameter combination with the highest adaptation degree from the parameter pool through the parameter similarity matching algorithm, generates a compensation coefficient (such as assembly tolerance correction value) for the new device based on transfer learning, and sends it to edge computing layer 2 for fine tuning. High-precision debugging can be completed within 30 minutes.

[0164] For the batch debugging of thousands of devices, fog computing layer 3 can enable cluster debugging mode. That is, when the first device is debugged, the optimized parameters obtained are desensitized and included in the parameter pool. When debugging subsequent devices, fog computing layer directly calls the parameters for adaptive correction, so that the single debugging time is compressed to within 15 minutes, and the debugging accuracy deviation is ≤0.5%. During the whole process, the original process parameters only flow between the edge layer and the fog computing layer in the factory, which not only reduces the leakage risk of cloud data transmission, but also realizes efficient collaborative debugging of batch devices through factory-level parameter aggregation, solving the debugging efficiency bottleneck of large-scale deployment.

[0165] In some embodiments, the data flow between the levels of the system adopts a layered optimized communication protocol stack. The data flow of the cloud-fog-edge four-level framework is as shown in Figure 18 The uplink data flow is: the sensor data of device layer 1 is aggregated by fog computing layer 3 in the form of encrypted gradient after feature extraction by edge computing layer 2, and finally uploaded to cloud computing layer 4. The downlink control flow: the global model parameters of cloud computing layer 4 are distributed to edge computing layer 2 by fog computing layer 3, and the optimized control instructions are generated and sent to the actuators of device layer 1.

[0166] Specifically, in some embodiments, between device layer 1 and edge computing layer 2: industrial real-time Ethernet (preferably EtherCAT or PROFINET IRT) or high-speed fieldbus (such as CANopen FD) is adopted to meet the extremely low delay and deterministic communication requirements of the μs-level control loop. It is worth mentioning that the uplink data of device layer 1 refers to the position, temperature, vision, etc. of the multi-source sensor array 11. The real-time control instructions sent by edge computing layer 2 to device layer 1 include motion trajectory correction, emergency stop, parameter fine tuning, etc.

[0167] The edge computing layer 2 and fog computing layer 3 communicate using a lightweight message queue protocol (preferably MQTT over TLS 1.3, QoS 1 or 2) or OPC UA PubSub mode to suit the low-bandwidth, asynchronous communication scenarios within the factory network. MQTT / OPC UA PubSub offers advantages such as publish / subscribe mechanisms, low bandwidth overhead, support for asynchronous communication and reconnection after disconnection, making it more suitable for the factory network environment between edge computing layer 2 and fog computing layer 3. TLS (Transport Layer Security) ensures transmission security, and OPC UA PubSub also provides a built-in information model for structured data transmission. Notably, the uplink data from edge computing layer 2 includes feature data, model gradient updates, non-sensitive data, and alarm information. Fog computing layer 3 sends optimized model parameters, collaborative scheduling instructions, and process templates to edge computing layer 2.

[0168] Between the fog computing layer 3 and the cloud computing layer 4, reliable enterprise-grade messaging protocols (such as AMQP 1.0 over TLS) or HTTPS RESTful APIs are used to ensure high reliability and strong security for transmitting critical business data and model parameters across the wide area network. Notably, the uplink data from the fog computing layer 3 includes locally aggregated model parameters, fused feature data, summarized alarms, and performance reports. The cloud computing layer 4 sends data to the fog computing layer 3, including globally optimized model parameters, process parameter recommendation strategies, remote control commands, and software update packages. The downlink control flow uses the same protocol as the uplink to ensure reliable, secure, and efficient delivery of commands.

[0169] The intelligent management system of this application achieves intelligent coverage of the entire wire bonding process lifecycle through a layered architecture of cloud, fog, and edge. Specifically, for the coordination of "vision-operation control-structure," the intelligent management system of this application designs a two-layer mechanism that is interconnected yet has different focuses:

[0170] (1) The collaborative mechanism located in the edge computing layer 2 focuses on real-time quality control in the production process. It uses high-frequency sensor data to predict the quality trend of a single solder joint in milliseconds and uses the vision-motion control linkage submodule 221 to pre-adjust parameters, thereby achieving "real-time interception" and "proactive prevention" of quality defects.

[0171] (2) The collaborative mechanism located in the fog computing layer 3 focuses on process planning before the production of new products. Based on the chip design parameters, it automatically generates and verifies the complete process scheme in the virtual space through the digital twin technology of the rapid process design engine submodule 331, realizing the "minute-level deployment" and "first-time excellence" of the new process, fundamentally reducing the dependence on on-site debugging.

[0172] By setting up the aforementioned two-layer mechanism, one real and one virtual, one fast and one intelligent, together they constitute a closed-loop intelligent system covering the entire process from process design to production execution.

[0173] The intelligent management system of this application realizes a closed-loop data system covering the entire lifecycle of wire bonding machines, from R&D, production, use, maintenance to scrapping. During the R&D phase, data-driven design optimization is achieved through an anonymized fault database and a process-structure mapping model. During the production phase, a dynamic matching library of material batches and optimal process parameters can be built in the fog computing layer 3. When a new batch of materials is replaced, the parameters of the new material are collected through the edge layer, and the optimal process parameters of similar historical batches are retrieved from the matching library in the fog computing layer and distributed, achieving collaborative optimization of materials and processes. During the scrapping phase, a residual value assessment and tiered utilization scheme for core components is also constructed in the equipment lifecycle management module 43. The remaining value of core components (such as ultrasonic transducers) in scrapped equipment is assessed through a lifespan-performance degradation model. If the conditions for degrading use are met, the components are matched to users with low-precision packaging requirements through the platform, achieving resource recycling.

[0174] The intelligent management system of this application provides a complete set of management functions, such as... Figures 19 to 26 As shown, it includes: a real-time data acquisition module, an equipment status monitoring module, a remote control module, a data analysis module, an alarm system module, a process parameter management module, a historical data query module, and a user management module. These modules work together in a cloud-fog-edge-device architecture to provide users with comprehensive monitoring, management, and traceability capabilities.

[0175] (1) The real-time data acquisition module is equipped with sensors that collect data at a preset frequency and add data-sensitive tags. The edge computing layer 2 performs preliminary filtering, noise reduction, and format conversion on the data to ensure data accuracy and consistency, such as the fog computing layer 3 forwarding the data. Figure 19 The data processing flow is shown.

[0176] (2) The equipment status monitoring module is suitable for building an equipment status monitoring model based on local data in the edge computing layer 2 to judge the equipment operating status in real time; the fog computing layer 3 analyzes the overall operating trend by aggregating equipment data within the region. Combined with the global fault diagnosis knowledge obtained by federated learning, it accurately locates the fault type and cause, triggers hierarchical alarms, and the monitoring logic is as follows: Figure 20 As shown.

[0177] (3) Remote control module: Users initiate control commands through the cloud computing layer 4 platform, which are then executed through the edge computing layer 2. This covers operations such as process parameter adjustment, equipment start / stop, and calibration. Encrypted communication and two-factor authentication are used to ensure command security, and the equipment execution status is fed back in real time. It also supports the transmission of model parameter update commands. The control flow is as follows: Figure 21As shown, after logging into the cloud computing layer 4 platform, users select the target bonding machine in the factory's digital twin map and adjust the temperature / pressure / ultrasonic power values ​​or trigger start / stop and calibration commands through the parameter control panel. After the system enforces two-factor authentication (such as dynamic token combined with fingerprint verification), the command is sent to the edge computing layer 2 for execution via an encrypted channel, and the operation status is fed back to the platform's progress bar and logs in real time. Engineers can remotely intervene in faulty equipment to improve efficiency and avoid production line downtime.

[0178] (4) Data analysis module: Edge-fog-cloud collaborative modeling is implemented based on a federated learning framework. For process optimization, each enterprise's edge devices train models locally, fog computing layer 3 aggregates parameters, and cloud computing layer 4 generates a globally optimal process parameter recommendation model. For equipment fault prediction, multi-source data is integrated to train a high-precision prediction model, and the training effect and data contribution are displayed through a visual interface. The analysis architecture is as follows: Figure 22 As shown.

[0179] (5) Alarm System Module: Based on fault diagnosis knowledge optimized by federated learning, alarm rules and thresholds are dynamically adjusted. A tiered notification strategy (internal system notification or external notification via SMS / email) is adopted according to the severity of the fault and data sensitivity. The alarm mechanism is as follows: Figure 23 As shown.

[0180] (6) Process Parameter Management Module: This module combines a global process optimization model to provide users with accurate parameter recommendations. It supports user-defined parameter templates, and users participating in federated learning can automatically synchronize optimization suggestions and use the adjusted data for local model iteration. The management process is as follows: Figure 24 As shown.

[0181] (7) Historical Data Query Module: Supports multi-dimensional querying of equipment operation data, process parameters, alarm records, etc., by time, equipment number, product batch, etc. It specifically provides data querying related to federated learning (model training logs, parameter update records, data contribution), and displays it in diverse formats such as tables and charts. The query function includes... Figure 25 As shown, users can retrieve historical process parameters, alarm records, and federated learning data contributions through multi-dimensional filters (time / equipment ID / product batch); the results are visualized in the form of time series curves, defect heatmaps, etc., and it supports exporting federated audit reports containing model training versions and parameter update trajectories. It can complete the correlation analysis of tens of millions of data within 10 seconds, improving traceability efficiency by 80%.

[0182] (8) User Management Module: To implement user registration, login, and hierarchical permission management, new federated learning-specific permissions (data upload, model download, training participation) have been added. User operation logs are recorded for easy auditing and traceability. The permission management architecture is as follows: Figure 26As shown. Administrators set user permission levels (e.g., Operator / Engineer / Administrator) in the role configuration interface, and specify the detailed rules for each user permission level: for example, an "Operator" user can only view real-time data and basic parameters; an "Engineer" user can adjust process parameters and participate in local model training; and an "Administrator" user can configure federated learning strategies and cross-plant collaboration permissions. An approval process for permission changes is added (e.g., multi-level approval or automatic activation conditions), and users participating in federated learning are individually authorized to upload data and download models. The system automatically records user logins, key operations, and model call behaviors to a blockchain audit database to ensure traceability. This achieves fine-grained permission control for cross-enterprise collaboration, meeting ISO 27001 compliance requirements.

[0183] This application constructs a cloud-fog-edge-device collaborative fusion architecture, consisting of "real-time data acquisition at the device end, real-time control at the edge, dynamic collaboration at fog computing layer 3, and intelligent decision-making at cloud computing layer 4." Through layered data processing and distributed modeling architecture, it achieves microsecond-level real-time control of the bonding process at edge computing layer 2, eliminates cross-device data silos using multi-protocol gateways at fog computing layer 3, and breaks down reliance on manual experience with intelligent models at cloud computing layer 4, forming a full-stack technology closed loop. This enables local processing of sensitive data and collaborative sharing of non-sensitive data. While ensuring data privacy and security, it integrates multi-party data resources to improve the accuracy and versatility of models for equipment fault prediction and process optimization, and enhances equipment lifecycle management functions. This, in turn, improves wire bonding production efficiency, reduces enterprise operating costs, and enhances product quality stability. It provides a highly reliable, adaptive, and scalable intelligent bonding solution for semiconductor packaging.

[0184] This application addresses the unique challenges of intelligent manufacturing in the semiconductor wire bonding field, characterized by "microscale, multi-physics coupling, and high reliability." Compared to traditional general-purpose industrial control systems, wire bonding processes require micrometer-level positioning accuracy and millisecond-level action response, while simultaneously controlling multiple physical field parameters such as mechanical pressure, ultrasonic power, and temperature field. These precision requirements far exceed those of typical industrial scenarios. Furthermore, compared to traditional federated learning models, wire bonding data distribution is strongly bound to chip type, bonding material, and equipment model, necessitating federated learning guided by a process knowledge graph. Edge nodes must complete defect detection within 10ms, fundamentally differing from the universal algorithms of general frameworks. The intelligent management system in this application, compared to general-purpose industrial control models, focuses on the semiconductor packaging industry characteristics of wire bonding machines, including microsecond-level process response, strong multi-physics coupling, high sensitivity to process parameters, and a strong correlation between the lifespan of core components and the process. This involves multiple aspects, including system composition refinement, full lifecycle data closed-loop, and cross-subsystem linkage control.

[0185] The basic principles, main features, and advantages of this application have been described above. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely the principles of this application. Various changes and modifications can be made to this application without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. A lead bonding machine intelligent management system fusing cloud-fog-edge-end collaboration, characterized in that, The system comprises: a device layer comprising a multi-source sensor array integrated or mounted on a wire bonding machine, a physical execution unit for performing bonding operations, and a microsecond-level closed-loop controller for responding to control instructions and driving the physical execution unit; an edge computing layer deployed on the wire bonding machine and in microsecond-level communication with the device layer, the edge computing layer comprising: an edge data processing and modeling module for processing the real-time multi-physical field data and building a local model representing the bonding process; a real-time control and decision module for making millisecond-level intelligent decisions based on the local model and generating real-time control instructions to be sent to the microsecond-level closed-loop controller; an edge learning module for training models locally using classified process data and uploading only model parameter updates; a fog computing layer deployed on a local network for factory-level coordination of data and models of at least two edge computing layers; a cloud computing layer deployed in a remote data center for global aggregation and analysis of data and models of at least one fog computing layer and providing cross-domain knowledge services and full-life-cycle management; The system further comprises a flexible activation mechanism configured to dynamically adjust the system architecture according to the number of wire bonding machines or process complexity, directly connect the edge computing layer to the cloud computing layer when a first preset condition is met, and activate and deploy the fog computing layer to perform factory-level coordination between the edge computing layer and the cloud computing layer when a second preset condition is met. The data flow between the layers of the system uses a layered optimized communication protocol stack: the industrial real-time Ethernet or high-speed field bus protocol is used between the device layer and the edge computing layer; the lightweight message queue protocol or OPC UA PubSub mode is used between the edge computing layer and the fog computing layer; and the enterprise-level message protocol or HTTPS RESTful API is used between the fog computing layer and the cloud computing layer.

2. The wire bonder intelligent management system of claim 1, wherein, The real-time control and decision module of the edge computing layer further comprises a bonding process emergency control submodule for real-time monitoring of abnormal conditions during the bonding process, performing microsecond-level emergency shutdown at the device layer when a preset abnormal condition is detected, and automatically tracing back and storing multi-source sensor data before the abnormal condition occurs for traceability analysis.

3. The wire bonding machine intelligent management system of claim 1 or 2, wherein, The edge data processing and modeling module of the edge computing layer comprises a multi-physical field coupling engine submodule that fuses the multi-physical field data through a high-precision time synchronization mechanism and dynamically calculates real-time prediction values of key indicators of weld quality based on a physical information neural network model to form a closed loop control for adaptive adjustment of process parameters.

4. The wire bonder intelligent management system of claim 1, wherein, The real-time control and decision module of the edge computing layer includes a vision-operation linkage submodule, which is based on a solder feature-operation parameter correlation model, predicts the quality risk of the next solder based on the image feature change trend of the current and historical solders, and generates an operation parameter pre-adjustment instruction before the defect occurs to achieve active prevention of bonding quality problems; the edge computing layer is also preconfigured with a reference parameter set for quick start when a new device is accessed; the fog computing layer maintains a debugging parameter pool which aggregates the optimized process parameters of the devices that have been stably running in the factory; when a new device is accessed, the system automatically migrates the optimized process parameters that best match the initial working condition of the new device from the debugging parameter pool, and compensates and fine-tunes them in combination with the unique deviation of the new device to achieve quick and high-precision debugging.

5. The wire bonder intelligent management system of claim 1, wherein, The fog computing layer includes a factory-level intelligent application module, and the factory-level intelligent application module includes a quick process design engine submodule, which automatically generates a recommended process parameter set by inputting chip design parameters, combining a process rule library and a machine learning model; And using digital twin technology to pre-play the bonding process in a virtual environment to detect and avoid physical risks, thereby optimizing the recommended process parameter set.

6. The wire bonder intelligent management system of claim 1, wherein, The cloud computing layer includes a global process knowledge and optimization module, which is configured to: build and maintain a material-device-process three-dimensional dynamic knowledge graph; Through a federated learning architecture, the model parameters uploaded by each fog computing layer are aggregated to continuously evolve the knowledge graph without accessing the original confidential process data; and based on the knowledge graph, process parameter rapid adaptation is achieved in new material or new device scenarios through transfer learning; the cloud computing layer also includes a device lifecycle management module, which is configured to: build a device health and process stability joint prediction model, dynamically associate the predicted device key component health status with the process parameters, and issue adaptive process parameter compensation instructions; And build a performance degradation analysis and design feedback model to feedback the anonymized failure and process data of multiple factories to the device R&D design end to form an "use-design" optimization closed loop.

7. A method for intelligent management of a wire bonding machine in a fusion of cloud-fog-edge-end collaboration, characterized in that, The lead bonding machine intelligent management method is applied to the lead bonding machine intelligent management system as claimed in any one of claims 1-6, and the lead bonding machine intelligent management method comprises the following steps: At the device layer, a multi-source sensor array is used to collect multi-physical field data of the bonding process in real time, and a microsecond-level closed-loop controller is used to execute control instructions from the edge computing layer; At the edge computing layer, the multi-physical field data are processed and modeled in real time, millisecond-level intelligent decisions are made based on the local model to generate real-time control instructions, and only model parameter updates are uploaded after the model is trained using local confidential process data; At the fog computing layer, model parameter updates from at least two edge computing layers are aggregated and cooperatively optimized at the factory level; At the cloud computing layer, the aggregated model from at least one fog computing layer is globally trained and analyzed, and cross-domain knowledge services and full-life-cycle management are provided.

8. The wire bonding machine intelligent management method of claim 7, wherein, The step of making intelligent decisions at the edge computing layer further includes real-time monitoring of abnormal conditions during the bonding process, triggering the device layer to perform microsecond-level emergency shutdown when a preset serious abnormality is detected, and automatically backtracking and storing multi-source sensor data before the abnormality occurs; The method further includes a zero-sample debugging step: when a new device is connected, first load the baseline parameter set preloaded in the edge computing layer to achieve fast startup; Then, from the debugging parameter pool deployed in the fog computing layer, intelligently retrieve and migrate the optimized process parameters that match the initial working condition of the new device; Finally, combined with the unique bias of the new device, the migrated parameters are fine-tuned to complete high-precision debugging.

9. The method of claim 7, wherein, The step of collaborative optimization at the fog computing layer includes a rapid process design step: receiving chip design parameters as input, automatically generating a preliminary recommended process parameter set; Using digital twin technology to pre-play the bonding process using the recommended process parameter set in a virtual environment; According to the pre-play results, detect and correct potential physical risks to output an optimized process parameter set; The step of global training and analysis at the cloud computing layer includes the step of constructing and applying a knowledge graph: through a federated learning architecture, aggregate model parameters uploaded by each fog computing layer to continuously evolve a material-device-process three-dimensional dynamic knowledge graph without touching the original confidential process data; When a new material or new device is introduced, based on the knowledge graph, relationship reasoning and similarity matching are performed, and combined with transfer learning, initial process parameters for rapid adaptation are generated for the edge computing layer.

Images