An AI-driven gold bonding wire process parameter optimization method and system

Abstract

The present invention discloses an AI-driven bonding wire process parameter optimization method and system, which relates to the field of semiconductor packaging technology, including: establishing a dual-channel deep learning network for bonding process; the dual-channel deep learning network includes a dynamic parameter optimization channel and a static parameter optimization channel; inputting the real-time collected bonding process dynamic data and the preset bonding process static data into the dynamic parameter optimization channel to obtain dynamic process evaluation indicators; inputting the preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators; fusing the dynamic process evaluation indicators and the static process evaluation indicators through an adaptive weight fusion algorithm to obtain a comprehensive evaluation indicator, and optimize the bonding process parameters. The present invention integrates dynamic and static data through a dual-channel deep learning network architecture, and realizes intelligent and precise optimization of bonding process parameters in combination with an adaptive weight fusion algorithm, which significantly improves the bonding quality and process stability.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor packaging technology, and specifically to an AI-driven bonding wire process parameter optimization method and system. Background Art

[0002] With the rapid development of semiconductor packaging technology, the gold wire bonding process is a key technology for connecting chips to external leads, and its parameter optimization plays a decisive role in product performance and reliability. Traditional gold wire bonding process parameter optimization mainly relies on engineer experience and simple statistical analysis methods, and is achieved through trial-and-error adjustments of key parameters such as bonding force, ultrasonic power, and bonding time. In recent years, with the development of artificial intelligence technology, some researchers have begun to try to apply machine learning methods to process parameter optimization, such as using BP neural networks to predict bonding quality and using support vector machines to establish parameter optimization models. These methods have improved the automation level of parameter optimization to a certain extent, but there are still problems such as insufficient feature extraction and limited model generalization capabilities.

[0003] The existing methods for optimizing process parameters of gold bonding wires have the following main deficiencies: First, most methods fail to fully consider the synergy of dynamic data and static data in the bonding process, resulting in the inability of the optimization model to effectively capture the complexity of the process. Second, traditional single optimization algorithms often show limitations when dealing with high-dimensional, nonlinear process parameters, and are difficult to adapt to complex and changing production environments. Third, the existing methods lack in-depth analysis of microscopic factors such as stress distribution and deformation characteristics of bonding points, resulting in insufficient reliability of optimization results. In addition, under abnormal conditions such as equipment performance attenuation and process environment fluctuations, the robustness and adaptability of existing methods are poor, which can easily cause fluctuations in bonding quality. Finally, the existing technology lacks a systematic construction of process evaluation indicators, making it difficult to achieve a comprehensive evaluation and precise control of bonding quality. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the present invention provides an AI-driven bonding wire process parameter optimization method and system, which can solve the problems mentioned in the background technology.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: an AI-driven bonding wire process parameter optimization method, comprising: establishing a dual-channel deep learning network for a bonding process; the dual-channel deep learning network comprises a dynamic parameter optimization channel and a static parameter optimization channel; inputting real-time collected bonding process dynamic data and preset bonding process static data into the dynamic parameter optimization channel to obtain dynamic process evaluation indicators; inputting preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators; fusing the dynamic process evaluation indicators and the static process evaluation indicators through an adaptive weight fusion algorithm to obtain a comprehensive evaluation indicator, and optimizing the bonding process parameters according to the comprehensive evaluation indicators.

[0007] As a preferred solution of the AI-driven bonding wire process parameter optimization method described in the present invention, wherein: the bonding process dynamic data includes bonding force, ultrasonic power, bonding time, first bonding point deformation rate, second bonding point deformation rate, shear strength, gold wire arc height, equipment working time, and equipment maintenance cycle; the bonding process static data includes gold wire diameter, gold wire hardness, pad metallization layer thickness, metallization layer hardness, lead frame material strength, pad center distance, process environment temperature, and ambient humidity.

[0008] As a preferred solution of the AI-driven bonding wire process parameter optimization method described in the present invention, wherein: the dynamic parameter optimization channel is constructed based on the LSTM deep neural network, including an input layer, an LSTM hidden layer, a fully connected layer, and an evaluation and correction layer; wherein the input layer receives the bonding process dynamic data and the bonding process static data; the LSTM hidden layer performs time series feature extraction on the bonding process dynamic data and the bonding process static data, and outputs a feature state vector; the fully connected layer converts the LSTM hidden layer output into an initial evaluation value; the evaluation and correction layer includes a deformation rate correction unit and a quality index correction unit, which perform two-level corrections on the initial dynamic parameters in turn to obtain an evaluation and correction layer and a dynamic process evaluation index.

[0009] As a preferred solution of the AI-driven bonding wire process parameter optimization method of the present invention, the dynamic process evaluation index is obtained by the following steps:

[0010] The initial evaluation value is calculated by the fully connected layer:

[0011] ;

[0012] ;

[0013] in, is the weight matrix of the fully connected layer, is the characteristic state vector, It is the function affecting the device status; is the initial evaluation value; F is the bonding force; P is the ultrasonic power; T is the process environment temperature; t is the bonding time; L is the equipment working time; M is the equipment maintenance cycle.

[0014] The deformation rate correction unit calculates the intermediate evaluation value:

[0015] ;

[0016] in, is the standard shear strength; is the dynamic process sensitivity coefficient; is the intermediate evaluation value; D1 is the deformation rate of the first bonding point; D2 is the deformation rate of the second bonding point; S is the shear strength.

[0017] The quality index correction unit calculates the dynamic process evaluation index:

[0018] ;

[0019] ;

[0020] ;

[0021] in, is the theoretical optimal arc height; is the dynamic arc influence coefficient; H v is the hardness of the gold wire; σ is the strength of the lead frame material; H m is the hardness of the metallization layer; δ is the thickness of the metallization layer of the pad; It is the dynamic process evaluation index, that is, the output result of the dynamic parameter optimization channel; is the bonding quality coefficient; is the bonding deformation coefficient.

[0022] As a preferred solution of the AI-driven bonding wire process parameter optimization method described in the present invention, wherein: the static parameter optimization channel includes a material parameter processing unit and an environmental parameter processing unit; wherein the material parameter processing unit calculates the material benchmark index according to the bonding wire and pad material parameters; the environmental parameter processing unit corrects the material benchmark index in combination with the process environment parameters to obtain the static process evaluation index.

[0023] As a preferred solution of the AI-driven bonding wire process parameter optimization method of the present invention, the static process evaluation index is obtained by the following steps:

[0024] The material parameter processing unit calculates the material benchmark index:

[0025] ;

[0026] in, is the static process sensitivity coefficient; It is a material benchmark indicator.

[0027] The environmental parameter processing unit calculates the static process evaluation index:

[0028] ;

[0029] in, is the temperature influence coefficient, is the humidity influence coefficient, is the standard process temperature; It is a static process evaluation indicator.

[0030] As a preferred solution of the AI-driven bonding wire process parameter optimization method described in the present invention, the adaptive weight fusion algorithm includes:

[0031] Set initial weight coefficient , and 0≤ ≤1;

[0032] Calculate the fusion result:

[0033] ;

[0034] in, is the fusion result.

[0035] Dynamically adjust based on the following conditions :

[0036] When the shear strength S is lower than the standard shear strength , and the deformation ratio When the preset range is exceeded, the ;

[0037] When the bond deformation coefficient When the stress distribution at the bonding point interface is uneven, the stress is adjusted based on the deformation gradient of the bonding point. ;

[0038] When the bonding quality index When the product specification is exceeded, the shear strength S and the standard shear strength Dynamic adjustment of the coupling relationship between the deviation rate and its change trend , and the adjustment amplitude changes nonlinearly with the deviation duration.

[0039] To further solve the above technical problems, the present invention provides the following technical solutions: an AI-driven bonding wire process parameter optimization system, comprising: a data acquisition module, used to collect bonding process dynamic data in real time and preset bonding process static data; a network construction module, used to establish a dual-channel deep learning network for the bonding process; the dual-channel deep learning network includes a dynamic parameter optimization channel and a static parameter optimization channel; a dynamic evaluation module, used to input the real-time collected bonding process dynamic data and the preset bonding process static data into the dynamic parameter optimization channel to obtain dynamic process evaluation indicators; a static evaluation module, used to input the preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators; a comprehensive optimization module, used to fuse the dynamic process evaluation indicators and the static process evaluation indicators through an adaptive weight fusion algorithm to obtain comprehensive evaluation indicators, and optimize the bonding process parameters according to the comprehensive evaluation indicators.

[0040] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the AI-driven bonding wire process parameter optimization method as described above when executing the computer program.

[0041] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the AI-driven bonding wire process parameter optimization method as described above.

[0042] Beneficial effects of the present invention: The present invention realizes the coordinated processing of dynamic data and static data of bonding process by constructing a dual-channel evaluation architecture based on LSTM deep neural network, wherein the dynamic evaluation channel extracts the time series characteristics of process parameters through the hierarchical structure of input layer, LSTM hidden layer, fully connected layer and evaluation correction layer, and the static evaluation channel evaluates the process benchmark characteristics through material parameter processing unit and environmental parameter processing unit, and then the evaluation indicators of the two channels are intelligently fused through adaptive weight fusion algorithm, and the weight coefficient is dynamically adjusted based on microscopic characteristics such as deformation gradient and stress distribution of bonding point, and finally the precise optimization of process parameters such as bonding force, ultrasonic power and bonding time is realized. This method not only improves the accuracy and reliability of parameter optimization, but also enhances the adaptability and robustness of the present invention by considering factors such as equipment performance attenuation and process environment fluctuations, and significantly improves the bonding yield and product quality stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0044] Figure 1 This is a schematic diagram of the overall process of an AI-driven bonding wire process parameter optimization method proposed in the present invention;

[0045] Figure 2 A diagram of computer equipment in an AI-driven bonding wire process parameter optimization method proposed in the present invention. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0047] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0048] Example 1, reference Figure 1 , as an embodiment of the present invention, provides an AI-driven bonding wire process parameter optimization method.

[0049] The existing methods for optimizing process parameters of gold bonding wires have the following main deficiencies: First, most methods fail to fully consider the synergy of dynamic data and static data in the bonding process, resulting in the inability of the optimization model to effectively capture the complexity of the process. Second, traditional single optimization algorithms often show limitations when dealing with high-dimensional, nonlinear process parameters, and are difficult to adapt to complex and changing production environments. Third, the existing methods lack in-depth analysis of microscopic factors such as stress distribution and deformation characteristics of bonding points, resulting in insufficient reliability of optimization results. In addition, under abnormal conditions such as equipment performance attenuation and process environment fluctuations, the robustness and adaptability of existing methods are poor, which can easily cause fluctuations in bonding quality. Finally, the existing technology lacks a systematic construction of process evaluation indicators, making it difficult to achieve a comprehensive evaluation and precise control of bonding quality.

[0050] The present application provides an effective solution to the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to implement the AI-driven bonding wire process parameter optimization method.

[0051] Figure 1 The overall flow chart of an AI-driven gold bonding wire process parameter optimization method is shown, including:

[0052] S1: Establish a dual-channel deep learning network for the bonding process. The dual-channel deep learning network includes a dynamic parameter optimization channel and a static parameter optimization channel.

[0053] Furthermore, the dynamic parameter optimization channel is constructed based on the LSTM deep neural network, including an input layer, an LSTM hidden layer, a fully connected layer, and an evaluation and correction layer. Among them, the input layer receives the bonding process dynamic data and the bonding process static data; the LSTM hidden layer extracts the time series features of the bonding process dynamic data and the bonding process static data, and outputs the feature state vector; the fully connected layer converts the LSTM hidden layer output into an initial evaluation value; the evaluation and correction layer includes a deformation rate correction unit and a quality index correction unit, which perform two-level corrections on the initial evaluation value in turn to obtain an intermediate evaluation value and a dynamic process evaluation index.

[0054] Preferably, by establishing a dynamic evaluation channel architecture based on LSTM deep neural network, the channel includes a hierarchical structure design of input layer, LSTM hidden layer, fully connected layer and evaluation correction layer, and realizes intelligent evaluation of bonding process parameters. Among them, the input layer simultaneously receives dynamic data of bonding process (such as bonding force, ultrasonic power, bonding time, etc.) and static data (such as gold wire diameter, material hardness, etc.). The LSTM hidden layer can effectively capture the timing characteristics of parameter changes in the bonding process through its unique gating mechanism and memory unit, overcoming the limitation that traditional feedforward neural networks cannot handle timing dependence. This evaluation architecture based on deep learning not only improves the system's perception of the dynamic characteristics of the bonding process, but also establishes a more comprehensive process parameter evaluation system through the coordinated input of dynamic and static data.

[0055] S2: Input the real-time collected bonding process dynamic data and the preset bonding process static data into the dynamic parameter optimization channel to obtain the dynamic process evaluation index.

[0056] Specifically, the dynamic data of the bonding process include bonding force F, ultrasonic power P, bonding time t, deformation rate of the first bonding point D1, deformation rate of the second bonding point D2, shear strength S, gold wire arc height H, equipment working time L, and equipment maintenance cycle M.

[0057] Static parameter data include wire diameter d, wire hardness H v , pad metallization thickness δ, metallization hardness H m, lead frame material strength σ, pad center distance l, process environment temperature T, and ambient humidity R.

[0058] The process of obtaining dynamic process evaluation indicators includes the following steps:

[0059] 1. Calculate the initial evaluation value through the fully connected layer:

[0060] ;

[0061] ;

[0062] in, is the weight matrix of the fully connected layer, is the characteristic state vector, It is the function affecting the device status; is the intermediate evaluation value; F is the bonding force; P is the ultrasonic power; T is the process environment temperature; t is the bonding time.

[0063] 2. The deformation rate correction unit calculates the intermediate evaluation value :

[0064] ;

[0065] in, is the standard shear strength; is the dynamic process sensitivity coefficient; is the intermediate dynamic parameter; D1 is the deformation rate of the first bonding point; D2 is the deformation rate of the second bonding point; S is the shear strength.

[0066] It should be noted that the timing features extracted by LSTM are converted into initial evaluation values ​​through the calculation mechanism of the fully connected layer. This evaluation value comprehensively considers the coupling relationship between process parameters such as bonding force, time, power and temperature, and introduces the equipment state influence function to correct the equipment operation status. This evaluation mechanism that takes the equipment state into consideration effectively solves the problem of ignoring the impact of equipment maintenance cycle in traditional methods and improves the reliability of process parameter optimization. Subsequently, the intermediate evaluation value is calculated by the deformation rate correction unit. This value establishes a quantitative relationship between bonding quality and process parameters by introducing correction terms for gold wire deformation rate and shear strength, providing a more accurate evaluation basis for parameter optimization.

[0067] 3. The quality index correction unit calculates dynamic process evaluation indicators:

[0068] ;

[0069] ;

[0070] ;

[0071] in, is the theoretical optimal arc height; is the dynamic arc influence coefficient; H v is the hardness of the gold wire; σ is the strength of the lead frame material; H m is the hardness of the metallization layer; δ is the thickness of the metallization layer of the pad; It is the dynamic process evaluation index, that is, the output result of the dynamic parameter optimization channel; is the bonding quality coefficient; is the bonding deformation coefficient.

[0072] It should be noted that through the final processing of the quality index correction unit, the system obtains a dynamic process evaluation index, which innovatively integrates the bonding quality coefficient , Bond deformation coefficient And arc influence factors, a complete bonding quality evaluation system is constructed. It reflects the matching degree between bonding strength and material properties. The adaptability of bonding deformation and pad structure is characterized, and the arc influence factor takes into account the influence of the arc height of the gold wire on the bonding quality. This multi-level evaluation and correction mechanism not only improves the accuracy of process parameter optimization, but also enhances the adaptability and robustness of the optimization scheme by considering the influencing factors of multiple dimensions such as materials and structures. Practice has proved that this evaluation system can effectively improve the bonding yield, significantly improve the stability of product quality, and provide innovative technical solutions for intelligent manufacturing in the field of semiconductor packaging.

[0073] S3: Input the preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators.

[0074] Specifically, the static evaluation channel includes a material parameter processing unit and an environmental parameter processing unit. The material parameter processing unit calculates the material reference index according to the bonding wire and pad material parameters; the environmental parameter processing unit corrects the material reference index in combination with the process environment parameters to obtain the static process evaluation index.

[0075] The process of calculating static process evaluation indicators includes the following steps:

[0076] ;

[0077] in, is the static process sensitivity coefficient; It is a material benchmark indicator.

[0078] The environmental parameter processing unit calculates static process evaluation indicators:

[0079] ;

[0080] in, is the temperature influence coefficient, is the humidity influence coefficient, is the standard process temperature; It is a static process evaluation indicator.

[0081] Preferably, by setting up a dual-unit architecture of a material parameter processing unit and an environmental parameter processing unit in the static evaluation channel, a systematic evaluation of the static characteristics of the bonding process is achieved. The calculation formula for the tensile strength of the gold wire is Structural characteristics of pads Perform coupling calculations and introduce deformation difference correction terms , and establish a mathematical relationship between bonding materials and structural parameters. This evaluation method based on material mechanics not only takes into account the intrinsic properties of the material, but also incorporates the pad structural parameters into the evaluation system, providing a reliable material benchmark for process parameter optimization. In particular, through the static process sensitivity coefficient α s The invention can sensitively capture the phenomenon of uneven deformation of bonding points and effectively prevent bonding quality problems caused by mismatch of material properties.

[0082] Through the calculation mechanism of the environmental parameter processing unit, the present invention performs environmental impact correction on the material benchmark index to obtain the final static process evaluation index This indicator is obtained by introducing the temperature change rate and humidity change rate The coupling correction term is used to establish a quantitative relationship between process environment fluctuation and bonding quality. In the form of, not only the impact of instantaneous changes in environmental parameters is considered, but also the deviation between process temperature and standard temperature is included in the evaluation scope. This correction mechanism based on the dynamic characteristics of the environment significantly improves the environmental adaptability of process parameter optimization, especially in production scenarios with large fluctuations in environmental conditions, which can achieve precise control of process parameters and ensure the stability of bonding quality. and humidity influence coefficient The present invention achieves differentiated treatment of the degree of influence of different environmental factors through synergistic effects, and provides a more targeted evaluation basis for process parameter optimization.

[0083] S4: The dynamic process evaluation index and the static process evaluation index are fused through an adaptive weight fusion algorithm to obtain a comprehensive evaluation index, and the bonding process parameters are optimized according to the comprehensive evaluation index.

[0084] Specifically, the adaptive weight fusion algorithm includes the following: setting the initial weight coefficient , and 0≤ ≤1.

[0085] Calculate the fusion results, that is, the comprehensive evaluation index:

[0086] ;

[0087] in, It is a comprehensive evaluation indicator.

[0088] Dynamically adjust based on the following conditions :

[0089] When the shear strength S is lower than the standard shear strength , and the deformation ratio When the preset range is exceeded, the ;

[0090] When the bond deformation coefficient When the stress distribution at the bonding point interface is uneven, the stress is adjusted based on the deformation gradient of the bonding point. ;

[0091] When the bonding quality index When the product specification is exceeded, the shear strength S and the standard shear strength Dynamic adjustment of the coupling relationship between the deviation rate and its change trend , and the adjustment amplitude changes nonlinearly with the deviation duration.

[0092] Furthermore, according to the comprehensive evaluation index Adjust controllable parameters in the bonding process dynamics data, including:

[0093] when When it is lower than the preset threshold, the bonding force F is increased according to the preset step size based on the deviation between the arc height H of the gold wire and the standard arc height, and the ultrasonic power P is adjusted according to the real-time ratio of the deformation rate D1 of the first bonding point to the deformation rate D2 of the second bonding point;

[0094] When the arc height H of the gold wire deviates from the target value and the working time L of the equipment exceeds the preset time, the stability of the shear strength S is comprehensively considered and the compensation amount of the bonding time t is dynamically adjusted;

[0095] In the predetermined time period from the equipment maintenance cycle M, based on the historical change trend of shear strength S, wire arc height H and D1 / D2, the equipment performance attenuation model is established to realize the adaptive compensation of process parameters. It should be noted that this can be achieved through the following schemes: Support vector regression SVR method: Using shear strength S, wire arc height H and D1 / D2 as feature vectors, a nonlinear mapping relationship is established, and the low-dimensional feature space is mapped to the high-dimensional space through the kernel function to realize the prediction modeling of equipment performance attenuation, so as to adaptively adjust the process parameters. Long short-term memory network LSTM: The time series data of shear strength S, wire arc height H and D1 / D2 are input into the LSTM network, and the long-term dependency is captured through the gate control unit to establish a dynamic prediction model of equipment performance changes over time, so as to realize the real-time compensation of process parameters. Fuzzy neural network FNN: Based on expert experience, a fuzzy rule base is constructed, and the historical data of shear strength S, wire arc height H and D1 / D2 are used as training samples. The membership function and weight coefficient are optimized through the adaptive neural network to establish a fuzzy reasoning system to realize the intelligent compensation of process parameters.

[0096] It should be noted that during the bonding process, when the bonding deformation coefficient When the stress distribution at the bonding point interface is uneven and the stress is lower than the target value, it indicates that there may be a local weak area in the bonding interface. At this time, by analyzing the deformation gradient characteristics of the area around the bonding point, the stress concentration degree and distribution law can be obtained, and then the bonding quality of the bonding interface can be judged. When the deformation gradient is too large, it means that the stress distribution is highly uneven and needs to be increased. To increase the influence weight of dynamic parameters; when the deformation gradient is relatively gentle, it means that the stress distribution tends to be uniform, so the corresponding reduction Through this adaptive adjustment mechanism based on deformation gradient, the stress distribution state of the bonding interface can be effectively improved and the stability of the bonding quality can be improved.

[0097] The bonding quality factor When the product specification is exceeded, the shear strength S and the standard shear strength should be considered comprehensively. The deviation rate of the bonding process and the dynamic change trend of the deviation are considered, and a coupling relationship model between the two is established. This model not only considers the absolute size of the deviation, but also includes the higher-order characteristics such as the rate of change and acceleration of the deviation into the calculation range. At the same time, considering the hysteresis effect of parameter adjustment during the bonding process, the duration of the deviation is taken as an important influencing factor. The adjustment amplitude of shows nonlinear variation with duration. When the duration is short, the adjustment amplitude is relatively small to avoid over-response; when the duration is prolonged, the adjustment amplitude will gradually increase according to the preset nonlinear function, thus achieving more accurate and robust parameter optimization.

[0098] In summary, the present invention realizes the collaborative processing of dynamic data and static data of bonding process by constructing a dual-channel evaluation architecture based on LSTM deep neural network, wherein the dynamic evaluation channel extracts the time series characteristics of process parameters through the hierarchical structure of input layer, LSTM hidden layer, fully connected layer and evaluation correction layer, and the static evaluation channel evaluates the process benchmark characteristics through material parameter processing unit and environmental parameter processing unit, and then the evaluation indicators of the two channels are intelligently fused through adaptive weight fusion algorithm, and the weight coefficient is dynamically adjusted based on microscopic characteristics such as deformation gradient and stress distribution of bonding point, and finally the precise optimization of process parameters such as bonding force, ultrasonic power and bonding time is realized. This method not only improves the accuracy and reliability of parameter optimization, but also enhances the adaptability and robustness of the present invention by considering factors such as equipment performance attenuation and process environment fluctuations, and significantly improves the bonding yield and product quality stability.

[0099] Embodiment 2 is an embodiment of the present invention, which provides an AI-driven bonding wire process parameter optimization system, including: a data acquisition module, used to collect bonding process dynamic data in real time and preset bonding process static data; a network construction module, used to establish a dual-channel deep learning network for the bonding process; the dual-channel deep learning network includes a dynamic parameter optimization channel and a static parameter optimization channel; a dynamic evaluation module, used to input the real-time collected bonding process dynamic data and the preset bonding process static data into the dynamic parameter optimization channel to obtain dynamic process evaluation indicators; a static evaluation module, used to input the preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators; a comprehensive optimization module, used to fuse the dynamic process evaluation indicators and the static process evaluation indicators through an adaptive weight fusion algorithm to obtain comprehensive evaluation indicators, and optimize the bonding process parameters according to the comprehensive evaluation indicators.

[0100] Example 3, reference Figure 2, is an embodiment of the present invention, which is different from the previous embodiment in that: if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0101] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0102] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0103] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0104] Example 4 is an embodiment of the present invention, which provides an AI-driven bonding wire process parameter optimization method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0105] In order to verify the beneficial effects of the present invention, the production line of a semiconductor packaging company was selected for experimental verification. The experimental data collection period is from June 1, 2023 to December 31, 2023, involving three different specifications of products (28nm, 40nm and 65nm process nodes), using gold wires with diameters of 20μm, 25μm and 30μm for bonding respectively. The experimental platform adopts ASMEagle Xtreme bonding machine, equipped with high-precision force sensors and real-time image acquisition system. Data of 10,000 bonding points are collected for each specification of product, including dynamic parameters such as bonding force, ultrasonic power, bonding time, deformation rate, shear strength, and static data such as gold wire material properties and pad structure parameters. During the experiment, a dynamic parameter evaluation model is first established through an LSTM deep neural network. The network contains a 2-layer LSTM structure with 128 neurons in each layer. It is trained using the Adam optimizer and the learning rate is set to 0.001. At the same time, a static evaluation model is constructed based on material parameters and process environment parameters. On this basis, the dynamic fusion of the two evaluation channels is realized through an adaptive weight fusion algorithm. In order to verify the advantages of the method of the present invention, tests were carried out under normal working conditions, high temperature conditions (85°C), high humidity conditions (85%RH) and vibration conditions (0.1g@60Hz), and compared with traditional empirical methods and single artificial intelligence model methods. In addition, a special equipment aging compensation experiment was designed to collect data at different maintenance cycle stages to verify the effectiveness of the present invention in dealing with equipment performance degradation.

[0106] Table 1 Performance comparison data of different optimization methods under various working conditions

[0107] Product Type Working condition type Optimization Methods Bond strength (gf) Deformation uniformity (%) Arc deviation(μm) Shear strength (gf) Yield(%) Optimization time (min) Stability score 28nm Products Normal working condition Traditional empirical method 6.2 85.3 2.8 45.6 95.2 45 0.82 28nm Products Normal working condition Single AI model 6.5 88.6 2.3 48.2 96.8 15 0.86 28nm Products Normal working condition The present invention 6.8 92.4 1.5 52.3 98.5 8 0.93 28nm Products High temperature conditions Traditional empirical method 5.8 82.1 3.2 42.3 91.3 60 0.76 28nm Products High temperature conditions Single AI model 6.2 85.8 2.6 45.7 93.5 20 0.82 28nm Products High temperature conditions The present invention 6.6 90.2 1.8 49.8 96.7 10 0.90 40nm Products High humidity conditions Traditional empirical method 6.0 83.5 3.0 43.5 90.8 55 0.78 40nm Products High humidity conditions Single AI model 6.3 86.9 2.4 46.8 92.9 18 0.84 40nm Products High humidity conditions The present invention 6.7 91.5 1.6 50.6 96.2 9 0.91 65nm Products Vibration conditions Traditional empirical method 5.7 81.8 3.4 41.8 89.5 65 0.75 65nm Products Vibration conditions Single AI model 6.1 85.2 2.7 44.9 91.7 22 0.81 65nm Products Vibration conditions The present invention 6.5 89.7 1.9 48.9 95.4 11 0.89

[0108] It can be seen from the experimental data in Table 1 that the present invention is significantly superior to the traditional empirical method and the single AI model method under various working conditions. Under normal working conditions, the bonding yield of the present invention reaches 98.5%, which is 3.3 percentage points higher than the traditional empirical method and 1.7 percentage points higher than the single AI model method. In particular, in terms of the two key indicators of bonding strength and deformation uniformity, the present invention achieved 6.8gf and 92.4% respectively, which are significantly higher than the other two methods. This is mainly due to the dual-channel deep learning network architecture adopted by the present invention, which can simultaneously consider the influence of dynamic parameters and static parameters to achieve more accurate parameter optimization.

[0109] In harsh environments, the advantages of the present invention are more obvious. Under high temperature conditions, the yield of the traditional empirical method drops to 91.3%, while the present invention can still maintain a high yield of 96.7%, reflecting extremely strong environmental adaptability. This is because the present invention accurately models the impact of temperature changes through a static evaluation channel and realizes dynamic compensation of parameters through an adaptive weight fusion algorithm. Similarly, under high humidity and vibration conditions, the performance degradation of the present invention is also significantly smaller than that of the other two methods. For example, under vibration conditions, the bond strength and shear strength of the present invention are maintained at 6.5gf and 48.9gf, respectively, while the traditional empirical rule drops to 5.7gf and 41.8gf, respectively.

[0110] In addition, the present invention also has significant advantages in optimization efficiency. Under normal working conditions, the traditional empirical method takes 45 minutes to complete parameter optimization, the single AI model takes 15 minutes, and the present invention only takes 8 minutes to complete. This high efficiency not only improves production efficiency, but also can respond to process fluctuations more quickly. It can be seen from the stability scoring index that the present invention maintains a high stability under various working conditions (the scores are all above 0.89), thanks to its compensation mechanism that takes into account the attenuation of equipment performance and can adaptively adjust parameters according to the equipment status. Overall, the present invention has shown obvious technical advantages in bonding quality, environmental adaptability, optimization efficiency and stability, and provides innovative solutions for intelligent manufacturing in the field of semiconductor packaging.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An AI-driven gold bonding wire process parameter optimization method, characterized in that: include: Establish a dual-channel deep learning network for bonding process; The dual-channel deep learning network includes a dynamic parameter optimization channel and a static parameter optimization channel; Inputting the real-time collected bonding process dynamic data and the preset bonding process static data into the dynamic parameter optimization channel to obtain a dynamic process evaluation index; Inputting preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators; The dynamic process evaluation index and the static process evaluation index are fused by an adaptive weight fusion algorithm to obtain a comprehensive evaluation index, and the bonding process parameters are optimized according to the comprehensive evaluation index; The bonding process dynamic data includes bonding force, ultrasonic power, bonding time, deformation rate of the first bonding point, deformation rate of the second bonding point, shear strength, arc height of gold wire, equipment working time, and equipment maintenance cycle; the bonding process static data includes gold wire diameter, gold wire hardness, pad metallization layer thickness, metallization layer hardness, lead frame material strength, pad center distance, process environment temperature, and environment humidity; The dynamic parameter optimization channel is constructed based on the LSTM deep neural network, including an input layer, an LSTM hidden layer, a fully connected layer, and an evaluation and correction layer; Wherein, the input layer receives the bonding process dynamic data and the bonding process static data; The LSTM hidden layer extracts time series features from the bonding process dynamic data and the bonding process static data, and outputs a feature state vector; The fully connected layer converts the LSTM hidden layer output into an initial evaluation value; The evaluation correction layer includes a deformation rate correction unit and a quality index correction unit, which perform two-level corrections on the initial dynamic parameters in sequence to obtain the evaluation correction layer and the dynamic process evaluation index; The static parameter optimization channel includes a material parameter processing unit and an environmental parameter processing unit; Wherein, the material parameter processing unit calculates the material reference index according to the bonding wire and pad material parameters; The environmental parameter processing unit corrects the material reference index in combination with the process environmental parameter to obtain the static process evaluation index.

2. The AI-driven gold bonding wire process parameter optimization method according to claim 1, characterized in that: The dynamic process evaluation index is obtained by the following steps: The initial evaluation value is calculated by the fully connected layer: ； ； in, is the weight matrix of the fully connected layer, is the characteristic state vector, It is the function affecting the device status; is the initial evaluation value; F is the bonding force; P is the ultrasonic power; T is the process environment temperature; t is the bonding time; L Working hours for the equipment; M Equipment maintenance cycle; The deformation rate correction unit calculates the intermediate evaluation value: ； in, is the standard shear strength; is the dynamic process sensitivity coefficient; is the middle evaluation value; D 1 is the deformation rate of the first bonding point; D 2 is the deformation rate of the second bonding point; S is the shear strength; The quality index correction unit calculates the dynamic process evaluation index: ； ； ； in, is the theoretical optimal arc height; is the dynamic arc influence coefficient; H v is the hardness of the gold wire; σ is the lead frame material strength; H m is the hardness of the metallization layer; δ is the thickness of the pad metallization layer; It is the dynamic process evaluation index, that is, the output result of the dynamic parameter optimization channel; is the bonding quality coefficient; is the bonding deformation coefficient.

3. The AI-driven bonding wire process parameter optimization method according to claim 2, characterized in that: The static process evaluation index is obtained by the following steps: The material parameter processing unit calculates the material benchmark index: ； in, is the static process sensitivity coefficient; It is a material benchmark indicator; The environmental parameter processing unit calculates the static process evaluation index: ； in, is the temperature influence coefficient, is the humidity influence coefficient, is the standard process temperature; It is a static process evaluation indicator.

4. The AI-driven bonding wire process parameter optimization method according to claim 3, characterized in that: The adaptive weight fusion algorithm includes: Set initial weight coefficient , and 0≤ ≤1; Calculate the fusion result: ； in, is the fusion result; Dynamically adjust based on the following conditions : When the shear strength S Below standard shear strength , and the deformation ratio When the preset range is exceeded, the ; When the bond deformation coefficient When the stress distribution at the bonding point interface is uneven, the stress is adjusted based on the deformation gradient of the bonding point. ; When the bonding quality index When the product specification is exceeded, based on the shear strength S Standard shear strength Dynamic adjustment of the coupling relationship between the deviation rate and its change trend , and the adjustment amplitude changes nonlinearly with the deviation duration.

5. A system using the AI-driven bonding wire process parameter optimization method as described in any one of claims 1 to 4, characterized in that: include: Data acquisition module, used to collect bonding process dynamic data in real time and preset bonding process static data; A network building module for building a dual-channel deep learning network for the bonding process; The dual-channel deep learning network includes a dynamic parameter optimization channel and a static parameter optimization channel; A dynamic evaluation module, used to input the real-time collected bonding process dynamic data and the preset bonding process static data into the dynamic parameter optimization channel to obtain a dynamic process evaluation index; A static evaluation module, used for inputting preset bonding process static data into the static parameter optimization channel to obtain static process evaluation indicators; The comprehensive optimization module is used to fuse the dynamic process evaluation index and the static process evaluation index through an adaptive weight fusion algorithm to obtain a comprehensive evaluation index, and optimize the bonding process parameters according to the comprehensive evaluation index.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the AI-driven bonding wire process parameter optimization method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the AI-driven bonding wire process parameter optimization method according to any one of claims 1 to 4 are implemented.

Images