Wire bonding method based on die interconnect

Abstract

This invention relates to the field of semiconductor packaging technology and discloses a bonding wire welding method based on die interconnects. This method utilizes a multimodal in-situ sensing mechanism integrating acoustic resonance and surface plasmon resonance to collect and fuse acoustic response and optical reflection signals at the bonding interface in real time. It dynamically calculates the thickness of the intermetallic compound layer, mechanical coupling deformation, and energy dissipation state. Combined with a model predictive control algorithm, it uses closed-loop adjustment of ultrasonic power and bonding force to ensure the welding process always evolves along a preset optimal physical path. The system includes a multimodal sensing welding head, a synchronous data acquisition module, a state calculation module, and an adaptive control module. This invention achieves transparency and intelligence in the welding process, significantly improving the bonding consistency, reliability, and yield of high-density interconnects.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor packaging technology, specifically relating to a bonding wire welding method based on die interconnect. Background Technology

[0002] In the fields of semiconductor packaging and advanced integration technology, die interconnects are a crucial link in realizing high-density, high-performance chip systems, and their reliability directly determines the lifespan and stability of the entire electronic system. Bonding wires, as a widely used interconnecting medium, connect die pads to package pins via thermoforming or ultrasonic welding, undertaking the core functions of signal transmission and power distribution. However, with the increasing integration density and operating frequency of chips, bonding wire structures are becoming increasingly miniaturized and denser. Their solder joints are highly susceptible to mechanical stress, thermal cycling fatigue, and electromigration, leading to failure modes such as increased contact resistance and connection breakage, ultimately causing system-level failures.

[0003] The bonding wire welding method based on bare die interconnects focuses on achieving highly reliable and traceable electrical connections at the micrometer scale. This method not only requires a strong physical connection but also the ability to sense and assess the connection status online. Traditional bonding processes rely on offline sampling or macroscopic electrical performance testing, which cannot independently identify and monitor the status of individual or specific groups of bonding wires. This makes it difficult to detect early, minor degradations in a timely manner, resulting in low fault location efficiency and a lack of dynamic modeling capabilities for the evolution of welding quality.

[0004] In existing technologies, bonding wires are typically treated as indiscriminate conductive paths, lacking individual identification information and in-situ sensing mechanisms. Even when some solutions attempt to introduce impedance measurement, the lack of a unique electrical identifier for each bonding wire makes it impossible to distinguish signal aliasing between multiple channels, hindering the establishment of an accurate impedance-health state mapping. Furthermore, in high-density packaging scenarios, the compact bonding wire layout and severe parasitic parameter coupling make conventional monitoring methods susceptible to interference from adjacent lines, resulting in low signal-to-noise ratios and an inability to achieve sensitive responses to microscopic degradation of solder joints (such as interface oxidation, void formation, and microcrack propagation).

[0005] Therefore, there is an urgent need for a welding method that can assign a unique electrical identifier to the bonding wire and integrate micro impedance monitoring capabilities, so as to achieve accurate, real-time, and traceable intelligent diagnosis of the health status of the die interconnect. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a bonding wire welding method based on bare die interconnection. This method introduces a multimodal in-situ sensing mechanism based on acoustic resonance and surface plasmon resonance, and combines it with a model predictive control algorithm to construct a closed-loop welding system that performs real-time calculation and adaptive control of the physical state of the bonding interface.

[0007] This method no longer relies on preset, unchanging process parameters. Instead, it dynamically adjusts ultrasonic power and bonding pressure by sensing the thickness of intermetallic compounds at the atomic scale, the mechanical coupling deformation at the microscale, and the energy dissipation at the macroscale of the bonding interface in real time. This ensures that the welding of each bonding wire reaches the preset optimal physical state, thereby fundamentally improving the bonding quality, consistency, and reliability of high-density die interconnects.

[0008] The bonding wire welding method based on bare die interconnection provided by the present invention includes the following steps: providing a welding wedge with integrated multimodal sensing function, wherein the welding wedge has an integrated piezoelectric transducer for capturing acoustic signals of the bonding interface, and a miniature optical fiber channel for conducting probe light is provided along its central axis.

[0009] During the bonding wire welding operation, the welding wedge synchronously performs real-time acquisition of multi-physics field signals, including the acoustic resonance response signal and optical reflection spectrum signal of the bonding interface.

[0010] The acoustic resonance response signal acquired in real time is subjected to spectrum analysis to extract the resonant frequency offset, which characterizes the change in mechanical stiffness of the coupled system composed of the welding wedge, bonding wire and solder pad, and the acoustic emission energy integral value, which characterizes the interface plastic deformation and energy dissipation.

[0011] The real-time acquired optical reflection spectrum signal is solved, and the real-time generated thickness of the intermetallic compound layer at the bonding interface is determined by analyzing the position change of the surface plasmon resonance peak.

[0012] The resonant frequency offset, acoustic emission energy integral value, and intermetallic compound layer thickness are fused into a multi-dimensional bonding state feature vector to characterize the physical evolution state of the bonding process in real time.

[0013] Within a preset control period, the real-time bonding state feature vector is compared with the pre-established target bonding state evolution model, and the deviation between the two is calculated.

[0014] Based on the deviation, a model prediction controller generates dynamic adjustment instructions for welding process parameters in real time, including the driving power of the ultrasonic transducer and the normal bonding force applied to the welding GEO cutter.

[0015] According to the dynamic adjustment command, the driving power of the ultrasonic transducer and the normal bonding force are adjusted in a closed loop until the bonding state feature vector converges to the endpoint state defined by the target bonding state evolution model, thus completing the single-point welding.

[0016] As one embodiment of the present invention, the welding wedge integrating multimodal sensing function specifically includes: the main body of the welding wedge is made of cemented carbide material, and its working end face has a geometric configuration for guiding and flattening the bonding wire; the piezoelectric transducer is an annular piezoelectric ceramic sheet, which is coaxially fixed to the shank of the welding wedge by an insulating adhesive, and its electrode lead-out end is connected to a high-frequency signal acquisition circuit through a micro coaxial cable; the micro optical fiber channel runs through the central axis of the welding wedge, its inner wall is coated with an antireflection film layer, and its entrance end is connected to a light source and a spectral detector through a standard optical fiber connector.

[0017] As one embodiment of the present invention, the real-time acquisition of the synchronous execution of multi-physics field signals specifically includes: continuously acquiring, through the piezoelectric transducer, a composite acoustic vibration signal generated by ultrasonic welding energy excitation and propagating between the welding blade and the bonding interface at a sampling rate of not less than 5 MHz;

[0018] A monochromatic laser source is used to focus a coherent beam of a specific wavelength onto the contact area between the bonding wire and the pad below the working end face of the welding wedge via the entrance end of the micro-fiber channel. The light signal reflected from the contact area is then guided into the spectral detector via the same entrance end of the micro-fiber channel using a beam splitter, and the reflection spectral data is acquired at a frequency of not less than 10 kHz.

[0019] As one embodiment of the present invention, the spectral analysis of the real-time acquired acoustic resonance response signal specifically includes: applying a short-time Fourier transform to the acquired time-domain acoustic vibration signal to generate a continuous time-frequency distribution map;

[0020] In the time-frequency distribution diagram, identify and track first-order or multi-order resonance peaks related to the overall mode of the welding system;

[0021] The resonant peak frequency at the start of welding is set as the reference frequency, and the offset of the current resonant peak frequency relative to the reference frequency is calculated in real time.

[0022] The amplitude of the time-domain acoustic vibration signal is squared and then integrated within a time window to obtain the acoustic emission energy integral value that is directly related to the intensity of the acoustic emission event.

[0023] As one embodiment of the present invention, the calculation of the real-time acquired optical reflection spectrum signal specifically includes:

[0024] The wavelength of the monochromatic laser source is pre-selected to a specific value that can effectively excite plasmon resonance on the surface of the pad metal material;

[0025] The spectral detector records the curve of the reflected light intensity as a function of wavelength or incident angle in real time.

[0026] On the curve, locate and trace the point of lowest reflectivity, i.e., the resonance peak, formed by surface plasmon resonance absorption;

[0027] During the bonding process, the formation of intermetallic compounds in aluminum-gold or copper-gold alters the interfacial dielectric constant, causing a measurable wavelength shift in the resonance peak.

[0028] By using a lookup table or mathematical function established in advance through calibration experiments, the wavelength shift of the resonance peak is directly mapped to the average thickness of the intermetallic compound layer, with a measurement resolution reaching the nanometer level.

[0029] As one embodiment of the present invention, the target bonding state evolution model is a four-dimensional time-series database stored in the controller. It defines the standard values ​​that the resonant frequency offset, acoustic emission energy integral value, and intermetallic compound layer thickness should reach at each time step from time zero to completion of the ideal welding process. The model is constructed by in-situ monitoring and statistical analysis of a large number of welding samples with optimal mechanical and electrical properties.

[0030] As one embodiment of the present invention, the real-time generation of dynamic adjustment instructions through the model prediction controller specifically includes:

[0031] The model predictive controller predicts the bonding state evolution trajectory for multiple future time steps within a finite time domain, based on a system state-space model that describes the dynamic relationship between welding process parameters and bonding state feature vectors.

[0032] A quadratic objective function is constructed that includes tracking error and control energy consumption, wherein the tracking error is the difference between the predicted trajectory and the target bonding state evolution model, and the control energy consumption is the rate of change of ultrasonic power and bonding force;

[0033] By solving the constrained optimization problem, a future control sequence that minimizes the quadratic objective function is calculated in each control cycle, and the first element of the sequence is output as the dynamic adjustment command for the current cycle.

[0034] According to another aspect of the invention, a bonding wire bonding system based on die interconnect is provided, comprising:

[0035] A multimodal sensing welding head, wherein the multimodal sensing welding head has a built-in piezoelectric transducer and a central optical fiber channel;

[0036] A multi-source data synchronous acquisition module is electrically connected to the piezoelectric transducer and connected to the central fiber optic channel via an optical fiber, for synchronously acquiring acoustic resonance response signals and optical reflection spectrum signals;

[0037] The real-time bonding state calculation module is connected to the multi-source data synchronous acquisition module and is used to process and fuse the acquired signals to calculate a multi-dimensional bonding state feature vector that characterizes the current bonding physical state in real time.

[0038] An adaptive closed-loop control module receives the bonding state feature vector and has a built-in target bonding state evolution model and model prediction controller to generate dynamic adjustment instructions for welding process parameters.

[0039] The process parameter actuator includes an ultrasonic power source and a high-precision force control actuator, which is used to receive the dynamic adjustment command and adjust the ultrasonic power and bonding force applied to the multimodal sensing welding head in real time.

[0040] As one embodiment of the present invention, the real-time bonding state calculation module is implemented by a field-programmable gate array, which has internal parallel processing logic for performing short-time Fourier transform, resonance peak tracking, acoustic energy integration, spectral analysis and surface plasmon resonance peak calculation, ensuring that the delay of state calculation is on the order of microseconds.

[0041] In one embodiment of the present invention, the adaptive closed-loop control module is implemented by a digital signal processor, which runs the model predictive control algorithm and stores the target bonding state evolution model; the model can be selectively loaded according to different bonding wire materials, pad metallization types and substrate materials to adapt to different welding application scenarios.

[0042] In one embodiment of the present invention, the system further includes a welding quality database for storing the evolution trajectory of the bonding state feature vector throughout each welding operation, the final quality assessment result, and the corresponding dynamic adjustment instruction sequence. The data in the database is used for iterative optimization of the target bonding state evolution model and to provide data support for fault diagnosis and process improvement.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] 1. This invention, by introducing a dual in-situ sensing mechanism of acoustic resonance and surface plasmon resonance, achieves for the first time direct, real-time, and multi-dimensional quantitative monitoring of the internal physical state of the bonding wire welding interface, making the traditionally invisible welding process transparent; the method fundamentally solves the problem that traditional open-loop control methods cannot cope with local uncertainties such as material surface cleanliness, metal layer thickness, and differences in underlying structure.

[0045] 2. The adaptive closed-loop control system constructed in this invention utilizes advanced model predictive control algorithms to transform the welding process from a fixed "execution" mode to a dynamic "optimization" mode, ensuring that the formation of each weld joint follows a verified optimal physical evolution path, thereby greatly improving the consistency of bonding quality and yield.

[0046] 3. This invention can record data in the entire time domain of the welding process and generate a unique "health record" for each weld point. This not only achieves 100% online quality monitoring, but also provides an unprecedented data foundation for realizing intelligent process optimization, preventive maintenance of equipment, and quality traceability throughout the product life cycle. It has great value for promoting the advancement of high-density, high-reliability semiconductor packaging technology. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall technical solution architecture of the bonding wire welding method based on bare die interconnection proposed in this invention;

[0048] Figure 2 This is a schematic diagram of the core principle framework of the multimodal in-situ sensing mechanism based on acoustic resonance and surface plasmon resonance in this invention.

[0049] Figure 3 This is a flowchart illustrating the logical process of real-time calculation and fusion of bonding state feature vectors in this invention.

[0050] Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the multimodal sensing welding head and its signal acquisition and execution mechanism in this invention;

[0051] Figure 5 This is a schematic diagram of the logical architecture of the welding process data closed loop and quality traceability system in this invention. Detailed Implementation

[0052] Please refer to Figures 1 to 5 This invention provides a bonding wire welding method based on die interconnects. Its core lies in constructing a closed-loop adaptive control system that integrates acoustic resonance and surface plasmon resonance dual sensing mechanisms to achieve real-time sensing, calculation, and control of the physical state of the bonding interface. This method simultaneously acquires multi-physics field signals during the welding process and dynamically adjusts the ultrasonic power and normal bonding force based on a model predictive control algorithm, ensuring that each solder joint completes its formation along a preset optimal physical evolution path. This significantly improves the consistency and reliability of bonding quality in high-density packaging.

[0053] The method first performs step S1: providing a welding wedge with integrated multimodal sensing capabilities. The main body of the welding wedge is made of cemented carbide material, which has excellent wear resistance and thermal stability. Its working end face is designed with a geometric configuration for guiding and flattening the bonding wire, including a V-groove or wedge-shaped pressure head structure, to ensure that the bonding wire forms a stable initial contact point when it contacts the pad.

[0054] At the shank of this welding wedge, a ring-shaped piezoelectric ceramic sheet is coaxially die-bonded as a piezoelectric transducer. This piezoelectric transducer is firmly bonded to the outer wall of the wedge with insulating adhesive. Its inner and outer electrodes are respectively led out and connected to a high-frequency signal acquisition circuit through a micro coaxial cable. The function of the piezoelectric transducer is to convert the mechanical vibration generated by ultrasonic energy excitation during the welding process into an electrical signal, thereby capturing the acoustic resonance response of the bonding interface.

[0055] Furthermore, a miniature optical fiber channel is formed through the central axis of the welding blade. The inner wall of this channel is coated with an antireflection film to reduce reflection loss of the optical signal during transmission. The entrance end of this optical fiber channel is connected to an external light source and a spectrometer via a standard optical fiber connector, used to conduct probe light and receive optical signals reflected from the bonding interface. This structure enables the welding blade to not only perform traditional mechanical crimping functions but also possess acoustic and optical in-situ sensing capabilities, forming the core carrier of multimodal sensing.

[0056] Then, step S2 is executed: During the bonding wire welding operation, real-time acquisition of multi-physics field signals is performed synchronously through the welding wedge. Specifically, when the bonding wire is fed above the pad and begins to be pressed down, the system simultaneously initiates two types of signal acquisition processes.

[0057] The first category is acoustic signal acquisition: piezoelectric transducers continuously acquire composite acoustic vibration signals generated by ultrasonic transducers at a sampling rate of not less than 5 MHz. This signal includes the overall vibration mode of the coupled system consisting of the welding wedge body, bonding wire, and solder pad, as well as transient acoustic emission events caused by inelastic behaviors such as interface plastic deformation, microslip, and crack initiation.

[0058] The second type is optical signal acquisition: a coherent beam emitted from a monochromatic laser source is transmitted through a miniature fiber optic channel and focused on the contact area between the bonding wire and the solder pad below the working end face of the welding wedge. The laser wavelength is pre-set to a specific value that can effectively excite plasmon resonances on the surface of the solder pad metal material (such as gold, aluminum, or copper). The reflected light returns through the same fiber optic channel and is guided to a spectrometer via a beam splitter. The spectrometer records the intensity of the reflected light as a function of wavelength at a frequency of at least 10 kHz, thereby obtaining real-time information on the dielectric environment of the bonding interface. The two types of signals are strictly synchronized in the time dimension to ensure physical consistency for subsequent fusion analysis.

[0059] Next, step S3 is executed: spectral analysis is performed on the real-time acquired acoustic resonance response signal to extract the resonant frequency offset characterizing the change in mechanical stiffness of the welding system, and the integral value of acoustic emission energy characterizing interface energy dissipation. This process first applies a short-time Fourier transform to the original time-domain acoustic signal to generate a time-frequency joint distribution map.

[0060] In this distribution diagram, the system identifies and tracks the first-order principal resonance peak associated with the overall welded structural modes. This resonance peak typically lies in the 100 kHz to 2 MHz range, and its frequency position is highly sensitive to system stiffness. The resonance peak frequency at the start of welding is set as the reference frequency. As the bonding process progresses, factors such as increased contact area between the bonding wire and the pad, formation of intermetallic compounds, and interfacial plastic flow cause changes in the overall system stiffness, leading to a shift in the resonant frequency towards higher or lower frequencies. The current resonant frequency is calculated in real time. and The difference As a key indicator characterizing the evolution of interfacial bonding strength.

[0061] At the same time, the original time-domain signal After squaring, the integral is performed within the sliding time window to obtain the integral value of the acoustic emission energy. This integral value directly reflects the total amount of energy dissipated by the interface due to inelastic deformation per unit time, and can be used to determine whether abnormal states such as excessive plastic flow, microcrack propagation, or insufficient bonding have occurred.

[0062] Then, step S4 is executed: the real-time acquired optical reflection spectrum signal is processed, and the real-time thickness of the intermetallic compound layer at the bonding interface is determined by analyzing the positional changes of the surface plasmon resonance peaks. Surface plasmons are electromagnetic modes formed by the coupling of collective oscillations of free electrons at the metal-dielectric interface with incident light, and their resonance conditions are extremely sensitive to the interface dielectric constant. In the initial stage of bonding, the pad surface is a pure metal (such as gold), with a fixed dielectric function corresponding to a specific resonance wavelength. As the bonding wire (such as aluminum or copper) comes into contact with the pad under high temperature and pressure, atoms diffuse into each other to form intermetallic compounds (such as...). , (etc.), this compound has dielectric properties different from pure metals, causing a change in the effective dielectric constant at the interface, which in turn causes a shift in the resonant wavelength. The spectrometer records the wavelength position corresponding to the point of lowest reflectivity in real time and inputs it into a pre-established mapping model. This mapping model is obtained through numerous calibration experiments and can take the form of a lookup table or an analytical function. , where h represents the average thickness of the intermetallic compound layer. This thickness resolution reaches the nanometer level, with a typical measurement range of 0 to 500 nanometers, covering the critical stages from initial contact to complete reaction. This parameter is directly related to the metallurgical bonding quality of the bonding interface and is a core indicator for evaluating electrical conductivity and long-term reliability.

[0063] Then, step S5 is performed: the resonant frequency offset is... Acoustic emission energy integral value and the real-time generation thickness of intermetallic compound layers Fuse into a multi-dimensional bonded state feature vector This feature vector is used to characterize the physical evolution of the bonding process in real time. It is updated in each control cycle, forming a quantitative description of the current synthesis state of the bonding interface.

[0064] in, Reflects the strength of macroscopic mechanical coupling. Characterizing microscopic energy dissipation behavior, This reveals metallurgical reaction processes at the atomic scale. Together, these three elements constitute a complete observational space covering multi-scale physical phenomena, avoiding the limitations of relying on a single parameter criterion. For example, depending solely on... Poor mechanical contact due to insufficient pressure may be overlooked; relying solely on It may be difficult to distinguish between normal plastic flow and microcrack formation. By integrating these three factors, the system can comprehensively identify the normal evolution path and abnormal deviation trends in the bonding process.

[0065] Next, step S6 is executed: within a preset control period, the real-time bonding state feature vector is... With the pre-established target bond state evolution model Compare them and calculate the deviation vector between them. The target bonding state evolution model It is a four-dimensional time-series database, stored in the controller's non-volatile memory.

[0066] This model defines the ideal welding process at each time step from time zero to completion. , , The model represents the standard numerical sequence that the three parameters should achieve. It was constructed through in-situ monitoring and statistical analysis of a large number of welding samples exhibiting optimal tensile strength, shear strength, and low contact resistance. Different material combinations (e.g., gold-aluminum, copper-gold), different wire diameters (e.g., 17 μm, 25 μm), and different substrate types (e.g., silicon, ceramic) correspond to different sub-models. The system automatically loads the matching sub-model based on the process recipe before welding. Deviation vector. Each dimension represents the degree of deviation between the current state and the ideal trajectory. The larger the absolute value, the more the current process parameters deviate from the optimal control point.

[0067] Then proceed to step S7: based on the deviation The model predictive controller generates dynamic adjustment commands for welding process parameters in real time. The model predictive controller operates within a finite prediction time domain N, and its core function is to describe the process parameters. (in For ultrasonic power, (Normal bond force) and eigenvectors of the bond state A system state-space model of the dynamic relationships between them:

[0068] ;

[0069] Where A is the state transition matrix and B is the input influence matrix, both obtained through online or offline calibration in system identification experiments. The controller solves the following constrained optimization problem in each control cycle k:

[0070] ;

[0071] ;

[0072] ;

[0073] Where Q and R are the weight matrices for the state tracking error and the control increment, respectively. Let k be the ideal state predicted for the i-th future step based on the current time k. To correspond to the predicted state, The variable represents the change in the control command. The solution to this optimization problem is a future control sequence of length N, where the controller executes only the first element of the sequence. This serves as a dynamic adjustment command for the current cycle. This strategy balances tracking accuracy with control smoothness, avoiding system oscillations or mechanical shocks caused by drastic adjustments.

[0074] Finally, step S8 is executed: based on the dynamic adjustment command, the driving power of the ultrasonic transducer and the normal bonding force are adjusted in a closed loop until the bonding state characteristic vector is obtained. The process converges to the endpoint state defined by the target bonding state evolution model M*, completing the single-point welding.

[0075] Specifically, dynamic adjustment commands are sent to the process parameter actuator. This actuator includes a high-frequency ultrasonic power source whose output power can be continuously adjusted from zero to its maximum rated value with an adjustment accuracy of not less than 0.5%; it also includes a high-precision force-controlled actuator, typically a voice coil motor or piezoelectric actuator, with a normal force control range of 10 millinewtons to 500 millinewtons and a resolution better than 0.1 millinewtons. The actuator adjusts its output in real time according to the commands, ensuring that the ultrasonic energy and mechanical pressure applied by the welding wedge precisely match the requirements of the current bonding interface. Within multiple consecutive control cycles, the deviation from the endpoint state of M* is less than a preset threshold. Once the system determines that the welding is complete, it lifts the cutting tool and ends the operation.

[0076] Throughout the welding process, all acquired raw signals, calculated feature vectors, generated control commands, and final quality assessment results are recorded in the welding quality database. This database generates a unique "health record" for each weld joint, containing time-series data from the entire process. This data is used not only for real-time quality assessment but also for iterative optimization of the target bond state evolution model—analyzing the feature differences between successful and failed cases using machine learning algorithms to automatically correct model parameters and improve its generalization ability. Simultaneously, the database supports fault diagnosis, for example, when a batch of products frequently exhibits… Slow growth When the temperature rises abnormally, the system can infer that there is oxidation contamination on the surface of the pads and trigger a cleaning process alarm.

[0077] The implementation of this method relies on a highly integrated hardware system. The multimodal sensing welding head, as the front-end sensing unit, integrates a piezoelectric transducer and a central fiber optic channel; the multi-source data synchronous acquisition module is responsible for acquiring acoustic and optical signals with nanosecond-level synchronization accuracy; the real-time bonding state calculation module is typically implemented using a field-programmable gate array (FPGA), which has internally embedded parallel processing logic capable of performing short-time Fourier transforms, resonance peak tracking, energy integration, and spectral calculations within microseconds; the adaptive closed-loop control module is implemented using a digital signal processor (DSP), running a model predictive control algorithm and managing the loading and switching of multiple target bonding state evolution models; and the process parameter execution mechanism ensures high-fidelity execution of control commands. All modules are interconnected via a high-speed bus, forming a low-latency, high-reliability closed-loop control system.

Claims

1. A bonding wire bonding method based on die interconnect, characterized in that, include: A welding cutter with integrated multimodal sensing function is provided. The welding cutter has an integrated piezoelectric transducer for capturing acoustic signals at the bonding interface and a miniature optical fiber channel for conducting probe light along its central axis. During the bonding wire welding operation, the welding wedge synchronously performs real-time acquisition of multi-physics field signals, including the acoustic resonance response signal and optical reflection spectrum signal of the bonding interface. The acoustic resonance response signal acquired in real time is subjected to spectrum analysis to extract the resonant frequency offset, which characterizes the change in mechanical stiffness of the coupled system composed of the welding wedge, bonding wire and solder pad, and the acoustic emission energy integral value, which characterizes the interface plastic deformation and energy dissipation. The real-time acquired optical reflection spectrum signal is solved, and the real-time generated thickness of the intermetallic compound layer at the bonding interface is determined by analyzing the position change of the surface plasmon resonance peak. The resonant frequency offset, acoustic emission energy integral value, and intermetallic compound layer thickness are fused into a multi-dimensional bonding state feature vector to characterize the physical evolution state of the bonding process in real time. Within a preset control period, the real-time bonding state feature vector is compared with the pre-established target bonding state evolution model, and the deviation between the two is calculated. Based on the deviation, a model prediction controller generates dynamic adjustment instructions for welding process parameters in real time, including the driving power of the ultrasonic transducer and the normal bonding force applied to the welding wedge. According to the dynamic adjustment command, the driving power of the ultrasonic transducer and the normal bonding force are adjusted in a closed loop until the bonding state feature vector converges to the endpoint state defined by the target bonding state evolution model, thus completing the single-point welding.

2. The bonding wire welding method based on bare die interconnection according to claim 1, characterized in that, The welding cutter that provides integrated multimodal sensing functionality specifically includes: The main body of the welding wedge is made of cemented carbide, and its working end face has a geometric configuration for guiding and flattening the bonding wire. The piezoelectric transducer is a ring-shaped piezoelectric ceramic sheet, which is coaxially fixed to the shank of the welding wedge with an insulating adhesive. Its electrode lead-out end is connected to a high-frequency signal acquisition circuit through a micro coaxial cable. The micro optical fiber channel runs through the central axis of the welding wedge, and its inner wall is coated with an antireflection film. Its entrance end is connected to a light source and a spectral detector through a standard optical fiber connector.

3. The bonding wire welding method based on bare die interconnection according to claim 2, characterized in that, The real-time acquisition of the synchronous execution of multi-physics field signals specifically includes: continuously acquiring, through the piezoelectric transducer, the composite acoustic vibration signal generated by ultrasonic welding energy excitation and propagating between the welding blade and the bonding interface at a sampling rate of not less than 5 MHz; A monochromatic laser source is used to focus a coherent beam of a specific wavelength onto the contact area between the bonding wire and the solder pad below the working end face of the welding wedge via the entrance end of the micro fiber channel. Through the same entry end of the micro-fiber channel, the light signal reflected from the contact area is introduced into the spectral detector by a beam splitter, and the reflection spectral data is acquired at a frequency of not less than 10 kHz.

4. The bonding wire welding method based on bare die interconnection according to claim 3, characterized in that, The spectral analysis of the real-time acquired acoustic resonance response signal specifically includes: applying a short-time Fourier transform to the acquired time-domain acoustic vibration signal to generate a continuous time-frequency distribution map; In the time-frequency distribution diagram, identify and track first-order or multi-order resonance peaks related to the overall mode of the welding system; The resonant peak frequency at the start of welding is set as the reference frequency, and the offset of the current resonant peak frequency relative to the reference frequency is calculated in real time. The amplitude of the time-domain acoustic vibration signal is squared and then integrated within a time window to obtain the acoustic emission energy integral value that is directly related to the intensity of the acoustic emission event.

5. The bonding wire welding method based on bare die interconnection according to claim 4, characterized in that, The specific steps of solving the real-time acquired optical reflection spectrum signal include: the wavelength of the monochromatic laser source is pre-selected as a specific value that can effectively excite plasmon resonance on the surface of the pad metal material; The spectral detector records the curve of the reflected light intensity as a function of wavelength or incident angle in real time. On the curve, locate and trace the point of lowest reflectivity, i.e., the resonance peak, formed by surface plasmon resonance absorption; During the bonding process, the formation of intermetallic compounds in aluminum-gold or copper-gold alters the interfacial dielectric constant, causing a measurable wavelength shift in the resonance peak. By using a lookup table or mathematical function established in advance through calibration experiments, the wavelength shift of the resonance peak is directly mapped to the average thickness of the intermetallic compound layer, with a measurement resolution reaching the nanometer level.

6. The bonding wire welding method based on bare die interconnect according to claim 5, characterized in that, The target bonding state evolution model is a four-dimensional time-series database stored in the controller. It defines the standard values ​​that the resonant frequency offset, acoustic emission energy integral value, and intermetallic compound layer thickness should reach at each time step from time zero to completion in the ideal welding process. This model is constructed by in-situ monitoring and statistical analysis of a large number of welding samples with optimal mechanical and electrical properties.

7. The bonding wire welding method based on bare die interconnection according to claim 6, characterized in that, The process of generating dynamic adjustment instructions in real time through the model prediction controller specifically includes: the model prediction controller predicts the bonding state evolution trajectory for multiple future time steps within a finite time domain based on a system state space model that describes the dynamic relationship between welding process parameters and bonding state feature vectors. A quadratic objective function is constructed that includes tracking error and control energy consumption, wherein the tracking error is the difference between the predicted trajectory and the target bonding state evolution model, and the control energy consumption is the rate of change of ultrasonic power and bonding force; By solving the constrained optimization problem, a future control sequence that minimizes the quadratic objective function is calculated in each control cycle, and the first element of the sequence is output as the dynamic adjustment command for the current cycle.

8. The bonding wire welding method based on bare die interconnect according to claim 7, characterized in that, The system state-space model is represented as follows: ; in, The feature vector of the bonding state. This is a vector of welding process parameters. Here is the state transition matrix. As the input influence matrix, the and Obtained through system identification and experimental calibration.

9. The bonding wire welding method based on bare die interconnect according to claim 8, characterized in that, The constrained optimization problem satisfies the following constraints: ； ； in, To predict the first in the control sequence One control quantity, Its change, , , and These are the preset process parameter boundaries and variation rate limits.

10. The bonding wire welding method based on bare die interconnection according to claim 9, characterized in that, The welding process parameter execution mechanism includes a high-frequency ultrasonic power source and a high-precision force control actuator. The power adjustment accuracy of the high-frequency ultrasonic power source is not less than 0.5%, and the normal force control range of the high-precision force control actuator is 10 millinewtons to 500 millinewtons, with a resolution better than 0.1 millinewtons.

Images