Micro-resistance welding process and post-weld stress relief method for microelectronic components

Abstract

The application discloses a micro-resistance welding process and a post-welding stress relieving method for micro electronic components, and relates to the technical field of micro electronic manufacturing and material processing. The method comprises the following steps: obtaining the three-dimensional topography of a welding area and the material properties of lead wires, and constructing a dynamic resistance prediction model; generating a non-uniform welding energy release strategy comprising pre-pressing time, welding current pulse slope and maintaining time through a self-adaptive optimization framework; performing the strategy to heat and melt, and starting in-situ electromagnetic field assisted stress regulation within the solidification window period of the molten core; dynamically adjusting the electromagnetic field frequency and duty cycle by real-time collection of acoustic emission signals; and transferring the welding assembly to a gradient temperature control annealing platform for phased thermal cycle stress homogenization treatment. The application realizes accurate timing distribution of welding energy and in-situ active regulation of residual stress in the solidification process, and effectively improves the welding quality and reliability of micro electronic components.

Description

Technical Field

[0001] This invention relates to the field of microelectronics manufacturing and materials processing technology, specifically to a micro-resistance welding process for microelectronic components and a method for stress relief after welding. Background Technology

[0002] In the microresistance welding process of microelectronic components, the weld quality is directly affected by the dynamic resistance changes at the contact interface. Due to the tiny size of the welding area of ​​microelectronic components, the contact interface is typically composed of numerous micro-protrusions, resulting in an extremely uneven initial contact resistance distribution that undergoes dramatic time-varying evolution due to the thermo-electric-mechanical coupling during welding. Currently, conventional microresistance welding processes typically use fixed welding parameters (such as constant welding current, fixed pre-pressure time, and fixed holding time) for energy release, lacking real-time response and adaptive adjustment capabilities to the dynamic evolution of the contact interface resistance. This makes it difficult to precisely match the welding energy to the actual needs of the contact interface, easily leading to welding defects such as localized overheating, irregular molten nucleus morphology, or lack of fusion. Furthermore, during the solidification of the molten nucleus during welding, the thermal contraction of the material and the volume effect of solid-state phase transition generate significant residual stress within the weld joint. Existing post-weld stress relief methods mostly employ overall annealing or natural aging treatment. These methods cannot intervene and actively control the emerging stress field in real time during the molten nucleus solidification window, leading to the accumulation of residual stress in the weld joint and affecting the long-term reliability of the welded assembly. For microelectronic components, their structural size is small and their material composition is diverse. Conventional overall annealing treatment can easily cause component performance degradation or dimensional deformation. Existing stress relief technology cannot effectively meet the stress homogenization requirements of miniaturized and high-precision welded components. Summary of the Invention

[0003] This invention aims to provide a micro-resistance welding process and post-weld stress relief method for microelectronic components, addressing the technical problem that existing conventional micro-resistance welding processes, due to their use of fixed welding parameters, cannot adapt to the dynamic resistance evolution process of the contact interface, resulting in a mismatch between welding energy release and actual interface requirements. Simultaneously, this invention aims to address the technical problem that existing post-weld stress relief methods cannot perform real-time in-situ control of the forming residual stress during the solidification window of the molten nucleus, leading to stress accumulation within the weld joint and affecting component reliability.

[0004] The objective of this invention can be achieved through the following technical solutions: This invention provides a micro-resistance welding process and post-weld stress relief method for microelectronic components, comprising: acquiring three-dimensional topographic data of the welding area of ​​the microelectronic component and material property parameters of the leads to be welded, and jointly constructing a dynamic resistance prediction model of the welding contact interface; inputting the contact resistance distribution sequence output by the dynamic resistance prediction model into a pulse welding parameter adaptive optimization framework for multi-objective welding waveform modulation processing to generate a non-uniform welding energy release strategy including pre-pressure time, welding current pulse slope, and holding time; executing the non-uniform welding energy release strategy to perform instantaneous heating and melting treatment on the welding contact interface, and simultaneously initiating in-situ electromagnetic field-assisted stress control processing during the solidification window of the molten core, wherein the in-situ electromagnetic field-assisted stress control processing dynamically adjusts the electromagnetic field frequency and duty cycle by real-time acquisition of the solid-state phase change acoustic emission signal of the molten core; transferring the welded assembly after stress control treatment to a gradient temperature-controlled annealing platform for staged thermal cycling stress homogenization treatment to obtain a finished microelectronic component welded product with eliminated residual stress.

[0005] As a technical solution of the present invention, the jointly constructed dynamic resistance prediction model for the welding contact interface includes: performing statistical processing on the distribution of micro-protrusions at the contact interface from the three-dimensional topography data of the welding area to generate a true contact area distribution map of the contact interface; extracting the conductivity and hardness correlation features from the material property parameters of the lead to be welded to obtain a spatial heterogeneity distribution matrix of material resistivity; performing a grid-by-grid multiplication operation on the true contact area distribution map and the spatial heterogeneity distribution matrix of material resistivity to generate an initial contact resistance distribution field; and calling a pre-trained time-varying resistance evolution network to perform thermo-electric-mechanical coupled iterative update processing on the initial contact resistance distribution field to output the contact resistance distribution sequence. This method can accurately predict the dynamic evolution of contact resistance during welding, providing a reliable input basis for the formulation of subsequent energy release strategies, thereby improving the consistency of welding quality.

[0006] Preferably, the step of calling the pre-trained time-varying resistance evolution network to perform thermo-electric-mechanical coupled iterative update processing on the initial contact resistance distribution field includes: inputting the initial contact resistance distribution field into the Joule heat generation layer of the time-varying resistance evolution network, calculating the instantaneous heat power density at each grid point when the welding current passes through; transferring the instantaneous heat power density to the heat diffusion simulation layer of the time-varying resistance evolution network, updating the temperature distribution field of the contact interface, and performing iterative calculation using the finite difference heat diffusion update formula, wherein the finite difference heat diffusion update formula is: in, For grid points In the Temperature values ​​at each time step, in K; The time step is expressed in seconds (s). The average density of the contact interface material is expressed in kg / m³. This is the specific heat capacity at constant pressure of the material, expressed in J / (kg·K). and These represent the spatial step size of the grid points in the x and y directions, respectively, in meters; For grid points In the The instantaneous heat power density at each time step, in W / m²; The thermal conductivity of the material is expressed in W / (m·K). The temperature field is represented by the discrete Laplace operator value, in K / m². This temperature distribution field is input into the thermal softening effect layer of the time-varying resistance evolution network. The local contact pressure at each grid point is updated based on the temperature-to-material hardness mapping. The updated local contact pressure is then fed back to the contact resistance correction layer of the time-varying resistance evolution network, recalculating the contact resistance value at each grid point to form the next time step output in the contact resistance distribution sequence. This iterative process accurately simulates the interface evolution behavior under the coupling of thermo-electrical-mechanical multiphysics fields, enabling the resistance prediction results to have dynamic adaptability.

[0007] As a technical solution of the present invention, the time-varying resistance evolution network is a time-series prediction model based on a long short-term memory network. Its input features include the contact resistance distribution field and welding current density field at the current moment. By introducing time-series memory capability, it is possible to capture the long-term dependent variation law of contact resistance and improve prediction accuracy.

[0008] As a technical solution of the present invention, the generation of a non-uniform welding energy release strategy including preload time, welding current pulse slope, and holding time includes: extracting the temporal variation trend features of the contact resistance distribution sequence to identify the critical inflection point when the contact resistance transitions from an initial high resistance state to a low resistance state; using the critical inflection point as a boundary, dividing the optimization objective of the pulse welding parameter adaptive optimization framework into a preheating stage melt nucleus initiation objective and a main welding stage melt nucleus expansion objective; under the constraint of the preheating stage melt nucleus initiation objective, performing discrete search processing on the initial combination of preload time and welding current pulse slope to determine the preload time and welding current pulse slope that make the contact resistance decrease rate reach a preset level; under the constraint of the main welding stage melt nucleus expansion objective, using the melt nucleus size at the end of the preheating stage as the initial state, performing gradient ascending optimization processing on the holding time and welding current peak value to obtain the holding time and welding current peak value. By segmenting and optimizing the energy release parameters, the differentiated energy requirements of the melt nucleus initiation and expansion stages can be accurately matched, which helps to avoid spatter and excessive heat effects and improve the mechanical properties of the welded joint.

[0009] Preferably, the pre-compression time and welding current pulse slope for achieving a preset contact resistance decrease rate include: constructing a crushing response curve between the pre-compression time and the crushing degree of the contact micro-protrusion; extracting the minimum pre-compression time required for the contact resistance to enter the linear decrease segment from the crushing response curve as a first candidate pre-compression time; constructing a melt response curve between the welding current pulse slope and the first appearance time of the molten nucleus at the contact interface; extracting the current pulse slope corresponding to the first appearance time of the molten nucleus from the melt response curve as a first candidate slope; pairing the first candidate pre-compression time and the first candidate slope, and inputting them into the resistance time-varying evolution network for virtual welding simulation; verifying whether the contact resistance decrease rate reaches the preset level during the virtual welding simulation; if not, synchronously increasing and adjusting the first candidate pre-compression time and the first candidate slope. Through a verification mechanism combining data-driven and physical simulation, it can be ensured that the determined pre-compression time and current pulse slope effectively achieve the required resistance decrease rate under actual process conditions, thereby providing a guarantee for the stable initiation of the molten nucleus.

[0010] As a technical solution of the present invention, the non-uniform welding energy release strategy is used to instantaneously heat and melt the welding contact interface, and in-situ electromagnetic field-assisted stress control is simultaneously initiated during the solidification window of the molten nucleus. This includes: sequentially outputting welding current waveforms according to the pre-pressure time, welding current pulse slope, and maintenance time in the non-uniform welding energy release strategy; resistively heating the welding contact interface until a complete molten nucleus is formed; monitoring the dynamic resistance signal of the welding circuit in real time, and determining that the molten nucleus has entered the solidification window when the first derivative of the dynamic resistance signal changes from negative to positive; activating an electromagnetic field generator during the solidification window, and applying an alternating electromagnetic field to the molten nucleus region according to the initial electromagnetic field frequency and duty cycle; continuously acquiring the acoustic emission signal waveform released during the solidification of the molten nucleus during the application of the alternating electromagnetic field, and extracting the acoustic emission event rate and average amplitude characteristics from the acoustic emission signal waveform. By accurately identifying the solidification window and acquiring the acoustic emission signal in real time, stress control can be initiated in a timely manner and information on the microstructure evolution during the solidification process can be obtained, providing a basis for subsequent dynamic adjustment of electromagnetic field parameters.

[0011] Preferably, the determination of the first derivative of the dynamic resistance signal turning from negative to positive is based on the zero-crossing detection of the derivative after smoothing by a sliding window. This smoothing process can suppress noise interference and improve the accuracy of solidification window determination.

[0012] As a technical solution of the present invention, the in-situ electromagnetic field-assisted stress regulation processing dynamically adjusts the electromagnetic field frequency and duty cycle by real-time acquisition of the solid-state phase transition acoustic emission signal of the molten nucleus. This includes: normalizing the acoustic emission event rate and average amplitude characteristics and then inputting them into a fuzzy rule inference engine. The output variables of the fuzzy rule inference engine are the electromagnetic field frequency adjustment coefficient and the duty cycle adjustment coefficient. When the acoustic emission event rate exceeds a high event rate threshold and the average amplitude exceeds a high amplitude threshold, the electromagnetic field frequency adjustment coefficient output by the fuzzy rule inference engine is a frequency boost coefficient, and the duty cycle adjustment coefficient is a duty cycle reduction coefficient. and duty cycle adjustment factor Calculate according to the following formulas respectively: in, is the electromagnetic field frequency adjustment coefficient, which is dimensionless; The membership degree is defined as the acoustic emission event rate being higher than the high event rate threshold, and its value ranges from [0,1]. This is the frequency boost factor, dimensionless, and takes a value greater than 1. The membership degree is defined as the acoustic emission event rate being lower than the low event rate threshold, and its value ranges from [0,1]. This is a frequency reduction coefficient, dimensionless, and its value is less than 1. This is the duty cycle adjustment factor, which is dimensionless. The membership degree is defined as the average amplitude being higher than the high amplitude threshold, and its value ranges from [0,1]. This is a duty cycle reduction factor, dimensionless, and its value is less than 1. The membership degree is defined as the average amplitude being lower than the low amplitude threshold, and its value ranges from [0,1]. The duty cycle enhancement factor is dimensionless and has a value greater than 1. The electromagnetic field frequency is updated based on the product of the electromagnetic field frequency adjustment factor and the current electromagnetic field frequency, and the duty cycle is updated based on the product of the duty cycle adjustment factor and the current duty cycle. When the acoustic emission event rate is lower than the low event rate threshold, the electromagnetic field frequency adjustment factor output by the fuzzy rule inference engine becomes the frequency maintenance factor, and the duty cycle adjustment factor becomes the duty cycle maintenance factor. This fuzzy inference mechanism can optimize electromagnetic field parameters in real time based on the acoustic emission characteristics of crack initiation or dislocation movement during solidification, thereby suppressing the formation of welding residual stress and improving weld reliability.

[0013] Preferably, the fuzzy rule inference engine uses a triangular membership function as its input membership function and a single-valued membership function as its output membership function. This design simplifies the computational complexity of inference while maintaining sufficient adjustment sensitivity.

[0014] As a technical solution of the present invention, the step of transferring the stress-controlled welded assembly to a gradient temperature-controlled annealing platform for staged thermal cycling stress homogenization includes: placing the welded assembly on the heating station of the gradient temperature-controlled annealing platform, heating it from room temperature to a first holding temperature at a first heating rate, and maintaining it at the first holding temperature for a first holding time to complete the first stage annealing treatment; heating it from the first holding temperature to a second holding temperature at a second heating rate, and maintaining it at the second holding temperature for a second holding time, wherein the second holding temperature is higher than the first holding temperature, to complete the second stage annealing treatment; cooling it from the second holding temperature to the first holding temperature at a first cooling rate, and maintaining it at the first holding temperature for a third holding time to complete the first intermediate cooling pause treatment; and cooling it from the first holding temperature to room temperature at a second cooling rate to complete the final cooling treatment; wherein the second heating rate is less than the first heating rate, and the second cooling rate is less than the first cooling rate. This multi-stage thermal cycling strategy can release internal stress in different temperature ranges in stages, avoid thermal shock cracks caused by rapid heating or cooling, and achieve gradual homogenization of residual stress.

[0015] As a technical solution of the present invention, the first holding temperature, the second holding temperature, the first holding time, the second holding time, and the third holding time are determined by the material thermal expansion coefficient of the welding component and the size of the molten core, including: extracting the material thermal expansion coefficients of the lead to be welded and the substrate of the microelectronic component, and calculating the absolute value of the difference between the two thermal expansion coefficients as the thermal mismatch coefficient; measuring the equivalent diameter and penetration depth of the molten core, and using the product of the equivalent diameter and penetration depth as the volume characteristic value of the molten core; inputting the thermal mismatch coefficient and the volume characteristic value of the molten core into a pre-constructed annealing parameter lookup table, and matching the first holding temperature range, the second holding temperature range, the first holding time range, the second holding time range, and the third holding time range from the annealing parameter lookup table; and using the centroid method to select the median from the first holding temperature range, the second holding temperature range, the first holding time range, the second holding time range, and the third holding time range as the first holding temperature, the second holding temperature, the first holding time, the second holding time, and the third holding time range, respectively, as the first holding temperature, the second holding temperature, the first holding time, the second holding time, and the third holding time. This method allows for adaptive setting of annealing parameters based on the material matching and geometry of the actual weld joint, thereby improving the targeting and effectiveness of stress relief.

[0016] As a technical solution of the present invention, the gradient temperature-controlled annealing platform applies low-frequency mechanical vibration to the welded assembly during the staged thermal cycling stress homogenization process to assist in stress release treatment. This includes: installing a piezoelectric ceramic vibration generator below the heating station of the gradient temperature-controlled annealing platform; activating the piezoelectric ceramic vibration generator when the temperature of the welded assembly rises to the first holding temperature, transmitting a mechanical vibration wave perpendicular to the weld interface to the welded assembly at a first vibration frequency and a first vibration amplitude; switching the vibration frequency of the piezoelectric ceramic vibration generator to a second vibration frequency, where the second vibration frequency is higher than the first vibration frequency, as the temperature of the welded assembly decreases from the second holding temperature to the first holding temperature; and turning off the piezoelectric ceramic vibration generator when the temperature of the welded assembly is lower than the first holding temperature but higher than the glass transition temperature. By superimposing mechanical vibration during the thermal cycling process, grain boundary slip and dislocation movement can be promoted, accelerating stress release. Simultaneously, using different vibration frequencies at different temperature ranges can avoid excessive vibration leading to fatigue damage at the weld interface, which is beneficial for further improving stress relief efficiency while ensuring the integrity of the weld joint.

[0017] The beneficial effects of this invention are: The non-uniform welding energy release strategy is based on the contact resistance distribution sequence output by the dynamic resistance prediction model. A multi-objective welding waveform modulation process is performed using an adaptive optimization framework for pulse welding parameters to generate a combination of welding parameters including preload time, welding current pulse slope, and holding time. The dynamic resistance prediction model utilizes the three-dimensional topography data of the welding area and the material properties of the lead wire to be welded to jointly construct a contact resistance distribution sequence that reflects the distribution of micro-protrusions at the contact interface, the spatial heterogeneity of material resistivity, and the thermo-electric-mechanical coupling evolution mechanism. This sequence can accurately describe the temporal characteristics of the contact resistance transitioning from an initial high-resistivity state to a low-resistivity state during welding. Based on these temporal characteristics, the adaptive optimization framework identifies the critical inflection point of the contact resistance and divides the optimization objectives into the preheating stage melt nucleus initiation objective and the main welding stage melt nucleus expansion objective, thereby optimizing the preload time and welding current pulse slope, and the holding time and welding current peak value, respectively. This energy release method based on real-time feedback of dynamic resistance enables the timing and magnitude of welding energy input to precisely match the evolution of the contact interface resistance, avoiding the problems of premature energy concentration or insufficient energy release that are prone to occur in conventional fixed-parameter welding, and improving the uniformity and stability of the molten nucleus initiation and expansion process.

[0018] In-situ electromagnetic field-assisted stress control is initiated simultaneously within the solidification window of the molten nucleus. The electromagnetic field frequency and duty cycle are dynamically adjusted using real-time acquired acoustic emission signals from the solid-state phase transition of the molten nucleus. This process monitors the dynamic resistance signal of the welding circuit, using the change in the first derivative of the dynamic resistance signal from negative to positive as the criterion for determining when the molten nucleus enters the solidification window, thus accurately determining the timing of stress control intervention. Applying an alternating electromagnetic field within the solidification window allows the electromagnetic force to act on the molten metal at the solidification front, influencing dendrite growth morphology and solute distribution, thereby altering the stress distribution caused by solidification shrinkage. By continuously acquiring acoustic emission event rates and average amplitude characteristics, and employing a fuzzy rule inference engine to adjust the electromagnetic field frequency and duty cycle in real time, the excitation parameters of the electromagnetic field can be adaptively adjusted to follow the dynamic changes in the solidification process. When the acoustic emission event rate and average amplitude reflect severe solidification shrinkage or high stress concentration, the fuzzy rule inference engine outputs corresponding frequency boosting and duty cycle reduction coefficients to enhance the stirring and homogenizing effect of the electromagnetic field on the solidified structure. This allows for proactive intervention and release of stress at the initial stage of stress formation, preventing excessive stress accumulation inside the welded joint. Attached Figure Description

[0019] The invention will now be further described with reference to the accompanying drawings.

[0020] Figure 1 This is a process flow diagram for micro-resistance welding and post-weld stress relief; Figure 2 This is a flowchart of a staged thermal cycling stress homogenization process. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See Figure 1This invention provides a micro-resistance welding process and post-weld stress relief method for microelectronic components. The method includes: acquiring three-dimensional topographic data of the welding area of ​​the microelectronic component and material property parameters of the leads to be welded, and jointly constructing a dynamic resistance prediction model of the welding contact interface; inputting the contact resistance distribution sequence output by the dynamic resistance prediction model into a pulse welding parameter adaptive optimization framework for multi-objective welding waveform modulation processing to generate a non-uniform welding energy release strategy including pre-pressure time, welding current pulse slope, and holding time; executing the non-uniform welding energy release strategy to perform instantaneous heating and melting treatment on the welding contact interface, and simultaneously initiating in-situ electromagnetic field-assisted stress control processing during the solidification window of the molten core, wherein the in-situ electromagnetic field-assisted stress control processing dynamically adjusts the electromagnetic field frequency and duty cycle by real-time acquisition of the solid-state phase change acoustic emission signal of the molten core; transferring the welded assembly after stress control treatment to a gradient temperature-controlled annealing platform for staged thermal cycling stress homogenization treatment to obtain a finished microelectronic component welded product with eliminated residual stress.

[0023] Example 1 In practical implementation, a dynamic resistance prediction model for the welding contact interface is jointly constructed. First, the distribution of micro-protrusions at the contact interface is statistically processed from the three-dimensional morphology data of the welding area. Gaussian filtering is used to denoise the three-dimensional morphology data of the welding area, and the height distribution of micro-protrusions at the contact interface is extracted. Areas where the height of micro-protrusions exceeds a preset threshold of 0.5 micrometers are marked as contact points. The area ratio of contact points within each grid point is statistically analyzed to generate a true contact area distribution map of the contact interface. The value of each grid point represents the proportion of the true contact area at that location, which is dimensionless.

[0024] The material property parameters of the leads to be soldered are subjected to conductivity and hardness correlation feature extraction processing. Conductivity and hardness data are extracted from the material property parameters. Using a resistivity-conductivity-hardness mapping function pre-fitted based on material sample experiments, the material resistivity at each spatial location is calculated to obtain the spatial heterogeneity distribution matrix of material resistivity, where the value of each grid point represents the material resistivity at that location, in ohm-meters.

[0025] The actual contact area distribution map is multiplied by the spatial heterogeneity distribution matrix of material resistivity on a grid-by-grid basis. This grid-by-grid multiplication means that for each identical grid point coordinate, the contact area ratio of that grid point in the actual contact area distribution map is multiplied by the material resistivity of that grid point in the spatial heterogeneity distribution matrix of material resistivity. The product is used as the initial contact resistance value for that grid point. The initial contact resistance values ​​of all grid points constitute the initial contact resistance distribution field, with units of ohms.

[0026] A pre-trained time-varying resistance evolution network is invoked to perform thermo-electric-mechanical coupled iterative update processing on the initial contact resistance distribution field, outputting the contact resistance distribution sequence. The time-varying resistance evolution network is a time-series prediction model based on a long short-term memory (LSTM) network. Its input features include the contact resistance distribution field and welding current density field at the current time step, and its output is the contact resistance distribution field at the next time step. The structure of the time-varying resistance evolution network includes an input layer, a LSM layer, and a fully connected output layer. The number of neurons in the LSM layer is set to 128. The number of neurons in the fully connected output layer is equal to the total number of grid points, and a linear activation function is used. The training of the time-varying resistance evolution network uses the contact resistance distribution field and welding current density field collected in continuous time steps during historical welding processes as samples, and the actually measured contact resistance distribution field at the next time step as the label. Training is performed by minimizing the mean square error loss function, with 200 training epochs and a learning rate of 0.001.

[0027] The specific steps for using a pre-trained time-varying resistance evolution network to perform thermo-electric-mechanical coupled iterative update processing on the initial contact resistance distribution field include: inputting the initial contact resistance distribution field into the Joule heat generation layer of the time-varying resistance evolution network. The Joule heat generation layer calculates the instantaneous heat power density at each grid point when the welding current passes through, based on the welding current density field and the initial contact resistance distribution field, according to Joule's law. The unit is watts per square meter. The instantaneous heat power density... The calculation formula is ,in For grid points In the The current density value at each time step, in amperes per square meter. For grid points In the The contact resistance value at each time step, in ohms.

[0028] The instantaneous heat power density is transferred to the thermal diffusion simulation layer of the time-varying resistance evolution network. The thermal diffusion simulation layer receives the instantaneous heat power density, updates the temperature distribution field at the contact interface, and performs iterative calculations using a finite-difference thermal diffusion update formula. The finite-difference thermal diffusion update formula is as follows: in, For grid points In the Temperature values ​​at each time step, in Kelvin; Set the time step to 1. Second; The average density of the contact interface material is taken from the arithmetic mean of the densities of the lead to be soldered and the substrate material of the microelectronic component, in kilograms per cubic meter. The isobaric specific heat capacity of the material is taken from the average specific heat capacity of the materials at the contact interface, and the unit is joules per kilogram Kelvin. and These represent the spatial step sizes of the grid points in the x and y directions, respectively, both set to... rice; For grid points In the The instantaneous heat power density at each time step, in watts per square meter; The thermal conductivity of the material is taken from the average thermal conductivity of the materials at the contact interface, and the unit is watts per meter Kelvin. The discrete values ​​of the Laplace operator for the temperature field are calculated using a five-point difference scheme, with units of Kelvin per square meter.

[0029] The temperature distribution field is input into the thermal softening effect layer of the time-varying resistance evolution network. The thermal softening effect layer includes a pre-constructed temperature-hardness mapping function, which is obtained by fitting experimental material data and describes the relationship between temperature and material hardness. Based on the temperature-hardness mapping function, the local contact pressure at each grid point is updated according to the temperature value. The method for updating the local contact pressure is as follows: substituting the current temperature into the temperature-hardness mapping function to obtain the material hardness at that temperature, and then applying the contact mechanics formula... Calculate the local contact pressure, where For grid points In the Local contact pressure at each time step, in Pascals. For grid points In the The material hardness at each time step is expressed in Pascals. For grid points The actual contact area, in square meters.

[0030] The updated local contact pressure is fed back to the contact resistance correction layer of the time-varying resistance evolution network. The contact resistance correction layer uses the contact resistance-pressure relationship formula based on the updated local contact pressure. Recalculate the contact resistance value for each grid point, where For grid points In the The contact resistance value at each time step, in ohms. For grid points In the The material resistivity at each time step, in ohmmeters. The contact pressure correction factor is dimensionless and is set to... The reciprocal of Pascal's coefficient is set based on empirical values ​​for typical metal contact pairs. The recalculated contact resistance values ​​for all grid points constitute the contact resistance distribution field for the next time step, serving as an output in the contact resistance distribution sequence. The iterative process from the Joule heating layer to the contact resistance correction layer is repeated until a preset number of iterations is reached, outputting the complete contact resistance distribution sequence.

[0031] Example 2 In specific implementation, a non-uniform welding energy release strategy is generated, including pre-pressure time, welding current pulse slope, and holding time. The temporal trend feature extraction process is performed on the contact resistance distribution sequence to identify the critical inflection point when the contact resistance transitions from an initial high-resistance state to a low-resistance state. This temporal trend feature extraction process employs a sliding window linear regression method, with a window length of 5 time steps. The global average value of the contact resistance distribution sequence within each window is linearly fitted. When the fitting slope changes from negative to positive, the time step corresponding to the transition is marked as the critical inflection point.

[0032] Using the critical inflection point as a boundary, the optimization objectives of the pulse welding parameter adaptive optimization framework are divided into the preheating stage melt nugget initiation objective and the main welding stage melt nugget expansion objective. The preheating stage melt nugget initiation objective is defined as the melt nugget appearing for the first time before the critical inflection point, and the contact resistance decreasing at a preset level. The main welding stage melt nugget expansion objective is defined as the melt nugget diameter expanding to a target value after the critical inflection point, with the target value set at 90% of the welding area width.

[0033] Under the constraint of the preheating section molten nucleus initiation target, a discrete search is performed on the initial combination of preload time and welding current pulse slope. The search step size for the discrete search is set to: preload time step size 0.5 ms, welding current pulse slope step size 0.1 amperes per ms. The search range is set to: preload time from 1 ms to 10 ms, welding current pulse slope from 0.5 amperes per ms to 5 amperes per ms. Under each discrete combination, the preload time and welding current pulse slope are input into the resistance time-varying evolution network for virtual welding simulation. The maximum value of the contact resistance decrease rate during the virtual welding simulation is recorded. If the maximum value reaches the preset level of 0.5 ohms per second, the search stops, and the current combination is determined to be the preload time and welding current pulse slope that makes the contact resistance decrease rate reach the preset level.

[0034] Under the constraint of the target molten core expansion in the main welding section, the molten core size at the end of the preheating section is used as the initial state, and a gradient-based optimization process is performed on the holding time and welding current peak value. The gradient direction of the gradient-based optimization process is determined based on the partial derivatives of the molten core expansion rate with respect to the holding time and welding current peak value. The partial derivatives are calculated using the finite difference method, with the difference step size set to 0.1 ms for the holding time and 0.5 amperes for the welding current peak value. In each gradient-based step, the current holding time and welding current peak value are input into the time-varying resistance evolution network to simulate and obtain the molten core diameter. The difference between the molten core diameter and the target value is calculated. If the difference is less than the threshold of 0.5 micrometers, the iteration stops, and the current holding time and welding current peak value are output as the holding time and welding current peak value.

[0035] In some embodiments, the preload time and welding current pulse slope for achieving a preset contact resistance decrease rate are determined using the following method: A crushing response curve is constructed between the preload time and the degree of contact micro-protrusion crushing. This crushing response curve is obtained by inputting different preload times into the resistance time-varying evolution network for virtual welding simulation, extracting the contact micro-protrusion crushing degree value at the corresponding preload time. The crushing degree is defined as the ratio of the actual contact area to the nominal area of ​​the contact interface. The minimum preload time required for the contact resistance to enter the linear decrease phase is extracted from the crushing response curve as the first candidate preload time. The criterion for determining whether the contact resistance enters the linear decrease phase is: on the crushing response curve, the preload time corresponding to the first time the rate of change of the crushing degree is less than 0.01 dimensionless milliseconds as the preload time increases.

[0036] A melt response curve is constructed relating the welding current pulse slope to the first appearance time of the molten nucleus at the contact interface. This melt response curve is obtained by inputting different welding current pulse slopes into a time-varying resistance evolution network for virtual welding simulation, recording the time from the start of the welding current to the first appearance of the molten nucleus. The first appearance of the molten nucleus is defined as the temperature at the center grid point of the contact interface first exceeding the material's melting point. The current pulse slope corresponding to the first appearance time of the molten nucleus is extracted from the melt response curve as the first candidate slope. The current pulse slope corresponding to the first appearance time of the molten nucleus refers to the welding current pulse slope corresponding to the minimum first appearance time of the molten nucleus on the melt response curve.

[0037] The first candidate preload time and the first candidate slope are paired and input into the time-varying resistance evolution network for virtual welding simulation. The simulation is then used to verify whether the contact resistance decrease rate reaches a preset level of 0.5 ohms per second. If not, the first candidate preload time and the first candidate slope are adjusted synchronously and incrementally. The step size for this synchronous adjustment is set as follows: the first candidate preload time increases by 0.1 milliseconds each time, and the first candidate slope increases by 0.05 amperes per millisecond each time. After each adjustment, the virtual welding simulation is repeated until the contact resistance decrease rate reaches the preset level.

[0038] Example 3 In specific implementation, the non-uniform welding energy release strategy is executed to perform instantaneous heating and melting treatment on the welding contact interface, and in-situ electromagnetic field-assisted stress control treatment is simultaneously initiated during the solidification window of the molten nucleus, including the following steps.

[0039] The welding current waveform is output sequentially according to the pre-pressure time, welding current pulse slope, and holding time in the non-uniform welding energy release strategy. The welding power source outputs a constant pre-pressure based on the pre-pressure time. After the pre-pressure time ends, the welding power source linearly increases the welding current from zero according to the welding current pulse slope. When the welding current reaches a peak current determined by the welding current pulse slope and the holding time, this peak current is maintained constant during the holding time. Then, the welding current is cut off, and the welding interface is subjected to resistance heating until a complete molten nucleus is formed. The criterion for a complete molten nucleus is that the temperature of the central region of the contact interface exceeds the melting point of the material, and the area of ​​the molten region covers more than 95% of the welding area.

[0040] The dynamic resistance signal of the welding circuit is monitored in real time. A precision resistor is connected in series in the welding circuit. The dynamic resistance signal is calculated by collecting the ratio of the voltage drop across the precision resistor to the circuit current. The dynamic resistance signal is smoothed using a sliding window with a window length of 5 sampling points and a linear smoothing filter. The first derivative of the smoothed signal is calculated. When the first derivative changes from a negative value to a positive value, the molten nucleus is determined to have entered the solidification window period. The solidification window period is defined as the time from the moment the first derivative crosses zero until the molten nucleus is completely solidified. The criterion for determining complete solidification of the molten nucleus is that the dynamic resistance signal recovers to 98% of its steady-state value.

[0041] An electromagnetic field generator is activated during the solidification window. The electromagnetic field generator is a hollow coil, fitted around the outside of the welding assembly. An alternating electromagnetic field is applied to the molten core region according to the initial electromagnetic field frequency and initial duty cycle. The initial electromagnetic field frequency is set to 50 kHz, and the initial duty cycle is set to 50%.

[0042] During the application of an alternating electromagnetic field, the acoustic emission signal waveforms released during the solidification of the molten nucleus are continuously acquired. A broadband acoustic emission sensor, model R15, is used, fixed to the surface of the welding assembly substrate, with a sampling rate set to 1 MHz. The acoustic emission event rate and average amplitude characteristics are extracted from the acoustic emission signal waveforms. The acoustic emission event rate is defined as the number of acoustic emission events exceeding a preset threshold of 0.1 volts per second, and the average amplitude is defined as the arithmetic mean of the peak amplitudes of all acoustic emission events.

[0043] The determination of the first derivative of the dynamic resistance signal changing from negative to positive is based on the zero-crossing detection of the derivative after sliding window smoothing. The specific method is as follows: for the dynamic resistance signal sequence after sliding window smoothing, the first derivative is calculated once every 10 sampling points. When the first derivative calculated three times consecutively is greater than 0, and the first derivative calculated in the previous time is less than 0, it is determined to be a zero-crossing.

[0044] The in-situ electromagnetic field-assisted stress control process dynamically adjusts the electromagnetic field frequency and duty cycle by real-time acquisition of the solid-state phase transition acoustic emission signal of the molten core, including the following steps.

[0045] The acoustic emission event rate and average amplitude characteristics are normalized. The normalization process uses a maximum-minimum normalization method, mapping the acoustic emission event rate and average amplitude to the 0-1 interval. The normalized acoustic emission event rate and average amplitude are then input variables into the fuzzy rule inference engine. The output variable of the fuzzy rule inference engine is the electromagnetic field frequency adjustment coefficient. and duty cycle adjustment factor The fuzzy rule inference engine comprises two input variables and two output variables. The membership functions of the input variables are all triangular membership functions, while the membership functions of the output variables are all single-valued membership functions. The parameters of the triangular membership functions are set as follows: the vertex of the triangle corresponding to the low event rate threshold is at 0.2, and the vertex of the triangle corresponding to the high event rate threshold is at 0.8; the vertex of the triangle corresponding to the low amplitude threshold is at 0.3, and the vertex of the triangle corresponding to the high amplitude threshold is at 0.7. The output value of the single-valued membership function includes a frequency boost coefficient. Set to 1.2, frequency reduction factor Set to 0.8, frequency maintenance factor to 1.0; duty cycle reduction factor. Set to 0.8, duty cycle boost factor Set it to 1.2, and set the duty cycle maintenance factor to 1.0.

[0046] When the acoustic emission event rate exceeds the high event rate threshold of 0.8 (normalized value) and the average amplitude exceeds the high amplitude threshold of 0.7 (normalized value), the electromagnetic field frequency adjustment coefficient output by the fuzzy rule inference engine... and duty cycle adjustment factor Calculate according to the following formulas respectively: in, The membership degree is defined for acoustic emission event rates that are higher than the high event rate threshold, with a value range of [0,1], and is calculated using the triangular membership function. This is the frequency boost factor, dimensionless, set to 1.2; The membership degree is defined for acoustic emission event rates below the low event rate threshold, with a value range of [0,1], and is calculated using the triangular membership function. The frequency reduction factor is dimensionless and set to 0.8. The membership degree is defined for average amplitudes that are higher than the high amplitude threshold, and its value ranges from [0,1]. It is calculated using the triangular membership function. The duty cycle reduction factor is dimensionless and set to 0.8. The membership degree is the average amplitude below the low amplitude threshold, with a value range of [0,1], and is calculated by the triangular membership function. This is the duty cycle boost factor, dimensionless, set to 1.2.

[0047] According to the electromagnetic field frequency adjustment coefficient The electromagnetic field frequency is updated by multiplying the current electromagnetic field frequency by the frequency of the electromagnetic field, and the duty cycle adjustment coefficient is used as the basis for this update. The duty cycle is updated by multiplying the current duty cycle by the current duty cycle. The updated electromagnetic field frequency is limited to the range of 20 kHz to 100 kHz, and the updated duty cycle is limited to the range of 10% to 90%.

[0048] When the acoustic emission event rate is lower than the low event rate threshold of 0.2 (normalized value), the electromagnetic field frequency adjustment coefficient output by the fuzzy rule inference engine is a frequency maintenance coefficient of 1.0, and the duty cycle adjustment coefficient is a duty cycle maintenance coefficient of 1.0. At this time, the electromagnetic field frequency and duty cycle remain unchanged.

[0049] The fuzzy rule inference engine uses a triangular membership function as its input membership function and a single-valued membership function as its output membership function. The mathematical expression for the triangular membership function is: Where parameters For the low event rate threshold membership function of acoustic emission event rate, corresponding to the left endpoint, vertex, and right endpoint of the base of the triangle, respectively, we set... For high event rate threshold membership functions, set... For the low-amplitude threshold membership function of the average amplitude, set... For high-amplitude threshold membership functions, set... The output value of a single-valued membership function is a constant, as described above.

[0050] Example 4 See Figure 2 In practice, the welded components, after stress regulation treatment, are transferred to a gradient temperature-controlled annealing platform for staged thermal cycling stress homogenization. The gradient temperature-controlled annealing platform includes a programmable temperature controller capable of controlling the temperature of the heating station according to a preset temperature-time curve. The processing steps include: The welding assembly is placed on the heating station of the gradient temperature-controlled annealing platform and heated at a first heating rate. Heating from room temperature to the first holding temperature and at the first insulation temperature Maintain first heat preservation time The first stage of annealing is completed. The first heating rate... The setting is 10 degrees Celsius per minute, and the room temperature is 25 degrees Celsius.

[0051] At the second heating rate From the first insulation temperature Heat to the second holding temperature At the second insulation temperature Maintain the second insulation time The second insulation temperature Higher than the first insulation temperature The second stage of annealing is then completed. The second heating rate... Set to 5 degrees Celsius per minute, which is less than the first heating rate. .

[0052] At the first cooling rate From the second insulation temperature Cool down to the first insulation temperature At the first insulation temperature Maintain the third insulation time The first intermediate cooling and holding process is completed. The first cooling rate... Set to 5 degrees Celsius per minute.

[0053] At the second cooling rate From the first insulation temperature Cool to room temperature to complete the final cooling process. Second cooling rate Set to 3 degrees Celsius per minute, which is less than the first cooling rate. .

[0054] First insulation temperature Second insulation temperature First heat preservation time Second heat preservation time and the third heat preservation time The size of the weld nugget is determined by combining the coefficient of thermal expansion of the welded component's material with the size of the weld nugget. The determination steps are as follows: Extract the linear thermal expansion coefficient of the lead material to be soldered The coefficient of linear thermal expansion of the substrate material of microelectronic components The unit is per degree Celsius. Calculate the absolute value of the difference between the two. , will the As the thermal mismatch coefficient, the unit is per degree Celsius.

[0055] Measuring the equivalent diameter of the molten core With penetration depth The unit is micrometers. The equivalent diameter is... With the penetration depth product The volume characteristic value of the molten core is expressed in square micrometers.

[0056] The thermal mismatch coefficient With the characteristic value of the molten core volume The parameters are input into a pre-built annealing parameter lookup table. This lookup table is a two-dimensional lookup table structure, with the row index being the thermal mismatch coefficient. The discrete value range, with the column index being the volume characteristic value of the molten core. The discrete value range, each cell stores a set of annealing parameter range data, including: the first holding temperature range. Second insulation temperature range First insulation duration interval Second insulation duration interval and the third insulation duration interval The annealing parameter lookup table is pre-constructed based on experimental data under different material combinations and melt core sizes, and the thermal mismatch coefficient... The step size of the discrete value interval is set to Each degree Celsius, ranging from to per degree Celsius; the characteristic value of the molten core volume The discrete value interval step size is set to 100 square micrometers, ranging from 0 to 10000 square micrometers. The annealing parameter lookup table is used to match the input... and The intervals in question are the first insulation temperature interval, the second insulation temperature interval, the first insulation duration interval, the second insulation duration interval, and the third insulation duration interval.

[0057] The median of each of the following temperature ranges was selected as the first insulation temperature using the centroid method: the first insulation temperature range, the second insulation temperature range, the first insulation duration range, the second insulation duration range, and the third insulation duration range. Second insulation temperature First heat preservation time Second heat preservation time and the third heat preservation time The centroid method is applied to each interval. , take the median The first insulation temperature The value is the median obtained from the first insulation temperature range, in degrees Celsius; the second insulation temperature... The value is the median obtained from the second insulation temperature range, in degrees Celsius; the first insulation duration is... The second insulation duration is the median value obtained from the first insulation duration interval, in minutes; The third insulation duration is the median value obtained from the second insulation duration interval, in minutes; The value is the median obtained from the third insulation duration interval, in minutes.

[0058] Example 5 In practice, the gradient temperature-controlled annealing platform applies low-frequency mechanical vibration to the welding assembly during the staged thermal cycling stress homogenization process to assist in stress release. A piezoelectric ceramic vibration generator is installed below the heating station of the gradient temperature-controlled annealing platform. This piezoelectric ceramic vibration generator is a stacked piezoelectric ceramic actuator, model PST150 / 7 / 100, with a maximum output force of 1000 Newtons and a maximum displacement of 100 micrometers. The piezoelectric ceramic vibration generator is bolted to the metal base plate of the heating station, and the metal base plate is rigidly connected to the heating station, allowing vibration to be transmitted to the welding assembly placed on the heating station.

[0059] When the temperature of the welding assembly rises to the first holding temperature At that time, the piezoelectric ceramic vibration generator is activated. The first insulation temperature... The start time is determined by the lookup table method in Example 4. The temperature controller of the gradient temperature-controlled annealing platform detects when the welding assembly temperature reaches [a certain value]. A trigger signal is issued at a certain time. After startup, the piezoelectric ceramic vibration generator vibrates at a first vibration frequency. and the first vibration amplitude A mechanical vibration wave perpendicular to the welding interface is transmitted to the welding assembly. The first vibration frequency... Set to 50 Hz, the first vibration amplitude The setting is 10 micrometers. The piezoelectric ceramic vibration generator continues to operate until the switching conditions are met.

[0060] When the temperature of the welding assembly drops from the second holding temperature Drop to the first insulation temperature During the process, the vibration frequency of the piezoelectric ceramic vibration generator is switched to a second vibration frequency. The switching trigger condition is that the temperature controller of the gradient temperature-controlled annealing platform detects that the temperature of the welding assembly has decreased from the second holding temperature. It begins to decrease, and the temperature value reaches the first insulation temperature for the first time. At that time, a switching signal is issued. The second vibration frequency... Higher than the first vibration frequency The frequency is set to 150 Hz. The piezoelectric ceramic vibration generator maintains the first vibration amplitude after switching. constant.

[0061] When the temperature of the welding assembly is lower than the first insulation temperature And above the glass transition temperature When the piezoelectric ceramic vibration generator is turned off, the glass transition temperature is [not specified]. The glass transition temperature is taken from the lower of the lead material to be soldered or the substrate material of the microelectronic component, and the specific value is obtained from the material data sheet. The shutdown trigger condition is that the temperature controller of the gradient temperature-controlled annealing platform detects that the temperature of the soldering assembly is lower than the first holding temperature. And higher than the glass transition temperature. When the temperature reaches any value within the specified range, a shutdown signal is issued. After shutdown, the piezoelectric ceramic vibration generator stops outputting vibration.

[0062] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A microresistance welding process for microelectronic components and a post-weld stress relief method, characterized in that, The method includes: We acquire the three-dimensional topographic data of the welding area of ​​microelectronic components and the material property parameters of the leads to be welded, and jointly construct a dynamic resistance prediction model of the welding contact interface. The contact resistance distribution sequence output by the dynamic resistance prediction model is input into the pulse welding parameter adaptive optimization framework for multi-objective welding waveform modulation processing to generate a non-uniform welding energy release strategy that includes pre-pressure time, welding current pulse slope and holding time. The non-uniform welding energy release strategy is executed to perform instantaneous heating and melting treatment on the welding contact interface, and in-situ electromagnetic field-assisted stress control treatment is simultaneously started during the solidification window of the molten core. The in-situ electromagnetic field-assisted stress control treatment dynamically adjusts the electromagnetic field frequency and duty cycle by real-time acquisition of the solid phase change acoustic emission signal of the molten core. The welded components, after stress regulation treatment, are transferred to a gradient temperature-controlled annealing platform for staged thermal cycling stress homogenization treatment to obtain finished microelectronic components with eliminated residual stress.

2. The micro-resistance welding process and post-weld stress relief method for microelectronic components according to claim 1, characterized in that, The jointly constructed dynamic resistance prediction model for the welded contact interface includes: The three-dimensional morphology data of the welding area is statistically processed to generate a true contact area distribution map of the contact interface. The material property parameters of the lead to be soldered are subjected to conductivity and hardness correlation feature extraction processing to obtain the spatial heterogeneity distribution matrix of material resistivity. The actual contact area distribution map is multiplied by the spatial heterogeneity distribution matrix of material resistivity on a grid-by-grid basis to generate the initial contact resistance distribution field. The initial contact resistance distribution field is updated iteratively by calling a pre-trained time-varying resistance evolution network and outputting the contact resistance distribution sequence.

3. The micro-resistance welding process and post-weld stress relief method for microelectronic components according to claim 2, characterized in that, The process of calling a pre-trained time-varying resistance evolution network to perform thermo-electric-mechanical coupled iterative update processing on the initial contact resistance distribution field includes: The initial contact resistance distribution field is input into the Joule heat generation layer of the time-varying resistance evolution network to calculate the instantaneous heat power density at each grid point when the welding current passes through. The instantaneous thermal power density is transferred to the thermal diffusion simulation layer of the time-varying resistance evolution network to update the temperature distribution field at the contact interface, and iterative calculation is performed using the thermal diffusion finite difference update formula. The temperature distribution field is input into the thermal softening effect layer of the time-varying resistance evolution network, and the local contact pressure of each grid point is updated according to the mapping relationship between temperature and material hardness. The updated local contact pressure is fed back to the contact resistance correction layer of the time-varying resistance evolution network, and the contact resistance value of each grid point is recalculated to form the next time step output in the contact resistance distribution sequence.

4. The micro-resistance welding process and post-weld stress relief method for microelectronic components according to claim 3, characterized in that, The resistance time-varying evolution network is a time-series prediction model based on a long short-time memory network. Its input features include the contact resistance distribution field and welding current density field at the current moment.

5. The microresistance welding process for microelectronic components and the post-weld stress relief method according to claim 3, characterized in that, Generate a non-uniform welding energy release strategy that includes preload time, welding current pulse slope, and holding time, including: The contact resistance distribution sequence is processed by extracting the time-series change trend features to identify the critical inflection point when the contact resistance changes from the initial high resistance state to the low resistance state. Using the critical inflection point as the boundary, the optimization objectives of the pulse welding parameter adaptive optimization framework are divided into the preheating stage melt nucleus initiation objective and the main welding stage melt nucleus expansion objective; Under the constraint of the preheating section melt nucleus initiation target, the initial combination of pre-compression time and welding current pulse slope is discretely searched to determine the pre-compression time and welding current pulse slope that make the contact resistance decrease rate reach the preset level. Under the constraint of the target expansion of the molten core in the main welding section, the molten core size at the end of the preheating section is taken as the initial state, and the holding time and welding current peak value are optimized by gradient increase to obtain the holding time and welding current peak value.

6. The micro-resistance welding process for microelectronic components and the post-weld stress relief method according to claim 1, characterized in that, The non-uniform welding energy release strategy is implemented to instantaneously heat and melt the welding interface, and in-situ electromagnetic field-assisted stress control is simultaneously initiated during the solidification window of the molten nucleus, including: According to the pre-pressure time, welding current pulse slope and holding time in the non-uniform welding energy release strategy, the welding current waveform is output sequentially to resistively heat the welding contact interface until a complete molten nucleus is formed. The dynamic resistance signal of the welding circuit is monitored in real time. When the first derivative of the dynamic resistance signal changes from negative to positive, it is determined that the molten nucleus has entered the solidification window period. During the solidification window, the electromagnetic field generator is activated to apply an alternating electromagnetic field to the molten core region according to the initial electromagnetic field frequency and duty cycle. During the application of an alternating electromagnetic field, the acoustic emission signal waveform released during the solidification of the molten nucleus is continuously acquired, and the acoustic emission event rate and average amplitude characteristics are extracted from the acoustic emission signal waveform.

7. The microresistance welding process for microelectronic components and the post-weld stress relief method according to claim 6, characterized in that, The determination of whether the first derivative of the dynamic resistance signal changes from negative to positive is based on the zero-crossing detection of the derivative after smoothing by a sliding window.

8. The microresistance welding process for microelectronic components and the post-weld stress relief method according to claim 6, characterized in that, The in-situ electromagnetic field-assisted stress control process dynamically adjusts the electromagnetic field frequency and duty cycle by real-time acquisition of the solid-state phase transition acoustic emission signal of the molten core, including: After normalizing the acoustic emission event rate and average amplitude characteristics, the data are input into the fuzzy rule inference engine. The output variables of the fuzzy rule inference engine are the electromagnetic field frequency adjustment coefficient and the duty cycle adjustment coefficient. When the acoustic emission event rate exceeds the high event rate threshold and the average amplitude exceeds the high amplitude threshold, the electromagnetic field frequency adjustment coefficient output by the fuzzy rule inference engine is a frequency boosting coefficient, and the duty cycle adjustment coefficient is a duty cycle reduction coefficient. The electromagnetic field frequency is updated according to the product of the electromagnetic field frequency adjustment coefficient and the current electromagnetic field frequency, and the duty cycle is updated according to the product of the duty cycle adjustment coefficient and the current duty cycle. When the acoustic emission event rate is lower than the low event rate threshold, the electromagnetic field frequency adjustment coefficient output by the fuzzy rule inference engine is the frequency maintenance coefficient, and the duty cycle adjustment coefficient is the duty cycle maintenance coefficient.

9. The microresistance welding process for microelectronic components and the post-weld stress relief method according to claim 8, characterized in that, The fuzzy rule reasoning engine uses a triangular membership function as its input membership function and a single-value membership function as its output membership function.

10. The microresistance welding process for microelectronic components and the post-weld stress relief method according to claim 1, characterized in that, The step of transferring the stress-modulated welded assembly to a gradient temperature-controlled annealing platform for staged thermal cycling stress homogenization includes: The welding assembly is placed on the heating station of the gradient temperature-controlled annealing platform and heated from room temperature to the first holding temperature at the first heating rate, and maintained at the first holding temperature for the first holding time to complete the first stage of annealing. The second heating process involves heating from the first holding temperature to the second holding temperature at a second heating rate, maintaining the second holding temperature for a second holding time, wherein the second holding temperature is higher than the first holding temperature, thus completing the second stage of annealing. Cool down from the second holding temperature to the first holding temperature at the first cooling rate, and maintain the third holding time at the first holding temperature to complete the first intermediate cooling and holding process. The temperature is reduced from the first holding temperature to room temperature at a second cooling rate to complete the final cooling process; The second heating rate is less than the first heating rate, and the second cooling rate is less than the first cooling rate.

Images