Ultra-small-caliber ceramic chopper machining method based on femtosecond laser and application of ultra-small-caliber ceramic chopper machining method

Abstract

The invention discloses a femtosecond laser-based ultra-small-caliber ceramic chopper processing method and application thereof, relates to the technical field of laser processing, guarantees the processing basic precision by optimizing a blank preparation and fixing mode, realizes accurate removal of materials through beam shaping and parameter adaptation, and improves the processing precision of the ultra-small-caliber ceramic chopper. The forming precision of a complex three-dimensional structure is guaranteed through six-axis linkage scanning and path optimization, secondary damage is avoided through inert gas protection and synchronous chipping removal, machining consistency and stability are guaranteed through AI self-adaptive closed-loop control, and therefore machining of the composite structure ceramic chopper with the tip diameter smaller than or equal to 50 micrometers is stably achieved. The surface of a processed product is free of a fused recasting layer, a heat affected zone and obvious microscopic cracks, the surface roughness and geometric accuracy are superior to those of the mainstream standard of the industry, the service life is greatly prolonged, the performance requirement of advanced semiconductor packaging is perfectly met, and a brand new technical path is provided for efficient and precise manufacturing of the ultra-small-caliber ceramic chopper.

Description

Technical Field

[0001] This invention relates to the field of laser processing technology, specifically to a method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers and its application. Background Technology

[0002] In the field of semiconductor packaging technology, ceramic wedges are an indispensable tool in semiconductor packaging. Their processing quality directly affects the bonding effect and the reliability of semiconductor devices. With the rapid development of the semiconductor industry, the performance requirements for ceramic wedges are becoming increasingly stringent. They not only need to have ultra-small tip diameters, but also extremely low surface roughness, high geometric accuracy, and long service life to meet the needs of advanced packaging technologies such as 2.5D / 3D and Chiplet. Therefore, developing a method to achieve high-quality, high-precision ceramic wedge processing has become a key problem that urgently needs to be solved in the current semiconductor packaging field.

[0003] Traditional ceramic wedges are primarily processed using ultra-precision grinding. However, due to the hard and brittle nature of ceramic materials, micro-cracks, residual stress, and edge chipping are easily generated during grinding, severely affecting the processing quality and service life of the wedges. Especially when the wedge tip diameter is less than 50μm, the processing difficulty increases dramatically, making it difficult to meet the stringent requirements of advanced semiconductor packaging technology for wedge tip diameters ≤25μm. To overcome the limitations of grinding processes, the industry has attempted to introduce laser processing technology, but existing laser processing methods still have significant shortcomings: the significant thermal effects generated during nanosecond laser processing can lead to thermal damage to the ceramic material; picosecond laser processing suffers from directional heat accumulation, affecting processing accuracy; and while traditional femtosecond laser processing can reduce the thermal impact to some extent, its unreasonable path planning makes it difficult to achieve high-quality forming of complex three-dimensional structures. Therefore, existing technologies cannot meet the comprehensive performance requirements of advanced semiconductor packaging for ceramic wedges in terms of tip diameter, surface roughness, geometric accuracy, and service life. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers and its application. It can fundamentally eliminate thermal and mechanical damage during processing through the cold processing mechanism of femtosecond lasers, ensuring that the intrinsic properties of ceramic materials are not affected. At the same time, it integrates a composite path planning algorithm that combines specific process parameter windows, spiral layering and contour offset, and an AI adaptive closed-loop control strategy that combines coaxial vision and spectral monitoring to build a complete ultra-precision processing system. It can achieve stable processing of ceramic chopping tools with a tip diameter ≤50μm and a composite geometric structure of inner hole, outer conical surface and back cavity, effectively improving the controllability of the processing process and the dimensional consistency of the product.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers, the method comprising the following specific steps: S1: Provide a blank made of zirconium oxide toughened alumina or silicon nitride material that meets specific ratio, density, dimensional tolerance and surface quality requirements, and fix it on a five-axis or higher ultra-precision motion platform by a combination of vacuum adsorption and flexible clamps and calibrate its coaxiality. S2: A femtosecond laser pulse with a specific wavelength, pulse width, single pulse energy and repetition frequency is configured, phase-modulated and shaped by a spatial light modulator, and then focused by a high numerical aperture long working distance objective lens to form a focused spot with specific parameters. S3: Based on the 3D model of the cleaver, a spiral layering and contour offset composite path planning algorithm is adopted to control the motion platform and the laser scanning galvanometer to perform six-axis linkage scanning and etching. When processing key areas, the deceleration scanning mode and the ring compensation path are automatically called and the pulse overlap rate is adjusted. S4: During the processing, argon or nitrogen gas with specific parameters is injected into the processing area, and debris is removed by a synchronous negative pressure suction device with specific parameters and equipped with a collection device with corresponding filtration precision. S5: The high-speed vision module and the paraxial spectrometer, which are integrated in a coaxial manner, collect relevant feature parameters. After fusion, the parameters are input into the deep learning evaluation model. The laser pulse energy or scanning speed is dynamically fine-tuned through the central control system.

[0006] Furthermore, in step S1, the mass ratio of zirconium oxide to alumina in the zirconium oxide toughened alumina green is 3:7 to 4:6, and 5% to 8% by mass of Y2O3-Al2O3 composite sintering aid is added to the silicon nitride green. The green body density is controlled at 55% to 65%, the dimensional tolerance is ≤ ±0.02 mm, and the surface is free of pores with a diameter > 5 μm and cracks with a length > 10 μm.

[0007] Furthermore, in step S1, after fixing, the coaxiality between the billet's central axis and the platform's rotation center is calibrated using a visual positioning system. Specifically, a high-resolution industrial camera and a coaxial white LED light source are fixed to the processing system frame. The camera's optical axis is parallel to the platform's rotation center axis, and the distance between them is preset and fixed. A clear side view image of the billet is acquired through the camera. The complete outline of the billet's cross-section is obtained using grayscale threshold segmentation and edge contour extraction techniques. The pixel coordinates of the billet's central axis are obtained by fitting the outline using the least squares method. Subsequently, the platform is driven to rotate 180° in steps around the rotation axis. The above image acquisition, contour extraction, and center positioning operations are repeated every 30° interval. The offset data of the billet's central axis relative to the camera's optical axis at each rotation angle is recorded. The actual coaxiality deviation between the billet's central axis and the platform's rotation center is calculated based on the offset values ​​acquired multiple times. The X / Y translation axis of the platform is driven by the central control system to make a slight compensation adjustment. During the adjustment process, images are acquired in real time and the deviation value is dynamically verified until the offset of the billet's central axis acquired multiple times within one rotation of the platform is stably controlled within the preset range, thus completing the coaxiality calibration.

[0008] Furthermore, in step S2, a femtosecond laser is used to generate processing pulses. The wavelength of the pulses is selected from 515nm or 1030nm, the pulse width is 100-280fs, the single pulse energy is 5-30μJ, and the repetition frequency is 300kHz-1.5MHz. For processing zirconia-toughened alumina ceramics, a 515nm wavelength is preferred because the 515nm wavelength has a higher absorption rate in the absorption spectrum of zirconia-toughened alumina, which can effectively improve the material removal efficiency and reduce energy loss. For processing silicon nitride ceramics, a 1030nm wavelength is preferred because the laser at this wavelength can penetrate the thin oxide film on the surface of silicon nitride and act directly on the substrate material, avoiding interference from the surface oxide film on energy transfer. At the same time, the energy attenuation of the 1030nm wavelength laser pulse inside silicon nitride is more gradual, which can achieve uniform etching removal. The pulse width is adjusted according to the blank thickness. When the blank thickness is >12mm, 100~180fs is selected, and when the blank thickness is ≤12mm, 180~280fs is selected.

[0009] Furthermore, in step S2, after the pulse is phase-modulated and shaped by a spatial light modulator, it is focused through a long working distance objective lens with a high numerical aperture ≥0.55. The working distance of the objective lens is 10~20mm, the diameter of the focused spot is 1~5μm, and the focusing depth is ±2μm.

[0010] Furthermore, in step S3, the specific steps for controlling the motion platform and the laser scanning galvanometer to perform six-axis linkage scanning and etching based on the three-dimensional model of the cleaver using a spiral layering and contour offset composite path planning algorithm are as follows: The three-dimensional model of the cleaver is imported into the path planning system; the model is sliced ​​layer by layer according to a preset layer thickness; after each slice, a continuous spiral scanning path in a clockwise direction is generated with the center of the blank as the spiral starting point; the layer thickness of the spiral layering is set to 0.5~2μm; the layer thickness is 0.5μm in the finishing stage and 1.0μm in the roughing stage. ~2.0μm, and the contour offset step is determined based on the preset ratio of the focused spot diameter, so that the spiral path and the contour offset path alternately connect along the edge of the slice contour to form a non-overlapping and non-omission scanning coverage area; the X / Y / Z linear axes of the motion platform realize the macroscopic position positioning and layer feeding of the blank according to the slicing sequence, and the A / B / C rotary axes adjust the blank posture in real time according to the spatial angle of each layer to be processed. The laser scanning galvanometer synchronously responds to the path signal and performs precise correction of the microscopic scanning trajectory, realizing the six-axis linkage and coordination of the motion platform and the galvanometer.

[0011] Furthermore, in step S3, when scanning to the transition area at the junction of the inner hole and the outer conical surface and the tip arc area, the path algorithm automatically reduces the scanning speed by 30% to 50% compared to the base value, adjusts the laser pulse overlap rate to 30% to 45%, and generates an annular compensation path that adapts to the curvature of the region, so that the focused spot can be smoothly and continuously etched along the preset composite path.

[0012] Furthermore, in step S4, during the processing, an inert gas with a purity ≥99.999% is sprayed onto the processing point at a pressure of 0.1-0.3 MPa, while a synchronous negative pressure suction device with a pressure of -0.05 to -0.1 MPa is used to remove debris; wherein the inert gas is argon or nitrogen, the spray angle is 30°~45° with the processing surface, the distance between the negative pressure suction port and the processing point is 5~8 mm, and the synchronous negative pressure suction device is equipped with a debris collection device with a filtration accuracy ≤1 μm.

[0013] Furthermore, in step S5, the processed morphology image acquired by the high-speed vision module is preprocessed to extract surface roughness, dimensional deviation, and contour integrity feature parameters. Simultaneously, baseline correction and peak extraction are performed on the plasma emission spectrum acquired by the paraxial spectrometer to obtain spectral feature parameters of specific element emission peak intensity and peak shift. The two types of feature parameters are normalized, spliced, and fused according to preset weights to form a comprehensive feature vector characterizing the processing state. This feature vector is input into a deep learning evaluation model pre-trained on a large number of processing samples and embedded in the system. The model performs real-time calculations and analysis on the feature vector, comparing it with pre-defined parameters. The system compares the output with the characteristic thresholds of the normal processing state. If the output exceeds the threshold range, it is judged as an abnormal signal and immediately fed back to the central control system. The central control system automatically matches and adjusts the strategy according to the characteristic deviation type corresponding to the abnormal signal. If the abnormality is due to morphological abnormalities such as excessive surface roughness or contour defects, the laser scanning speed is finely adjusted according to the preset amplitude. If the abnormality is due to energy abnormalities such as fluctuations in spectral peak intensity or abnormal element emission characteristics, the laser pulse energy is finely adjusted according to the preset amplitude. During the fine adjustment process, the adjusted morphological image and spectral data are collected in real time and continuously input into the evaluation model to verify the adjustment effect until the characteristic parameters return to the normal range.

[0014] An application of an ultra-small diameter ceramic chopper based on femtosecond laser, wherein the ceramic chopper is applied in the field of advanced semiconductor packaging, specifically including 2.5D / 3D packaging and chiplet integrated packaging processes for products such as integrated circuits, microelectromechanical systems, and optoelectronic devices.

[0015] Compared with existing technologies, this method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers and its application have the following advantages: This invention ensures basic machining accuracy by optimizing the blank preparation and fixing methods, achieves precise material removal through beam shaping and parameter adaptation, ensures the forming accuracy of complex three-dimensional structures through six-axis linkage scanning and path optimization, avoids secondary damage through inert gas protection and simultaneous debris removal, and ensures machining consistency and stability through AI adaptive closed-loop control. This allows for the stable machining of composite ceramic chopping tools with a tip diameter ≤50μm. The machined product surface is free of molten recast layer, heat-affected zone, and obvious microcracks. Surface roughness and geometric accuracy are superior to mainstream industry standards, and service life is significantly extended. It perfectly meets the performance requirements of advanced semiconductor packaging, providing a new technical path for the efficient and precise manufacturing of ultra-small diameter ceramic chopping tools.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a flowchart of a method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers; Figure 2 This is a schematic diagram of a method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers; Figure 3 This is a schematic diagram of spiral layered scanning path planning for a femtosecond laser-based method for machining ultra-small diameter ceramic chopping tools. Figure 4 This is a schematic diagram of the scanning coverage area of ​​a femtosecond laser-based method for processing ultra-small diameter ceramic chopping tools. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] Example 1 S1: Provide a blank made of zirconium oxide toughened alumina or silicon nitride material that meets specific ratio, density, dimensional tolerance and surface quality requirements, and fix it on a five-axis or higher ultra-precision motion platform by a combination of vacuum adsorption and flexible clamps and calibrate its coaxiality. like Figure 1 As shown, a zirconia-toughened alumina green billet with a zirconia to alumina mass ratio of 3:7 was selected. The green billet size was designed to be φ0.6mm×15mm. The density of the green billet was controlled to 60% using a density testing method. The dimensional tolerance of the green billet was measured to be ±0.015mm using a high-precision dimensional measuring instrument. The surface of the green billet was observed using a metallographic microscope to confirm that there were no pores with a diameter greater than 5μm or cracks with a length greater than 10μm, meeting the preset surface quality requirements. A combination of vacuum adsorption and flexible clamps was used for fixing. The clamps were made of polytetrafluoroethylene, which has both flexibility and wear resistance to avoid damage to the green billet. After the green billet was placed in the positioning groove of the clamp, the vacuum adsorption system was turned on and the adsorption pressure was set to -0.04MPa. The adsorption status was monitored in real time by a pressure sensor to ensure that the green billet was stable and free from deformation under stress.

[0021] A high-resolution industrial camera (1024×768) and a coaxial white LED light source are fixed to the processing system frame. The camera position is adjusted to ensure the optical axis is parallel to the platform's rotation center axis, with a preset distance of 50mm between them. The position is locked using frame positioning pins to prevent displacement during processing. A clear side view image of the billet is acquired using the camera. Gray-scale thresholding segmentation technology is used to separate the image background from the billet area. Then, the complete cross-sectional contour of the billet is extracted using the Canny edge detection algorithm. The contour curve is fitted using the least squares method, and the fitting formula is: ,in, , The pixel coordinates of any point on the cross-sectional contour of the blank; , , The coefficients for the quadratic fitting of the contour curve are obtained by minimizing the sum of squared distances from all points on the contour to the fitted curve. After fitting using this formula, the pixel coordinates of the billet's central axis can be further calculated. .

[0022] The rotary axis driving the six-axis ultra-precision motion platform rotates 180° in steps around the central axis, pausing every 30°. After each pause, the above image acquisition, contour extraction, and center positioning operations are repeated, recording the offset data of the billet's central axis relative to the camera's optical axis at a total of 6 rotation angles. 6) The actual coaxiality deviation is calculated using the arithmetic mean method based on the 6 sets of offset data. The deviation calculation formula is as follows: ,in, This represents the actual coaxiality deviation between the billet's central axis and the platform's rotation center. For the first The offset of the billet's central axis in the X direction relative to the camera's optical axis during the second measurement; For the first The offset of the billet's central axis in the Y direction relative to the camera's optical axis during the second measurement; The number of measurements (values ​​1 to 6, corresponding to 6 rotation angles) was used to calculate the deviation, which was found to be 0.04 μm. Figure 2 As shown, the central control system drives the platform's X / Y translation axis to make micro-compensation adjustments based on the deviation value (single adjustment step size 0.01μm). During the adjustment process, images are acquired in real time and the deviation value is dynamically verified. After three fine adjustments, the offset of the billet's central axis acquired multiple times within one rotation of the platform is stably controlled within ±0.02μm, thus completing the coaxiality calibration.

[0023] S2: A femtosecond laser pulse with a specific wavelength, pulse width, single pulse energy and repetition frequency is configured, phase-modulated and shaped by a spatial light modulator, and then focused by a high numerical aperture long working distance objective lens to form a focused spot with specific parameters. A femtosecond laser is used to generate processing pulses. Based on the material properties of zirconia-toughened alumina, a laser wavelength of 515nm is selected because it has a higher absorption rate in the absorption spectrum of zirconia-toughened alumina, effectively improving material removal efficiency and reducing energy loss. Since the green body thickness is 15mm, the pulse width is set to 230fs according to the technical solution. The basic energy per pulse is set to 18μJ. Considering the precision requirements of the tip processing, the energy in the tip processing area is increased by 5.6% from the basic energy using an energy adjustment module. The energy increase formula is as follows: ,in, The laser single-pulse energy for the advanced processing area; This is the basic energy of a single laser pulse; The coefficient for improving the energy efficiency of advanced processing is given by substituting the data. =19μJ; set the repetition frequency to 800kHz to ensure it matches the subsequent scan speed and meets the pulse overlap rate requirements.

[0024] The laser pulse is phase-modulated and shaped using a spatial light modulator, and the energy distribution of the spot is optimized using a phase correction algorithm to obtain a flat-top spot with an energy uniformity of 93%. The shaped pulse is then fed into a long working distance objective lens with a numerical aperture of 0.6, and the laser pulse is focused into a focused spot with a diameter of 3μm and a focusing depth of ±2μm using an objective lens focusing system. The spot parameters are monitored in real time using a spot quality analyzer to ensure that they meet the processing requirements.

[0025] S3: Based on the 3D model of the cleaver, a spiral layering and contour offset composite path planning algorithm is adopted to control the motion platform and the laser scanning galvanometer to perform six-axis linkage scanning and etching. When processing key areas, the deceleration scanning mode and the ring compensation path are automatically called and the pulse overlap rate is adjusted. like Figure 3 As shown, the 3D model of the cleaver is imported into the path planning system. A layered slicing algorithm is used to slice the material according to a preset layer thickness. The layer thickness is set to 1.0 μm in the roughing stage and reduced to 0.5 μm in the finishing stage. The slicing direction is parallel to the axis of the blank to avoid cross-sectional tilting. After each slicing layer, a continuous clockwise spiral scanning path is generated using the center of the blank as the starting point of the spiral. The spiral path parameters satisfy the following: , ,in, , Let be the coordinates of any point on the spiral path; The scanning radius increases linearly with the scanning progress, with an initial value of 0 and a maximum value of the maximum radius of the slice outline. The scanning angle linearly increases from 0° to integer multiples of 360° as the scanning progresses. Simultaneously, the contour offset step size is set to 1.2 μm based on the focused spot diameter. The offset step size formula is: ,in, The offset step size is the profile offset step size. The diameter of the laser focusing spot; The ratio of the offset step size to the spot diameter is used as a factor. A path connection algorithm alternates between the spiral path and the contour offset path along the edge of the slice contour. Figure 4 As shown, a non-overlapping and non-omitted scanning coverage area is formed to ensure the integrity of the processing surface.

[0026] The X / Y / Z linear axes of the motion platform perform macroscopic positioning and layered feeding of the blank according to the slicing sequence. The A / B / C rotary axes detect the spatial angle of each surface to be processed in real time through angle sensors, and dynamically adjust the blank posture according to the angle data to ensure that the area to be processed is always perpendicular to the laser focusing spot. The laser scanning galvanometer responds synchronously to the path signal through a high-speed driver to accurately correct the microscopic scanning trajectory (correction accuracy ±0.05μm), realizing six-axis linkage and coordination between the motion platform and the galvanometer.

[0027] When scanning reaches the transition area between the inner hole and the outer conical surface, as well as the tip arc area, the path planning system automatically triggers a deceleration scanning mode. The scanning speed is reduced by 40% from the base value of 500 mm / s via the speed adjustment module. The speed adjustment formula is as follows: ,in, Adjusted scanning speed for critical areas; The baseline scanning speed for non-critical areas; Let be the deceleration coefficient for the scanning speed. Substituting the data, we get... =300mm / s; simultaneously adjust the laser pulse overlap rate, the formula for calculating the overlap rate is: ,in, The laser pulse overlap ratio; The laser pulse repetition frequency; The scan line spacing is equal to the contour offset step size. Substitute the current scanning speed into... =800kHz =1.2μm, =300mm / s, calculated as follows =38%; and a ring-shaped compensation path with a compensation radius of 1.0 μm is generated by the curvature fitting algorithm. This ring-shaped path enables the focused spot to be continuously etched along the curvature of the region smoothly, avoiding processing defects in key areas.

[0028] S4: During the processing, argon or nitrogen gas with specific parameters is injected into the processing area, and debris is removed by a synchronous negative pressure suction device with specific parameters and equipped with a collection device with corresponding filtration precision. Argon gas with a purity of 99.999% is selected as the protective gas. The injection pressure is set to 0.2MPa through a gas pressure reducing valve. The nozzle angle is adjusted so that the gas injection direction is 35° with the processing surface. The distance between the injection port and the processing point is controlled at 8mm to form a local sealed protective atmosphere. The oxygen concentration in the processing area is monitored in real time by an oxygen content sensor to ensure that the concentration is below 0.1%. The synchronous negative pressure suction device is turned on and the suction pressure is set to -0.08MPa. The suction port is fixed 6mm below the processing point. The device is equipped with a ceramic filter element with a filtration accuracy of 1μm. The filter element intercepts the ceramic debris generated during processing to prevent the debris from adhering to the surface of the blank or entering the equipment and causing damage. The filter element clogging is monitored regularly by a differential pressure sensor to ensure stable suction efficiency.

[0029] S5: The coaxially integrated high-speed vision module and the paraxial spectrometer collect processing-related feature parameters, which are then fused and input into the deep learning evaluation model. The laser pulse energy or scanning speed is dynamically fine-tuned through the central control system. The high-speed vision module acquires a processing topography image for each layer scanned. After grayscale correction and noise reduction, the image is processed by image analysis algorithms to extract feature parameters such as surface roughness, dimensional deviation, and contour integrity. The paraxial spectrometer acquires the plasma emission spectrum generated during the processing in real time. The baseline correction algorithm eliminates background interference, and the peak detection algorithm extracts spectral feature parameters such as emission peak intensity and peak offset of specific elements (Al, Zr).

[0030] The morphological and spectral feature parameters are normalized according to preset weights (morphological parameter weight 0.6, spectral parameter weight 0.4). A feature concatenation algorithm is used to form a comprehensive feature vector representing the processing state. This vector is input into a deep learning evaluation model that has been trained on 500 sets of samples with different processing states and is fixed in the system. The deep learning evaluation model adopts a convolutional neural network (CNN) model. The training data comes from 800 sets of good and defective samples processed in history, covering processing scenarios of blanks with different materials and thicknesses. The sample data are all morphological-spectral fusion feature data collected synchronously by a high-speed vision module and a rangefinder spectrometer. After 100 rounds of iterative training, the model has a recognition accuracy of 98.5%. The model performs real-time calculation and analysis on the feature vector and compares it with the preset normal processing state feature threshold.

[0031] During the processing, the model detected that the surface roughness parameter exceeded the normal threshold, which was identified as a morphological anomaly and immediately fed back to the central control system. The control system automatically matched and adjusted the strategy, prioritizing the fine adjustment of the laser scanning speed, reducing the speed from 300mm / s to 280mm / s. During the fine adjustment, the adjusted morphological image and spectral data were collected in real time and continuously input to evaluate the model to verify the adjustment effect. After three layers of scanning, the surface roughness parameter returned to the normal range, ensuring that the processing quality was stable and controllable.

[0032] Tests were conducted under the same working conditions, comparing the zirconia-toughened alumina ceramic wedge processed in this embodiment with a wedge of the same specification processed by traditional ultra-precision grinding. The results are as follows: Number of consecutive bonding cycles: The cleaver processed in this embodiment has a continuous and stable bonding cycle of 362,000 cycles, while the traditional grinding cleaver only has 231,000 cycles, resulting in a 56.5% increase in service life; Surface roughness: In this embodiment, the surface roughness of the critical area of ​​the cleaving tool is Ra=0.045μm, while that of a traditional grinding cleaving tool is Ra=0.135μm, resulting in a 66.7% improvement in surface smoothness; Dimensional accuracy: The diameter deviation of the chopping tool tip in this embodiment is ±0.8μm, while that of a traditional grinding chopping tool is ±3.2μm, representing a 75% improvement in dimensional accuracy; Metal wire damage rate: In this embodiment, the metal wire damage rate of the cleaving tool during the bonding process is only 0.08%, while that of the traditional grinding cleaving tool is 0.52%, representing a reduction in damage rate of 84.6%. Batch processing consistency: For the cleavers of this embodiment with a batch processing of 100 pieces, the key parameter deviation is ≤0.03μm, while that of traditional grinding cleavers is ≤0.12μm, resulting in a 75% improvement in consistency.

[0033] The above comparative data fully demonstrate that the processing method of the present invention effectively solves the problems of thermal damage, low dimensional accuracy, and short service life of traditional processes, significantly improves the overall performance of ceramic cleavers, and fully meets the high precision and high stability requirements of advanced semiconductor packaging for micro-nano bonding tools.

[0034] Example 2 A silicon nitride ceramic green body was prepared, in which 6% Y2O3-Al2O3 composite sintering aid was added. The green body size was φ0.5mm×12mm, and the density was controlled at 58%. After testing, the green body size tolerance was ±0.01mm, and there were no pores with a diameter greater than 5μm or cracks with a length greater than 10μm on the surface.

[0035] The green blank was fixed on a six-axis ultra-precision motion platform by a combination of vacuum adsorption and flexible clamps. The adsorption pressure was set to -0.035MPa. After fixing, the same coaxiality calibration procedure as in Example 1 was performed to finally stabilize the offset of the green blank's central axis within ±0.02μm.

[0036] A femtosecond laser pulse with a wavelength of 1030 nm was selected. Since the green body thickness was 12 mm, the pulse width was set to 200 fs, the single pulse energy was 12 μJ, and the repetition frequency was set to 600 kHz. The pulse was phase-modulated and shaped by a spatial light modulator to obtain a flat-top spot with an energy uniformity of 92%. The spot was then focused by a long working distance objective lens with a numerical aperture of 0.55. The working distance of the objective lens was 12 mm, the focused spot diameter was 2.5 μm, and the focusing depth was ±2 μm.

[0037] After importing the 3D model of the cleaver into the system, roughing and slicing are performed with a layer thickness of 0.8μm. In the finishing stage, the layer thickness remains at 0.5μm. A clockwise spiral scanning path is generated starting from the center of the blank, and the contour offset step is set to 1.0μm (i.e., 40% of the spot diameter). The motion platform and the laser scanning galvanometer work together to achieve six-axis linkage. When scanning to the transition area between the inner hole and the outer conical surface and the tip arc area, the scanning speed is reduced by 40% from the basic value of 450mm / s to 270mm / s, the laser pulse overlap rate is adjusted to 35%, and a ring compensation path with a compensation radius of 0.8μm is generated.

[0038] Nitrogen gas with a purity of 99.999% was selected as the protective gas, with an injection pressure of 0.18 MPa, an injection angle of 30° to the processing surface, and an injection nozzle distance of 7 mm from the processing point. The suction pressure of the synchronous negative pressure suction device was set to -0.07 MPa, the suction nozzle distance was 5 mm from the processing point, and it was equipped with a debris collection device with a filtration accuracy of 1 μm.

[0039] During online monitoring and dynamic fine-tuning, the high-speed vision module acquires morphological images at a frame rate of 1000fps, and the paraxial spectrometer simultaneously acquires spectral data. After feature fusion, the data is input into the evaluation model. No obvious abnormalities occurred during the processing, except for a slight fluctuation in the intensity of the spectral peaks. The central control system fine-tuned the single-pulse energy to 12.8μJ according to the adjustment strategy. Subsequent continuous monitoring showed that the characteristic parameters remained stable within the normal range, ensuring the consistency and reliability of the processing.

[0040] Tests were conducted under the same working conditions, comparing the silicon nitride ceramic wedge processed in this embodiment with a wedge of the same specification processed by traditional ultra-precision grinding. The results are as follows: Number of consecutive bonding cycles: The cleaver processed in this embodiment has a continuous and stable bonding cycle of 385,000 cycles, while the traditional grinding cleaver only has 237,000 cycles, resulting in a 62.4% increase in service life; Surface roughness: In this embodiment, the surface roughness of the critical area of ​​the cleaving tool is Ra=0.042μm, while that of a traditional grinding cleaving tool is Ra=0.145μm, resulting in a 71.1% improvement in surface smoothness; Dimensional accuracy: The diameter deviation of the chopping tool tip in this embodiment is ±0.7μm, while that of a traditional grinding chopping tool is ±3.2μm, representing a 78.1% improvement in dimensional accuracy; Metal wire damage rate: In this embodiment, the metal wire damage rate of the cleaving tool during the bonding process is only 0.07%, while that of the traditional grinding cleaving tool is 0.52%, representing a reduction in damage rate of 86.5%. Batch processing consistency: For the cleavers of this embodiment processed in batches of 100 pieces, the key parameter deviation is ≤0.025μm, while that of traditional grinding cleavers is ≤0.12μm, resulting in a consistency improvement of 79.2%.

[0041] The comparative data above fully demonstrate that the processing method of the present invention effectively solves the problems of heat-affected zone, structural delamination, and low dimensional accuracy that exist in traditional processes for processing silicon nitride ceramics. It significantly improves the overall performance of ceramic wedges and fully meets the high precision, high stability, and long lifespan requirements of semiconductor chiplet integrated packaging and 3D packaging for micro-nano bonding tools. The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for machining ultra-small diameter ceramic chopping tools based on femtosecond lasers, characterized in that, The method includes the following specific steps: S1: Provide a blank made of zirconium oxide toughened alumina or silicon nitride material that meets specific ratio, density, dimensional tolerance and surface quality requirements, and fix it on a five-axis or higher ultra-precision motion platform by a combination of vacuum adsorption and flexible clamps and calibrate its coaxiality. S2: A femtosecond laser pulse with a specific wavelength, pulse width, single pulse energy and repetition frequency is configured, phase-modulated and shaped by a spatial light modulator, and then focused by a high numerical aperture long working distance objective lens to form a focused spot with specific parameters. S3: Based on the 3D model of the cleaver, a spiral layering and contour offset composite path planning algorithm is adopted to control the motion platform and the laser scanning galvanometer to perform six-axis linkage scanning and etching. When processing key areas, the deceleration scanning mode and the ring compensation path are automatically called and the pulse overlap rate is adjusted. S4: During the processing, argon or nitrogen gas with specific parameters is injected into the processing area, and debris is removed by a synchronous negative pressure suction device with specific parameters and equipped with a collection device with corresponding filtration precision. S5: The high-speed vision module and the paraxial spectrometer, which are integrated in a coaxial manner, collect relevant feature parameters. After fusion, the parameters are input into the deep learning evaluation model. The laser pulse energy or scanning speed is dynamically fine-tuned through the central control system.

2. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S1, the mass ratio of zirconium oxide to alumina in the zirconium oxide toughened alumina green is 3:7~4:

6. 5%~8% Y2O3-Al2O3 composite sintering aid is added to the silicon nitride green. The green body density is controlled at 55%~65%, the dimensional tolerance is ≤±0.02mm, and the surface is free of pores with a diameter >5μm and cracks with a length >10μm.

3. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S1, after fixing, the coaxiality between the billet's central axis and the platform's rotation center is calibrated using a visual positioning system. Specifically, a high-resolution industrial camera and a coaxial white LED light source are fixed to the processing system frame. The camera's optical axis is parallel to the platform's rotation center axis, and the distance between them is preset and fixed. A clear side view image of the billet is acquired through the camera. The complete outline of the billet's cross-section is obtained using grayscale threshold segmentation and edge contour extraction techniques. The pixel coordinates of the billet's central axis are obtained by fitting the outline using the least squares method. Subsequently, the platform is driven to rotate 180° in steps around the rotation axis. The above image acquisition, contour extraction, and center positioning operations are repeated every 30° interval. The offset data of the billet's central axis relative to the camera's optical axis at each rotation angle is recorded. The actual coaxiality deviation between the billet's central axis and the platform's rotation center is calculated based on the offset values ​​acquired multiple times. The X / Y translation axis of the platform is driven by the central control system to make a slight compensation adjustment. During the adjustment process, images are acquired in real time, and the deviation value is dynamically verified until the offset of the billet's central axis acquired multiple times within one rotation of the platform is stably controlled within the preset range, thus completing the coaxiality calibration.

4. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S2, a femtosecond laser is used to generate processing pulses. The wavelength of the pulse is selected from 515nm or 1030nm, the pulse width is 100-280fs, the single pulse energy is 5-30μJ, and the repetition frequency is 300kHz-1.5MHz. For processing zirconia toughened alumina ceramics, a wavelength of 515nm is preferred, and for processing silicon nitride ceramics, a wavelength of 1030nm is preferred. The pulse width is adjusted according to the thickness of the blank. When the blank thickness is >12mm, 100~180fs is selected, and when the blank thickness is ≤12mm, 180~280fs is selected.

5. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 4, characterized in that, In step S2, the pulse is phase-modulated and shaped by a spatial light modulator, and then focused by a long working distance objective lens with a high numerical aperture ≥0.

55. The working distance of the objective lens is 10~20mm, the diameter of the focused spot is 1~5μm, and the focusing depth is ±2μm.

6. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S3, the specific steps of controlling the motion platform and the laser scanning galvanometer to perform six-axis linkage scanning and etching based on the three-dimensional model of the cleaver using a spiral layering and contour offset composite path planning algorithm are as follows: the three-dimensional model of the cleaver is imported into the path planning system, and the model is sliced ​​layer by layer according to the preset layer thickness. After each slice, a continuous spiral scanning path in the clockwise direction is generated with the center of the blank as the spiral starting point. The layer thickness of the spiral layering is set to 0.5~2μm, the layer thickness in the finishing stage is 0.5μm, and the layer thickness in the roughing stage is 1.0~2μm. 0μm, and at the same time, based on the preset ratio of the focused spot diameter, the contour offset step is determined so that the spiral path and the contour offset path alternately connect along the edge of the slice contour to form a non-overlapping and non-omission scanning coverage area; the X / Y / Z linear axes of the motion platform realize the macroscopic position positioning and layer feeding of the blank according to the slicing sequence, and the A / B / C rotary axes adjust the blank posture in real time according to the spatial angle of each layer to be processed. The laser scanning galvanometer synchronously responds to the path signal to make precise correction of the microscopic scanning trajectory, realizing the six-axis linkage and coordination of the motion platform and the galvanometer.

7. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S3, when scanning to the transition area between the inner hole and the outer conical surface and the tip arc area, the path algorithm automatically reduces the scanning speed by 30% to 50% compared to the base value, adjusts the laser pulse overlap rate to 30% to 45%, and generates an annular compensation path that adapts to the curvature of the region, so that the focused spot can be smoothly and continuously etched along the preset composite path.

8. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S4, during the processing, inert gas with a purity ≥99.999% is sprayed onto the processing point at a pressure of 0.1-0.3 MPa, while a synchronous negative pressure suction device with a pressure of -0.05 to -0.1 MPa is used to remove debris. The inert gas is argon or nitrogen, the injection angle is 30°~45° with the processing surface, the negative pressure suction port is 5~8mm away from the processing point, and the synchronous negative pressure suction device is equipped with a debris collection device with a filtration accuracy of ≤1μm.

9. The method for processing ultra-small diameter ceramic chopping tools based on femtosecond lasers according to claim 1, characterized in that, In step S5, the processed morphology image acquired by the high-speed vision module is preprocessed to extract surface roughness, dimensional deviation, and contour integrity feature parameters. Simultaneously, baseline correction and peak extraction are performed on the plasma emission spectrum acquired by the off-axis spectrometer to obtain spectral feature parameters of specific element emission peak intensity and peak offset. The two types of feature parameters are normalized, spliced, and fused according to preset weights to form a comprehensive feature vector representing the processing state. This feature vector is input into a deep learning evaluation model pre-trained on a large number of processing samples and embedded in the system. The model performs real-time computation and analysis on the feature vector, comparing it with a preset normal value. The processing status characteristic threshold is compared. If the output result exceeds the threshold range, it is judged as an abnormal signal and immediately fed back to the central control system. The central control system automatically matches and adjusts the strategy according to the characteristic deviation type corresponding to the abnormal signal. If the abnormality is due to morphological abnormality such as excessive surface roughness or contour defects, the laser scanning speed is finely adjusted according to the preset amplitude. If the abnormality is due to energy abnormality such as fluctuation of spectral peak intensity or abnormal element emission characteristics, the laser pulse energy is finely adjusted according to the preset amplitude. During the fine adjustment process, the adjusted morphological image and spectral data are collected in real time and continuously input into the evaluation model to verify the adjustment effect until the characteristic parameters return to the normal range.

10. An application of an ultra-small diameter ceramic chopping tool based on femtosecond laser, characterized in that, The ceramic chopping blade is used in the field of advanced semiconductor packaging, specifically including 2.5D / 3D packaging and chiplet integrated packaging processes for products such as integrated circuits, microelectromechanical systems, and optoelectronic devices.

Images