Polishing method for wafer-level plane of brittle-hard material based on femtosecond laser

Abstract

The present application relates to the application in the field of laser precision machining based on wafer level brittle hard material plane polishing method of femtosecond laser, comprising: shaping femtosecond laser into flat-top line spot with uniform energy to scan workpiece surface;In the process of machining, the OCT module integrated coaxially is used to obtain the interference signal of workpiece surface in real time;Real-time analysis of OCT axial scanning signal spectrum feature and / or attenuation slope is used as the diagnostic basis of subsurface damage risk;According to the diagnostic result, the laser energy density is reduced in real time and / or the energy distribution of the spot is dynamically modulated through the spatial light modulator to inhibit the subsurface damage.The present application realizes the real-time, in-situ diagnosis and active inhibition of brittle hard material subsurface damage in femtosecond laser machining, combines flat-top line spot and dynamic light field regulation, solves the problem of roughness deterioration in fine grinding stage, and can realize subsurface nanometer level surface finish, non-damage and high efficiency planarization machining.

Description

Technical Field

[0001] This invention relates to a method for polishing wafer-level brittle and hard material planes using femtosecond lasers, applicable to the field of laser precision machining. Background Technology

[0002] With the rapid development of third-generation semiconductor technology, the demand for wafers made of brittle and hard materials such as silicon carbide and gallium nitride is increasing. After growth or early processing, these materials often have microscopic defects, damaged layers or undulations on their surfaces, requiring ultra-precision planarization for subsequent photolithography and other processes.

[0003] Currently, the mainstream planarization technologies include chemical mechanical polishing (CMP) and mechanical grinding. Although CMP has high precision, it is inefficient, has high material costs, and is not very effective for some superhard materials; mechanical grinding, on the other hand, is prone to introducing mechanical stress and microcracks, leading to a decrease in wafer yield.

[0004] In recent years, femtosecond lasers, with their ultrashort pulse characteristics, have enabled near-thermal-effect-free cold processing, bringing hope for the non-destructive planarization of brittle and hard materials. To improve efficiency, shaping the beam into a line spot for scanning has become a development direction. However, existing technologies still face two key bottlenecks: First, in the fine grinding stage, uneven energy distribution of the laser spot (such as a Gaussian spot) easily leads to alternating over-ablation and under-ablation of the material, causing a reverse increase in surface roughness instead of a decrease, making it difficult to achieve sub-nanometer-level smoothness; Second, and more fundamentally, existing online, non-contact process monitoring methods have sensing blind spots. For example, even optical coherence tomography (OCT) modules integrated with the laser coaxially are usually only used in existing technologies to measure surface height or reconstruct macroscopic morphology. These methods are difficult to perceive, in real time and non-destructively, subsurface damage states such as amorphization, lattice distortion, or microcrack initiation caused by the interaction between the femtosecond laser and the material below the surface during processing. This invisible damage poses a fatal threat to the long-term reliability of devices, making current high-precision laser polishing essentially an "unknown" state of critical risk, which severely restricts the final application of this technology in high-reliability semiconductor manufacturing. Summary of the Invention

[0005] In view of the above-mentioned prior art, the technical problem to be solved by the present invention is: how to overcome the fundamental defect of existing femtosecond laser planarization technology that makes it difficult to perceive and actively control subsurface damage of materials in real time, and to provide a grinding method and system that can diagnose and actively suppress subsurface damage in real time while processing efficiently, thereby achieving truly non-destructive and ultra-precision planarization.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a method for polishing wafer-level brittle and hard material planes using femtosecond lasers, comprising the following steps:

[0007] S1: Generates femtosecond laser pulses and shapes them into flat-top line spots with uniform energy distribution, focusing them on the surface of the workpiece to be processed;

[0008] S2: Control the flat-top line spot to perform two-dimensional scanning relative to the workpiece surface to remove material;

[0009] S3: During the grinding process, the optical coherence tomography (OCT) module, which is coaxially integrated with the femtosecond laser optical path, is used to acquire the interference signal on the workpiece surface in real time.

[0010] S4: Analyze the axial scanning signal of the OCT module in real time, and extract its spectral characteristics and / or signal attenuation slope as a basis for diagnosing subsurface damage risk;

[0011] S5: Based on the diagnostic criteria, the pulse energy density of the femtosecond laser is reduced in real time and / or the wavefront phase of the beam is modulated by a spatial light modulator, thereby dynamically changing the local energy density distribution of the flat-top line spot in the scanning direction to suppress subsurface damage.

[0012] As a further improvement of the present invention, step S4 specifically includes:

[0013] S4.1: Pre-stored health signal fingerprint of workpiece material. The health signal fingerprint is used to characterize the standard signal features of the corresponding material in a non-damaged state, including reference spectral features and reference attenuation slope.

[0014] S4.2: Compare the spectral characteristics and / or attenuation slope of the real-time acquired axial scan signal with the health signal fingerprint;

[0015] S4.3: When the spectral characteristics shift by more than the first preset threshold and / or the attenuation slope exceeds the second preset threshold, it is determined that there is a risk of subsurface damage.

[0016] As a further improvement of the present invention, the real-time adjustment of the femtosecond laser processing parameters in step S5 includes:

[0017] After determining that there is a risk of damage, the pulse energy density of the femtosecond laser is instantaneously reduced before the next or subsequent laser pulse; and / or, the beam phase is modulated by a spatial light modulator, thereby dynamically modulating the local energy distribution of the flat-top line spot in the scanning direction and reducing the energy density of the corresponding risk area.

[0018] The spatial light modulator modulates the local energy density of the flat-top line spot as it scans through the risk area, based on the location and severity of the risk area, and attenuates the energy density relative to the current processing setting, corresponding to the damage risk level.

[0019] As a further improvement of the present invention, in step S3, the OCT module is also used to reconstruct the three-dimensional morphology of the surface; the polishing method further includes step S6:

[0020] Based on three-dimensional topographic information, the local energy density distribution of the flat-top line spot is dynamically adjusted by a spatial light modulator to perform closed-loop modification of the surface topography.

[0021] As a further improvement of the present invention, before step S6, a spectral confocal displacement sensor coaxially integrated with the femtosecond laser optical path is used to measure the absolute height of multiple specific reference points on the OCT scanning path, and the measurement data is used as a reference to correct the systematic error of the overall field shape data acquired by the OCT module.

[0022] As a further improvement of the present invention, for wafer-level brittle and hard materials with heterojunction structures, the polishing method further includes step S0 performed before step S1:

[0023] The location of the heterojunction interface is identified using the tomographic imaging capability of the OCT module;

[0024] In step S2, different processing parameters are applied to different material layers according to the interface position, and the processing endpoint is determined based on the real-time OCT axial signal characteristics; wherein, when the real-time monitored removal depth approaches a preset critical distance of the interface, the processing parameters are automatically switched to conservative.

[0025] As a further improvement of the present invention, in step S2, when the edge region of the workpiece is scanned, the edge adaptive processing step S7 is performed:

[0026] The energy distribution of the flat-top line spot is adjusted to an asymmetric form by using a spatial light modulator, so that the energy on the outside of the spot is higher than that on the center by a preset compensation ratio.

[0027] And / or, switch the scan path from raster to spiral and reduce the scan speed by a preset amount.

[0028] As a further improvement of the present invention, a micro-stress fixed-point annealing step S8 is also included, which is performed after the polishing process is completed:

[0029] Based on the surface morphology data obtained by the OCT module, the surface roughness fluctuation of the whole domain is analyzed, and the area with roughness fluctuation amplitude greater than a preset stress judgment threshold is identified as a micro-stress concentration area.

[0030] In-situ fixed-point annealing of the micro-stress concentration area is performed using a femtosecond laser with a pulse width of 500 fs and an energy density of a preset annealing energy density for a preset annealing time.

[0031] Secondly, the present invention provides a polishing system for implementing the above-described polishing method, comprising:

[0032] Femtosecond laser source;

[0033] The beam shaping module is used to shape the femtosecond laser beam into a flat-top line spot;

[0034] A mobile platform is used to carry and move workpieces;

[0035] The coaxial integrated optical monitoring module includes at least an OCT module for real-time acquisition of interference signals;

[0036] A spatial light modulator, disposed in the optical path of the beam shaping module, is used to dynamically modulate the local energy density distribution of the flat-top line spot in the scanning direction by modulating the phase of the beam according to control commands.

[0037] The central control unit is communicatively connected to the femtosecond laser source, beam shaping module, mobile platform, spatial light modulator, and coaxial integrated optical monitoring module.

[0038] The central control unit is configured to: perform subsurface damage risk diagnosis based on the axial scanning signal of the OCT module, and generate control commands to adjust femtosecond laser processing parameters and / or drive the spatial light modulator for energy modulation.

[0039] As a further improvement of the present invention, the coaxial integrated optical monitoring module also includes a spectral confocal displacement sensor; the system also includes a low-power femtosecond laser annealing module.

[0040] The central control unit contains a material property database that includes the laser absorption threshold, delamination threshold, and health signal fingerprints of OCT axial signals for various brittle and hard materials.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This invention creatively mines and utilizes the deep physical information (spectral characteristics and attenuation slope) contained in the OCT axial scanning signal, using it as a subsurface health fingerprint. For the first time, it achieves real-time, in-situ diagnosis of subsurface damage risk in brittle and hard materials during femtosecond laser processing. Based on this diagnostic result, laser processing parameters are dynamically adjusted at the millisecond level, achieving a fundamental shift from post-damage detection to pre-damage prediction and suppression, thus ensuring the long-term reliability of the device from the source.

[0043] 2. Based on the above core closed loop, the flat-top line light spot with uniform energy effectively avoids the reverse effect of Gaussian light spot. Through dynamic light field modulation based on high-precision topographic feedback calibrated by spectral confocal calibration, the synergistic effect enables the surface roughness to be stably converged to the sub-nanometer level and achieves active control of nanometer-level surface shape accuracy.

[0044] 3. This invention integrates dual monitoring calibration, a material database, intelligent control algorithms, and an in-situ annealing module to construct a fully automated, end-to-end quality assurance system. This system can automatically identify and adapt to complex scenarios such as heterojunction interfaces and wafer edges, enabling one-click high-quality processing of different materials and regions, significantly reducing process complexity and improving batch consistency. Attached Figure Description

[0045] Figure 1 This is a process flow diagram of the polishing method of the present invention;

[0046] Figure 2 This is a structural block diagram of the polishing system of the present invention;

[0047] Figure 3 This is a schematic diagram illustrating the principle of correlating OCT axial signals with subsurface damage in this invention.

[0048] Figure 4 This is a schematic diagram of intelligent adaptation processing of heterojunction materials in the second embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of adaptive wafer edge processing in the second embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of in-situ low-stress annealing in the second embodiment of the present invention. Detailed Implementation

[0051] The two embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0052] First implementation method:

[0053] I. System Hardware Components

[0054] The method for polishing wafer-level brittle and hard material planes based on femtosecond lasers in this invention is implemented using a polishing system. For example... Figure 2 As shown, the polishing system mainly includes the following core components:

[0055] Femtosecond laser sources: used to generate ultrashort laser pulses, which are the physical basis for cold processing. In a specific example, a ytterbium-doped fiber femtosecond laser can be used, with an output pulse width of less than 500 femtoseconds (e.g., 100 fs); a center wavelength of 1030 nm (or 515 nm after frequency doubling); and a repetition frequency adjustable from 1 kHz to 1 MHz. The single-pulse energy must meet the material ablation requirements (e.g., up to several millijoules), and the energy stability must be better than ±3% to ensure that the thermal diffusion time is much longer than the pulse duration, thereby minimizing the heat-affected zone.

[0056] Beam shaping module: Used to convert the Gaussian beam of the femtosecond laser into a flat-topped line spot with uniform energy. This module includes, in sequence along the optical path:

[0057] Beam expander assembly: used to adjust the beam diameter to accommodate subsequent optical components;

[0058] Polarization adjustment components (such as half-wave plates) and thin-film polarization beam splitters: used to adjust the laser polarization state and serve as the beam combining / splitting interface between the processing optical path and the monitoring optical path;

[0059] The optical components used to generate the flat-top spot consist of one or more pairs of cylindrical lenses and a Powell prism. The cylindrical lenses are responsible for pre-focusing and stretching the beam in one direction, while the Powell prism converts it into a flat-top spot with highly uniform energy distribution along a straight line. After this shaping, a focal line with a length range of 5-10 mm and a width of 50-200 μm can be formed on the working surface. Its energy uniformity is better than ±5%, which can cover a large linear scanning area and significantly improve processing efficiency.

[0060] Spatial light modulator: As a key actuator of this invention, it is independently located after the optical path of the beam shaping module. A reflective liquid crystal spatial light modulator is preferred. It receives the phase modulation map from the central control unit and performs millisecond-level real-time, local phase modulation on the wavefront of the flat-top line laser beam formed after passing through the beam shaping module, thereby dynamically changing the local energy density distribution of the flat-top line spot at different positions in the scanning direction.

[0061] Focusing objective: Precisely focuses the modulated light beam onto the surface of the workpiece.

[0062] Motion platform: Used to support and drive the workpiece (wafer) to perform precise relative movement with the light beam. The platform has nanometer-level movement accuracy (e.g., ±0.1μm) in the X and Y directions, and the scanning speed is continuously adjustable from 1mm / s to 100mm / s, supporting various scanning paths such as grating and spiral.

[0063] Coaxial integrated optical monitoring module: This is the core sensing element of the invention. It is strictly coaxial with the processing optical path via a thin-film polarization beam splitter, ensuring zero parallax between the processing point and the monitoring point. This module mainly includes:

[0064] The white-light optical coherence tomography (OCT) module employs a broadband light source (e.g., a center wavelength of 830 nm and a bandwidth > 100 nm) to achieve micrometer-level (e.g., < 2 μm) axial resolution. The signals from its interferometer reference arm and measurement arm are combined and received by a high-speed spectrometer or detector. This module performs a dual function: first, it reconstructs the three-dimensional microstructure of the workpiece surface through lateral scanning to calculate roughness and surface shape errors; second, it outputs the original axial scanning interferometric signal, which contains backscattering information from different depth layers of the material and is the sole data source for subsurface health diagnosis in this embodiment.

[0065] Spectral confocal displacement sensor: To improve the absolute accuracy of topography feedback, the system preferably integrates this sensor coaxially. Its optical path can be partially integrated with the OCT optical path via a dichroic mirror. Utilizing the principle of chromatic aberration, this sensor can provide sub-nanometer (e.g., <0.5nm) absolute height measurement accuracy. In this embodiment, its main function is to periodically measure specific, fixed reference points on the OCT scanning path, providing an absolute accuracy calibration benchmark for the relative topography data of the OCT.

[0066] Central Control Unit: This is the system's decision-making brain, typically a high-performance industrial computer integrated with a real-time controller. It communicates with the femtosecond laser source, spatial light modulator, mobile platform, and monitoring modules. The central control unit has a built-in material property database, pre-stored with key parameters for various brittle and hard materials (such as 4H-SiC, GaN, and diamond), including but not limited to: femtosecond laser ablation threshold, delamination threshold, thermal conductivity, and most importantly, OCT health signal fingerprint. Furthermore, the unit runs core control algorithms for signal processing, feature extraction, decision logic generation (such as PID control and model predictive control), and spatial light modulator phase map calculation (e.g., using the Gerchberg-Saxton algorithm).

[0067] II. Core Diagnostic Principle: Correlation between OCT Axial Signals and Subsurface Damage

[0068] like Figure 3 As shown, the core of this invention lies in the in-depth analysis and application of OCT axial signals. When a femtosecond laser interacts with brittle and hard materials, if the energy density is too high or there is a cumulative effect, damage may be induced at the subsurface level (several micrometers to tens of micrometers below the surface), mainly including amorphization (lattice structure destruction) and microcrack initiation. These microscopic defects can significantly alter the local optical scattering properties and effective refractive index of the material.

[0069] Signal characteristics of amorphization: After a material becomes amorphous, its electronic band structure and phonon modes change, which may lead to a change in the absorption coefficient for light of a specific wavelength, thus affecting the backscattering spectrum of the OCT probe light at that depth. This manifests as a detectable shift in the spectral distribution (frequency domain characteristics) of the OCT axial signal after Fourier transform. For example, characteristic peaks may redshift or blueshift, or the energy proportions in specific frequency bands (such as the near-infrared band) may show characteristic changes.

[0070] Signal characteristics of microcrack initiation: Microcracks introduce a large number of subwavelength-scale gas-solid interfaces, leading to a sharp increase in light scattering. In OCT signals, this manifests as a significantly steeper decay rate (i.e., decay slope) of signal intensity with increasing depth starting from the damage initiation depth, and its decay curve deviates significantly from the exponential decay model under healthy conditions.

[0071] A health signal fingerprint is essentially pre-acquired data, obtained through calibration experiments, characterizing the "identity" of a specific material's OCT axial signal under a known undamaged state. Acquisition methods include: performing numerous OCT axial scans at different locations on a perfectly polished and annealed sample; averaging the spectra of these signals to obtain baseline spectral characteristics (e.g., a standard spectrogram); and fitting the signal amplitude-depth curve to obtain a baseline attenuation slope. The collection of these characteristic data constitutes the material's health signal fingerprint and is stored in a database. During real-time processing, by comparing the real-time signal with the "fingerprint," changes in the subsurface's health status can be diagnosed.

[0072] III. Detailed Steps of the Polishing Method

[0073] Please see Figure 1 The polishing method of this embodiment includes the following steps, which are automatically executed by the above-described system:

[0074] S1: Generates femtosecond laser pulses and shapes them into flat-top line spots with uniform energy distribution, focusing them on the surface of the workpiece to be processed.

[0075] The central control unit retrieves initial safety parameters from the database based on the selected material and sets the energy density of the femtosecond laser source (e.g., 2.5 J / cm² for 4H-SiC, with its processing window located between the ablation threshold (e.g., 2.0-2.2 J / cm²) and the delamination threshold (e.g., 3.3-3.7 J / cm²)). The laser beam sequentially passes through a beam shaping module and a spatial light modulator, with the beam shaping module forming an initial flat-top line spot. In this step, the spatial light modulator can initially load a uniform phase map (i.e., no modulation) or load a corrected phase map to compensate for inherent aberrations in the system. Finally, the spot is focused onto the workpiece surface by a focusing objective.

[0076] S2: Control the flat-top line spot to perform two-dimensional scanning relative to the workpiece surface to remove material.

[0077] The central control unit plans the scanning path (e.g., a grating path starting from the center of the wafer) and drives the moving platform to carry the workpiece at a set speed (e.g., 10 mm / s). The flat-top line spot scans the workpiece surface, and its uniform energy distribution achieves consistent ablation and removal of the material surface layer, fundamentally avoiding the "ploughing" effect and reverse roughness increase problem caused by Gaussian spots.

[0078] S3: During the grinding process, the OCT module, which is coaxially integrated with the femtosecond laser optical path, is used to acquire the interference signal on the workpiece surface in real time.

[0079] During the scanning process, the coaxially integrated OCT module continuously probes the area that has just been scanned by the light spot (the hysteresis distance is usually <1mm). The raw interference signals it acquires are transmitted to the central control unit in real time.

[0080] S4: Analyze the axial scan signal of the OCT module in real time, and extract its spectral characteristics and / or signal attenuation slope as a basis for diagnosing subsurface damage risk.

[0081] This is the core diagnostic step of the grinding method. The central control unit performs the following operations for each received axial scan signal:

[0082] S4.1: Retrieve the health signal fingerprint of the currently processed material from the database.

[0083] S4.2: Convert the real-time signal into a spectrum using Fast Fourier Transform and calculate its spectral characteristics (such as the energy centroid and peak wavelength in a specific frequency band); at the same time, perform linear or exponential fitting on the amplitude-depth curve of the signal and calculate its signal attenuation slope in the key depth range (such as 5-20 μm below the surface).

[0084] S4.3: The calculated real-time spectral features and attenuation slope are compared with the baseline value in the health signal fingerprint. When the spectral feature offset exceeds a first preset threshold (e.g., peak wavelength offset > 0.1 nm) and / or the increase in the signal attenuation slope relative to the baseline value exceeds a second preset threshold (e.g., increase > 20%), the system determines that there is a risk of damage to the subsurface of the material corresponding to the current scan point.

[0085] S5: Based on the diagnostic criteria, adjust the processing parameters of the femtosecond laser and / or modulate the energy distribution of the flat-top line spot in real time to suppress subsurface damage.

[0086] Once a damage risk is determined in step S4, the central control unit immediately (within milliseconds) initiates the damage suppression protocol:

[0087] Instantaneous energy reduction: A command is sent to the femtosecond laser source to instantaneously reduce the pulse energy density in the next or several subsequent pulse cycles. The reduction can be dynamically adjusted according to the risk level, for example, by 20%-50%.

[0088] Dynamic optical field modulation: Almost synchronously, the central control unit generates a new phase map based on the location of the risk point and the determined risk level, and sends it to the spatial light modulator. The spatial light modulator refreshes the phase map, so that when the flat-top line spot passes through the risk area or an adjacent area in the next scan, its local energy density is selectively attenuated and modulated. For example, a phase pattern is applied at the spot position corresponding to the risk point, reducing its energy to 60%-80% of the original set value.

[0089] This closed loop continues until the OCT signal characteristics collected from the area recover to a healthy range, at which point the system gradually restores normal processing parameters.

[0090] S6 (optional): Closed-loop trimming based on 3D topography.

[0091] Simultaneously, the signals acquired by the OCT module are also used to reconstruct the three-dimensional morphology of the workpiece surface. The central control unit calculates the real-time roughness and surface profile. When a local protrusion (high point) is detected, in order to accelerate surface convergence, the system can drive the spatial light modulator to enhance the local energy of the light spot in the corresponding protrusion area in subsequent scans (e.g., increase it by 5%-10%), achieving intelligent shaping through "surface-point collaboration".

[0092] IV. Dual Monitoring Calibration Process

[0093] To ensure the absolute reliability of the three-dimensional topography data used for shaping in step S6, this embodiment performs a calibration process before or periodically before step S6:

[0094] Using a coaxially integrated spectral confocal displacement sensor, the absolute height of multiple pre-defined reference points (such as several ceramic spheres of known height on the workpiece stage or special marks on the wafer edge) along the OCT scanning path is measured. The central control unit uses the true absolute height measured by the spectral confocal sensor as a benchmark and, through algorithms such as least-squares fitting, corrects for systematic errors (such as nonlinearity and drift) in the relative height data measured by the OCT module at the same batch of reference points. After this calibration, the absolute accuracy of the overall field shape data acquired by the OCT is guaranteed (e.g., reducing the overall field error from ±15nm to within ±3nm), providing a reliable sensing basis for high-precision closed-loop control.

[0095] V. Examples and Effects

[0096] Taking the fabrication of a 4-inch (approximately 100 mm in diameter) 4H-SiC wafer as an example, its initial roughness Ra is 1.5 μm, with a damage layer of approximately 10 μm. The goal is to planarize it to Ra < 0.5 nm.

[0097] Initialization: The workpiece is loaded, the system calls the SiC parameters, and establishes the morphological benchmark and health signal fingerprint.

[0098] Processing and Diagnosis: The system operates according to the S1-S6 process described above. When processing a certain area, the OCT diagnostic module detects an abnormal shift in spectral characteristics of 0.12nm, triggering a damage warning.

[0099] Suppression and Recovery: Based on the above diagnostic results (offset 0.12nm > threshold 0.1nm), the system immediately executes the damage suppression protocol of S5, instantly reducing the laser energy density from 2.8J / cm² to 1.9J / cm² (a reduction of approximately 32%), and attenuates and modulates the energy of subsequent light spots in this area using a spatial light modulator. After approximately 3 scan cycles, the OCT signal at this point returns to normal.

[0100] Shaping and Calibration: At the same time, the system performs dynamic optical field shaping on several micro protrusions based on the calibrated OCT morphology data.

[0101] Results: The entire processing time was approximately 40 minutes. Final testing showed that the surface roughness Ra reached 0.32 nm, and cross-sectional transmission electron microscopy analysis confirmed that no subsurface amorphous layer or microcracks were observed in the area that triggered the warning, verifying the effectiveness of this method in real-time diagnosis and damage suppression.

[0102] Second implementation method:

[0103] This embodiment, based on the first embodiment, further discloses one or more integrable advanced functional modules and their control methods. These functions can be enabled individually or in combination to achieve fully automated, high-quality planarization of complex workpieces.

[0104] I. Intelligent Adaptive Processing of Heterojunction Materials

[0105] like Figure 4 As shown, this system provides lossless planarization capability for heterojunction materials such as GaN-on-SiC (gallium nitride epitaxial layer grown on silicon carbide substrate).

[0106] Interface recognition: Before grinding begins (before step S1), the system uses the tomographic imaging mode of the OCT module to perform a depth scan of the workpiece. By analyzing the intensity of the reflected signals at different depths, the interface position between the GaN layer and the SiC substrate can be clearly identified and located with an accuracy of ±0.5μm.

[0107] Layered processing control: The central control unit calls up independent safe processing parameters from the material database for the GaN layer and the SiC layer respectively. For example, the energy density is set to 2.8 J / cm² and the scanning speed is 10 mm / s for the GaN layer; and more conservative parameters are set for the SiC layer: energy density of 1.2 J / cm² and scanning speed of 8 mm / s.

[0108] Endpoint Determination: The system monitors the OCT signal in real time to determine the removal depth. When the real-time monitoring shows that the remaining GaN layer thickness is close to a preset critical distance (e.g., 1.0 μm) at the interface, the system automatically issues an early warning and switches to SiC layer processing parameters. When the OCT axial signal shows a sudden change in the intensity of the characteristic reflection peak at the interface (corresponding to the difference in refractive index between the two layers), the system determines that the processing has ended and stops immediately.

[0109] Results: Processing a 2-inch GaN-on-SiC wafer using this process can polish the GaN layer surface to Ra < 0.4 nm, while cross-sectional SEM observation confirms that the interface is clear, complete, and without damaged layer transitions.

[0110] II. Adaptive processing of wafer edges

[0111] like Figure 5 As shown, the system integrates an edge adaptive module to address the issue of decreased processing quality caused by laser scattering and stress concentration in the wafer edge region.

[0112] Edge recognition: Based on the surface topography image acquired in real time by the OCT module, the area within a certain width (e.g., 2mm) from the physical edge is automatically labeled as the "edge processing area" through image processing algorithms (such as threshold segmentation and edge detection).

[0113] Optical field compensation: When processing enters the edge region, the central control unit instructs the spatial light modulator to modulate the originally energy-symmetrical flat-top line spot into an energy-asymmetric distribution perpendicular to the scanning direction (Y direction). Specifically, by loading a linearly or gradient-varying phase map, the energy distribution gradient on the outer side of the spot (the side closer to the wafer edge) is temporarily enhanced, achieving an 8%-12% energy compensation, thereby effectively offsetting the laser energy scattering loss at the edge.

[0114] Path and speed optimization: The scanning path automatically switches from a grating-type path in the central area to a spiral path to improve edge stress distribution; at the same time, the scanning speed is reduced by 15%-20% (e.g., from 10mm / s to 8.5mm / s) to ensure uniform material removal.

[0115] Results: After full-area processing of the 6-inch SiC wafer, measurements showed that the average Ra of the edge region (outermost 2mm ring) was 0.31nm, with a difference of less than 7% from the center region (Ra=0.29nm), achieving full-wafer consistency.

[0116] III. In-situ low-stress annealing

[0117] like Figure 6 As shown, to eliminate any residual micro-stress that may remain after femtosecond laser cold processing, the system can integrate a separate low-power femtosecond laser annealing module. This module typically uses a femtosecond laser from the same source as the main light source, but its energy density is attenuated to a lower level.

[0118] Stress diagnosis: After the main grinding process is completed, the central control unit analyzes the global final surface morphology data acquired by the OCT module and calculates the fluctuation of its micro-roughness. Continuous areas with roughness fluctuation amplitude (such as local roughness standard deviation) greater than a preset stress judgment threshold (e.g., >0.05μm) are identified as potential micro-stress concentration areas.

[0119] Point-to-point annealing: The control platform sequentially moves the aforementioned stress concentration areas to the focal point of the annealing module. The annealing module uses a femtosecond laser with a pulse width of approximately 500 fs and a low energy density (e.g., 0.2-1.0 J / cm², 1 / 5 to 1 / 3 of the main processing energy) to irradiate each stress point for a preset annealing time (e.g., 1-2 seconds). This low-energy, long-duration laser irradiation promotes local atomic relaxation, thereby eliminating micro-stress.

[0120] Effect verification: After annealing, the stress region was detected by micro Raman spectroscopy. It was observed that the average half-width of the characteristic peaks of the lattice vibration was reduced by about 15%, indicating that the lattice distortion had been recovered and the residual stress was estimated to be less than 50 MPa.

[0121] In light of current practical needs, the above-described embodiments of this invention are not limited to these specific implementations. Any changes made within the scope of knowledge possessed by those skilled in the art, without departing from the concept of this invention, still fall within the protection scope of this invention.

Claims

1. A method for polishing wafer-level brittle and hard material planes using femtosecond lasers, characterized in that, Includes the following steps: S1: Generates femtosecond laser pulses and shapes them into flat-top line spots with uniform energy distribution, focusing them on the surface of the workpiece to be processed; S2: Control the flat-top line spot to perform a two-dimensional scan relative to the workpiece surface to remove material; S3: During the polishing process, the OCT module, which is coaxially integrated with the femtosecond laser optical path, is used to acquire the interference signal on the surface of the workpiece in real time. S4: Analyze the axial scanning signal of the OCT module in real time, and extract its spectral characteristics and / or signal attenuation slope as a diagnostic basis for subsurface damage risk; S5: Based on the diagnostic criteria, reduce the pulse energy density of the femtosecond laser in real time and / or modulate the energy distribution of the flat-top line spot using a spatial light modulator to suppress subsurface damage.

2. The polishing method according to claim 1, characterized in that, Step S4 specifically includes: S4.1: Pre-store the health signal fingerprint of the workpiece material. The health signal fingerprint is used to characterize the standard signal features of the corresponding material in a non-damaged state, including the reference spectral features and the reference attenuation slope. S4.2: Compare the spectral characteristics and / or attenuation slope of the real-time acquired axial scan signal with the health signal fingerprint; S4.3: When the spectral characteristics shift by more than a first preset threshold and / or the attenuation slope exceeds a second preset threshold, it is determined that there is a risk of subsurface damage.

3. The polishing method according to claim 2, characterized in that, The real-time adjustment of the femtosecond laser processing parameters mentioned in step S5 includes: After determining that there is a risk of damage, the pulse energy density of the femtosecond laser is instantaneously reduced before the next or subsequent laser pulse; and / or, the local energy distribution of the flat-top line spot in the scanning direction is dynamically modulated by a spatial light modulator to reduce the energy density of the corresponding risk area. The spatial light modulator, based on the location and severity of the risk area, modulates the local energy density of the flat-top line spot as it scans through the risk area, attenuating it relative to the currently set energy density value and corresponding to the damage risk level.

4. The polishing method according to claim 1, characterized in that, In step S3, the OCT module is also used to reconstruct the three-dimensional morphology of the surface; the polishing method further includes step S6: Based on the three-dimensional topography information, the local energy density distribution of the flat-top line spot is dynamically adjusted by the spatial light modulator to perform closed-loop trimming of the surface topography.

5. The polishing method according to claim 4, characterized in that, Before step S6, the absolute height of multiple specific reference points on the OCT scanning path is measured using a spectral confocal displacement sensor coaxially integrated with the femtosecond laser optical path. Based on this measurement data, the overall field shape data acquired by the OCT module is systematically corrected for errors.

6. The polishing method according to claim 1, characterized in that, The wafer-level brittle and hard material has a heterojunction structure; the polishing method further includes step S0 performed before step S1: The location of the heterojunction interface is identified using the tomographic imaging capability of the OCT module; In step S2, different processing parameters are applied to different material layers according to the interface position, and the processing endpoint is determined based on the real-time OCT axial signal characteristics; wherein, when the real-time monitored removal depth approaches a preset critical distance of the interface, the processing parameters are automatically switched to conservative.

7. The polishing method according to claim 6, characterized in that, In step S2, when the edge region of the workpiece is scanned, the edge adaptive machining step S7 is performed: The energy distribution of the flat-top line spot is adjusted to an asymmetric form by the spatial light modulator, so that the energy on the outside of the spot is higher than that on the center by a preset compensation ratio. And / or, switch the scan path from raster to spiral and reduce the scan speed by a preset amount.

8. The polishing method according to claim 7, characterized in that, It also includes the micro-stress point annealing step S8 performed after the polishing process is completed: Based on the surface morphology data obtained by the OCT module, the global surface roughness fluctuation is analyzed, and the area with a roughness fluctuation amplitude greater than a preset stress judgment threshold is identified as a micro-stress concentration area. The micro-stress concentration area is subjected to in-situ fixed-point annealing for a preset annealing time using a femtosecond laser with a pulse width of 500 fs and an energy density of a preset annealing energy density.

9. A polishing system for implementing the polishing method according to any one of claims 1-8, characterized in that, include: Femtosecond laser source; The beam shaping module is used to shape the femtosecond laser beam into a flat-top line spot; A mobile platform is used to carry and move workpieces; The coaxial integrated optical monitoring module includes at least an OCT module for real-time acquisition of interference signals; A spatial light modulator is disposed in the optical path of the beam shaping module and is used to dynamically modulate the local energy density distribution of the flat-top line spot in the scanning direction according to control commands. The central control unit is communicatively connected to the femtosecond laser source, beam shaping module, moving platform, spatial light modulator, and coaxial integrated optical monitoring module. The central control unit is configured to: perform subsurface damage risk diagnosis based on the axial scanning signal of the OCT module, and generate the control commands to adjust the femtosecond laser processing parameters and / or drive the spatial light modulator to perform energy modulation.

10. The system according to claim 9, characterized in that, The coaxial integrated optical monitoring module also includes a spectral confocal displacement sensor; the system also includes a low-power femtosecond laser annealing module. The central control unit contains a material property database that includes laser absorption threshold, delamination threshold, and health signal fingerprints of OCT axial signals for various brittle and hard materials.

Images