a State Key Laboratory of Tribology in Advanced Equipment, Tsinghua University, Beijing 100084, China
b Beijing Key Laboratory of Transformative High-End Manufacturing Equipment and Technology, Department of Mechanical Engineering, Tsinghua University, Beijing 100084, China
c School of Mechanical Engineering and Automation, Beihang University, Beijing 100191, China
The non-physical-contact property of lasers poses significant challenges for the alignment procedure in precision engineering. Particularly in galvanometer-based laser processing systems, the requirement for multiple-step coordinate conversion further complicates the alignment procedure, thereby increasing the potential for error accumulation. To address the alignment issues during galvanometer laser scanning, this paper proposes an alignment-error-free solution for full-in-situ imaging and laser processing system, which eliminates the alignment error at the principal level by skipping the coordinate conversion and directly extracting angular coordinates for laser scanning from the captured images. Compared with the existing galvanometer-based laser processing systems, the main advantage of the proposed method is its ability to achieve alignment-error-free without requiring calibration, making it particularly suitable for small-batch, highly customized, and complex processing tasks. This system specifically facilitates in-line inspection, detection, and measurement during laser fabrications. Furthermore, two experimental cases in pan-semiconductor manufacturing, which includes flexible printed circuits (FPC) cutting and micro-light emitting diodes (Micro-LEDs) defect detection, have been conducted to demonstrate the validation of the proposed full-in-situ processing system. Correspondingly, the current experimental comparisons highlight the superiority of the proposed system for simultaneously achieving a maximum range of 27 mm × 27 mm and a minimum resolution of 0.412 µm, with a maximum processing error <15 µm. Demonstrations in detecting and processing the complex patterns illustrate its exceptional capabilities in alignment-error-free laser processing for precision manufacturing.
Laser processing technologies have been widely recognized as a promising approach for precision engineering tasks, including micro-patterning [1,2], micro-fabrication [3,4], surface modification [5,6], and so on, owing to their attributes of high speed, high precision, and non-contact nature. Particularly in the domain of pan-semiconductor manufacturing, the non-contact feature of lasers confers minimum stress to the substrate, thereby significantly reducing the damage and side-effects on the devices [7]. Galvanometer laser scanning technology is one of the laser steering technologies, whereby the deflection angles of the galvanometer mirrors are manipulated to steer the laser beam and precisely determine the laser spot's position. The rapid rotation of these mirrors at high angular speed enables the ultra-fast scanning of the laser spot. In applications demanding high throughput, such as high-density laser drilling [8], laser polishing [9], and selective laser melting [10,11], galvanometer laser scanning technology is extensively utilized due to its remarkable advantage in achieving ultra-fast scanning speeds and superior selectivity.
Ensuring precise alignment of the laser beam with the desired processing object is mandatory for laser systems to guarantee precision. However, given the non-contact nature of laser beams, traditional physical alignment methods are constrained as direct physical contact is unfeasible [12]. Within galvanometer systems, the task of converting the intended processing path's planar coordinates into the galvanometers’ deflection angles presents an even greater challenge. Hence, addressing the alignment issue has emerged as a critical concern for galvanometer systems [13].
A crucial approach to address this alignment challenge is the integration of in-situ inspection subsystems. Numerous studies have demonstrated that in-situ/on-machine inspection and measurement serve as pivotal technologies for monitoring processing procedures [14-18] and enhancing the alignment efficiency and accuracy of ultra-precision manufacturing [19-22]. Specific to the galvanometers, back-reflected detection can identify focus errors from parameters such as position [23], intensity [24], and beam size [25,26], which is utilized for axial alignment [27]. Structured light is employed to obtain the three-dimensional (3D) shape of the surface, thereby aiding in three-dimensional positioning alignment [28-31]. Vision-based methods capture the operating plane’s images directly and determine the laser path through image analysis. Notably, the vision-based methods can acquire more comprehensive information of the plane from the images, offering greater advantages in the identification and processing of planar features [32-34], thus is widely applied.
In the existing galvanometer-based in-situ systems, pattern images are captured by cameras in pixel coordinates. To enable laser processing, the trajectories must be specified in terms of the galvanometer mirrors’ deflection angles. As a result, a conversion process is necessary, initially translating the camera's pixel coordinates into the scanning plane's planar coordinates and subsequently into the galvanometers' angular coordinates (Fig. 1(a)). The accuracy of this multiple-step conversion process significantly influences the alignment precision of the entire system. The conversion accuracy is affected by numerous factors, including assembly errors in the galvanometer scanning system [35,36], field distortion of the F-theta lens [37,38], and the precise positional relationship between the inspection subsystem and the galvanometer scanner [34]. Consequently, most existing studies focus on enhancing alignment accuracy through calibrations addressing these factors. The field distortion of the F-theta lens is significantly reduced through calibrations in Refs. [39,40], achieving calibration errors of 7.3 and 11 μm, respectively. The alignment problem in the scanning laser Doppler vibrometer is addressed in Refs. [33,34] using calibrations, enabling the targeting of points 2 m away with an accuracy of 10 mm [34]. Additionally, the calibration of the triangulation device is optimized in Ref. [41], realizing about 0.5 mm three-dimensional calibration error with a 0.7 m distance between the scanning system and the targets.
However, the alignment accuracies achieved above are highly dependent on the accuracy of multiple calibrations for error compensation, which is an essential step in the existing approaches. For instance, in Ref. [40], after 6 rounds of calibration iterations, the accuracy of converting planar coordinates to angular coordinates reached 11 μm. Besides, the errors during the conversion from pixel coordinates to planar coordinates remain to be addressed. To address the above problems, this paper proposes a galvanometer-based full-in-situ imaging and laser processing system that directly obtains angular coordinates for processing from the imaging results. The proposed system eliminates the need for two-stepped coordinate conversion, thereby preventing the introduction of errors associated with conversion (Fig. 1(b)). Consequently, it has the advantage of achieving alignment-error-free laser processing at the principal level without the need for any calibrations. Furthermore, the existing vision-based in-situ methods are limited by the imaging capabilities of the camera, making it difficult to simultaneously achieve both large-range and high-precision imaging. In contrast, the proposed system could achieve variable ranges and resolutions, simultaneously supporting both large-range imaging and high-resolution imaging.
For those straightforward tasks like repetitive processing of identical patterns on fixed samples, or tasks where position precision is not critical, the existing galvanometer scanning system performs adequately with a set of calibration procedures. The advantages of the proposed system may not be obvious in such scenarios. However, for tasks requiring high flexibility, such as small-batched, highly customized, and complex processing, in-line inspection, detection, and precise trimming are strongly demanded, making the proposed system particularly well-suited for such applications.
The rest of the paper is organized as follows. Section 2 introduces the full-in-situ system and provides analyses on the factors influencing the image quality. Section 3 details the system's implementation along with a demonstration of its performance. In Section 4, two experimental cases are conducted: the special-shaped hole cutting of flexible printed circuits (FPCs) and the quality control of micro-light emitting diode (Micro-LED) chips. Finally, the conclusion is presented in Section 5.
2. Method analyses
This section analyzes the structure of the proposed system and further compares the workflow of the proposed system with those of the existing systems, showing the proposed system’s elimination of the coordinate conversion, thereby avoiding the necessity for calibrations and eliminating the alignment errors. Furthermore, a discussion on the primary factors influencing the quality of the captured image is provided.
2.1. System structure
A full-in-situ digital galvanometer imaging and laser processing system is proposed in this study, as shown in Fig. 2(a). To achieve full-in-situ laser processing, it’s required to first acquire the scanning plane’s image, and subsequently decide the processing coordinates according to the imaging result. Accordingly, the proposed system consists of two subsystems: the imaging subsystem (Fig. 2(b)) and the processing subsystem (Fig. 2(c)), tasked with managing imaging and processing operations, respectively.
In the imaging subsystem, the scanning plane is precisely positioned at the focal plane of the F-theta lens. Simultaneously, the luminance sensor is positioned at the focusing point of the measurement focusing lens. This configuration establishes a double telecentric imaging system, which effectively focuses the light emitted from the scanning plane onto the luminance sensor and virtually eliminates perspective angle error [36,37]. By varying the deflection angle of the galvanometer, the luminance sensor captures and records the light intensities at various points on the scanning plane. This system operates in a point-by-point scanning manner, where the image of the scanning plane is generated by organizing the recorded light intensities of each point as grayscale values. These values are organized based on their respective positions in the angular coordinate system, which directly correspond to the deflection angles of the galvanometers for each point. This configuration allows for the acquisition of images with various sizes and resolutions by regulating the scanning range and sampling interval of the galvanometer scanner. Consequently, a large-ranged fine scan with simultaneously high resolution is applicable. Furthermore, a "large-ranged coarse scan, zoom-in, small ranged fine scan" procedure can be implemented, balancing both speed and image clarity (Note S1, Figs. S1 and S2 in Appendix A).
The laser processing subsystem follows a typical galvanometer laser scanning configuration, as depicted in Fig. 2(c). In this subsystem, the laser beam emitted from the laser source is steered by the galvanometer mirrors. These mirrors adjust and control the beam’s deflection angle to achieve precise scanning. After being steered, the laser beam passes through the F-theta lens and gets focused onto the scanning plane. The F-theta lens guarantees a constant spot size for the laser beam across the entire scanning area, enabling precise and uniform laser processing.
By combining these two subsystems, the full-in-situ imaging and laser processing can be achieved by carefully aligning the position of the half-transparent mirror and the luminance sensor, ensuring that the luminance sensor and the laser source share the same equivalent optical path. The system’s operation process begins with obtaining an image of the scanning plane without the laser. Subsequently, it identifies the specific points that require laser processing according to the image. Upon activating the laser, the laser processing subsystem can precisely target the desired points by setting the galvanometer deflection angles exactly the same as the angular coordinates of those specific points, which were obtained during the initial imaging phase.
According to the principle of reversible optical path, when the galvanometers are arranged to a set of specific deflection angles, the light emitted from a particular point on the scanning plane can reach the luminance sensor, and this set of deflection angles is recorded as the angular θx-θy coordinates of this point on the image. Subsequently, by maintaining the same galvanometer deflection angles, since the sensor and the laser source share the same equivalent optical path, the laser emitted from the source can also be reflected by the galvanometers to exactly reach this point for laser processing. This approach ensures that no calibration or alignment error occurs, enabling exact localization of the laser on the intended points. As a result, full-in-situ imaging and laser processing can be achieved.
This concept can be extended to various imaging and processing scenarios where intensive and complex calibrations are required, such as the dynamic focus 3D galvanometer scanning system. The system described in this paper currently concentrates solely on imaging and processing within the X-Y plane. When confronted with a 3D surface with variation along the Z- axis, this system encounters defocusing challenges, leading to a decrease in both imaging and laser processing quality. By integrating a two-dimensional (2D) galvanometer system with a dynamic (pre)focus module, the resulting dynamic focus 3D galvanometer scanning system adjusts the height of the laser focus point by modifying the position of the dynamic focus module, ensuring precise focusing even on 3D surfaces. However, the calibration of the dynamic focus module's position with respect to the Z-axis position of the laser focus point also poses great difficulties. By applying the automatic focusing algorithm based on sharpness optimization, it is feasible to generate the 3D surface topography maps in three dimensions without calibration: the galvanometers’ deflection angles as the X-Y axes, and the dynamic focus module’s displacement as the Z- axis. Subsequently, by utilizing the original data from surface countering, precise laser focusing on the desired points can be achieved, thus enabling full-in-situ 3D laser processing.
2.2. Comparison between existing systems and alignment-error-free system
The workflow of the existing in-situ galvanometer-based laser processing systems is illustrated in Fig. 3(a), using the defect detection of a Micro-LED chip array as an example, where selective laser scanning is regarded as the most promising approach [42-44]. The misaligned dies in the chip array should be detected and removed by laser scanning. The above process comprises two primary steps: The first one is to capture the planar image, locate the misaligned dies, and calculate the coordinates (x,y) of the target processing points in the planar coordinate system XOY. The origin O is located at the position of the beam focus spot when the galvanometers operate with the angular coordinates θx=0 and θy=0. The second step involves converting the coordinates (x,y) into the angular coordinates (θx,θy), which are utilized for the subsequent laser processing. During step 1 of Fig. 3(a), the fine calibration of the camera is necessary to ascertain the accurate coordinates of the target point in the XOY coordinates from the captured images. While in step 2, the errors in the mechanical assembly [35,36] and the optical path of focusing lenses [37,38] result in deviations in the conversion from the planar coordinates to the angular coordinates (e.g., the field distortion). This leads to coordinates results that deviate from the theoretical predictions, thereby making it essential to perform additional calibration and compensation to ensure precise conversion.
Importantly, the proposed system is designed to directly obtain the angular coordinates of target processing points, eliminating the need for the two-step coordinate conversion. Therefore, errors stemming from the conversion are not introduced, ensuring alignment-error-free in-situ laser processing. In comparison to the existing systems, the proposed system also advantages in its capability of achieving both large working range and high resolution. As depicted in Figs. 3(b) and (c), vision-based approaches could be categorized into off-axis and coaxial ones according to the camera’s position. An off-axis camera could achieve a large image range [28,29], and coaxial camera could obtain an image with high resolution [36]. It is noticed that achieving both large range and high precision simultaneously is constrained by the imaging capability of the camera. Employing regional observation and stitching techniques is a feasible method to achieve both large-range and high-resolution imaging. However, the adoption of stitching techniques inevitably introduces image stitching errors and thereby reducing processing accuracy, which also necessitates time-consuming calibrations for compensation [45,46]. In contrast, the imaging range and resolution of the proposed system keep consistent with those of the galvanometer scanning system, enabling both large-field imaging and high-resolution imaging without any stitching error, as shown in Fig. 3(d).
2.3. Image quality analyses
While the proposed system achieves alignment-error-free in-situ laser processing, its capacity for feature recognition is influenced by the image quality. Theoretically, an ideal luminance sensor would be able to measure the light intensity of an infinitely small ideal point on a 2D plane, thus fully restoring all the features of the observed pattern. However, in practice, the sensor itself encompasses a certain area of measurement region, and can only measure the average light intensity across a small region rather than a specific point. This limitation leads to the blurring of the acquired image, compromising observation efficacy, and making it challenging to differentiate features that are too close to each other. In this subsection, the main factors affecting the quality of the acquired image are analyzed, aiming to minimize the impact of image blurring and thereby enhancing the system's processing capabilities.
To provide further clarification, as stated in Section 2.1, the imaging subsystem employed in this study is a double telecentric system, with the equivalent light path diagram showed in Fig. 4(a). As a result, the scanning plane is projected onto the luminance sensor at a constant magnification. To facilitate comprehension in subsequent discussions, this projection process can be equivalent to projecting the sensor onto the scanning plane, resulting in the formation of a conjugate image of the sensor’s measurement region on the plane, as depicted in Fig. S3 in Appendix A. The recorded measurement signifies the weighted average light intensity within the region on the scanning plane, which is encompassed by the projected conjugate image of the sensor’s measurement region. As the deflection angle of the galvanometer alters, the system scans across the plane, causing the position of the conjugate image to shift correspondingly.
Since the sensor captures the weighted average of light intensity values within its measurement region, the conjugate image of this region can be likened to a convolution kernel. Consequently, the imaging process of this system can be likened to convolving the original image. As the system scans across the targeted scanning plane, the conjugate image also traverses the plane, resulting in the convolutional outcome of the original image—a blurred image [47-49]. Notably, a smaller conjugate image of the sensor’s measurement region, implying a reduced convolution kernel, leads to less blurring in the resulting image, as illustrated in Figs. 4(b)-(d).
The size and characteristics of the conjugate image are determined by both the properties of the sensor’s measurement region and the optical magnification of the imaging subsystem. The magnification can be computed using the following method. Fig. 4(a) illustrates the equivalent light path diagram of the imaging subsystem. In this diagram, L1 represents the focusing lens for measurement, while L2 represents the F-theta lens. An object O (which corresponds to the measurement region of the luminance sensor) with a height of h1 is positioned at h1, the focal point of F1. The conjugate image O′ of O is projected at F2, the focal point of L2, with a height of h2. As O is located at F1, the light passing through L1 from O should be parallel. Therefore, the following relationships hold:
where M represents the optical magnification, which is determined by the focal lengths of the lenses. θ1, θ2 θ3, and θ4 represents the angle between the light rays emitted from the top of object O passing through the optical center, and the optical axis. Typically, the value of f2 is determined by the specific requirements of the laser processing, while f1 can be adjusted as desired. Consequently, selecting a measurement focusing lens with a longer focal length will yield a smaller magnification, resulting in a reduced size of the conjugate image and a smaller convolution kernel. Ultimately, this leads to a less blurred image. Meanwhile, choosing a more precise luminance sensor with a smaller measurement region also contributes to reducing the kernel size and obtaining a clearer image.
Apart from reducing the convolution effect by reducing the size of the convolution kernel, improving the hardware performance will also help to improve the image quality. For example, utilizing a more refined design of the focusing lens system and employing optical coatings with superior performance can help enhance the dispersion correction, and reducing the aperture size can improve the image sharpness, which will all make contributions to the image quality. However, reducing the aperture size can also lead to reduced incident light, thereby increasing the required exposure time and diminishing the signal-to-noise ratio, consequently extending the overall imaging time. Moreover, excessively small apertures can induce diffraction effects that, conversely, degrade image quality and constrain the laser processing power [50].
Spatial resolution (SR) is one of the evaluation metrics commonly employed by researchers to quantitatively assess the image quality, reflecting the sharpness and clarity of the captured image. It signifies the imaging system's capability to differentiate visual features that are very close to each other [51-53]. Adjacent features with a distance smaller than the SR value will merge all together in the obtained image, rendering them indistinguishable. Even with the application of image enhancement techniques such as sharpening, distinguishing these features remains impossible due to the already compromised original image information. More detailed examples with detailed explanation of the SR can be found in Note S2, Figs. S4 and S5 in Appendix A.
For a certain kernel characteristic, a larger kernel size will result in a more blurred image with lower quality and exhibit lower SR, leading to limitations of processing capabilities for adjacent features. Enlarging the focal length results in a smaller conjugate image of the measurement region, a smaller kernel size, and a higher SR, as illustrated in Figs. 4(b) and (c). More detailed experiments are conducted in Section 3.3.2 to evaluate the SR of the proposed system.
3. Prototyping system
To validate the proposed method, a prototyping system is constructed. In this section, the detailed descriptions and analyses are provided regarding the prototype's configuration and performance.
3.1. Prototype configuration
Fig. 5(a) illustrates the prototyping system utilized for subsequent full-in-situ imaging and laser processing experiments. A self-developed galvanometer scanner enables a maximum scanning range of 0.26 rad (optical angle) and achieves a minimum resolution of 0.5 µrad (optical angle). The F-theta lens employed is the JENar lens from Jenoptik, Germany, possessing a focal length of 100 mm. The luminance sensor module is derived from the TCD1304 module from TOSHIBA, Japan, covering a measurement region of 200 µm × 200 µm. Additionally, a convex lens with a focal length of 150 mm, serving as the measurement focusing lens, is set inside the optical tubes together with the half-transparent mirror (indicated by the yellow dashed lines in Fig. 5(a)). A 355 nm laser beam emitted from an Nd:YAG laser source (Bellin Laser, China) is introduced into the system through multiple reflections. It is worth noting that all optical instruments in the system are optimized for the unique operating wavelength of 355 nm to prevent image overlap caused by repeated imaging at multiple wavelengths. As for the scanning plane, a black Z-axis motion stage is utilized.
Theoretically, the maximum range and minimum resolution of the system are 50 mm × 50 mm (the maximum field of view of the current F-theta lens) and 0.103 µm, respectively. In this study, with the current deployed communication protocol, a maximum scanning area of 27 mm × 27 mm and a minimal scanning step (i.e., resolution) of 0.412 µm are achieved, both for imaging and laser processing. With future upgradations of the protocol, it is expected that the theoretical performance will be attained.
The scanning speed is determined by both the time required per scanning point and the size of the scanning step. The shorter time needed to collect luminance data for each point, and an increase in the step size, will both result in a higher scanning speed. The current data transfer bandwidth allows for luminance measurement of maximum around 14 000 points per second. While enabling large-scale and high-resolution imaging simultaneously, this technique necessitates traversing 4 × 109 points when using the maximum range (27 mm × 27 mm) and highest resolution (i.e., minimum scanning step, 0.412 μm), leading to excessively long imaging time. In practice, the need for maximum range and highest resolution imaging simultaneously is rare. Therefore, for most cases, a large-ranged coarse scan is first conducted with large scanning step and resolution to identify and locate the pattern, followed by a small-ranged fine scan with much smaller scanning step and resolution to acquire a higher-quality image, which helps to balance between image quality and efficiency. In cases of extreme requirements, imaging with maximum range and highest resolution is still feasible, though at the expense of extended time. With higher bandwidth communication protocols (such as universal serial bus (USB) at 30 MB⋅s-1), the measurement rate could be increased to approximately 10 million points per second, which will significantly enhance the imaging speed.
A magnetic suction blue lamp is installed as an auxiliary illumination source for imaging purposes, corresponding to a sensor exposure time of 2 μs. For the patterns with low contrast and strong background noise, increasing the exposure time or boosting the power of auxiliary lighting sources will help to enhance the contrast and the signal-to-noise ratio, making it easier to identify the desired patterns. The exposure time should be kept not too long (in this study, less than 35 μs, which is half of the data transfer time) to prevent slowing down the luminance data acquisition for each point and the overall imaging speed.
3.2. One-step focusing and alignment
A one-step focusing and alignment of the sensor needs to be completed at the initial stage. No further calibrations and adjustments are needed after this one-step procedure. To guarantee the optimal image quality and precision, exact focusing of both the F-theta lens and the measurement focusing lens is essential. Initially, the vertical position of the scanning plane is determined to achieve optimal focus for the F-theta lens. This is accomplished by reducing the laser source’s output power to a low level and adjusting the height of the scanning plane until the laser spot’s focal area on the plane is minimized. Subsequently, the vertical position of the measurement focusing lens is fine-tuned by rotating the optical tube. Due to the lack of intuitive judgment regarding the lens’s focusing status, an experiment was devised to ascertain its focus. To assess the lens’s focus status, a single line scanning process is performed on a black-to-white sharp edge, yielding a grayscale figure of the line (Figs. 5(b) and (c)). To evaluate the lens’s focus, two commonly utilized functions for 2D image focusing evaluation are introduced: the energy of gradient (EOG) function and the variance (VAR) function [54,55]. The functions are shown below:
where fx,y stands for the grayscale of the image at the x,y coordinate point, μ stands for the average of the whole image’s grayscale. These focus measurement functions attain their maximum values when the lens is fully focused. In our specific case, the aforementioned formulas can be simplified to their one-dimensional (1D) counterparts:
In this experimental procedure, the system is initiated to scan a single line and subsequently record the values of EOG1D and VAR1D from the resulting 1D image. Subsequently, we proceed to adjust the position of the focusing lens and repeat the aforementioned process for the same scanning line. The measurement focusing lens is considered adequately focused when the maximum values of both functions are attained, as illustrated in Fig. 5(d).
It is necessary for the luminance sensor and laser source to share an equivalent optical path, as outlined in Section 2.2. In this prototype, the half-transparent mirror is fixed within the optical tubes without adjustability. Therefore, adjustments are made to the position of the luminance sensor. Once the focusing process is completed, the alignment procedure can be easily achieved by initially configuring the laser source to a low power output and subsequently adjusting the sensor’s position until peak light intensity data is recorded, indicating successful alignment with the laser's center. Failure to align the sensor's position with the laser's center would result in an amplified mismatch magnified by the optical magnification factor M (in this prototype, 0.67) onto the processing plane. This would lead to a mismatch between the laser spot’s actual processing position and the intended processing position, introducing alignment errors. After the described one-step procedure, no more adjustments are needed in all the following steps.
3.3. Performance demonstration
3.3.1. Variable range imaging
An imaging process is first conducted using the maximum scanning range, and the result is compared with that obtained from a coaxial camera, as depicted in Fig. 5. The scanning range of the galvanometer is configured at 0x0000-0xFFFF with a sampling interval of 0x0200 counts, and the acquired imaging result is depicted in Fig. 5(e). The imaging range in this instance corresponds to the actual size of 27 mm × 27 mm on the scanning plane, which matches the scanning range of the galvanometers mentioned in Section 3.1. In contrast, when capturing images directly using a coaxial camera, the range of the resulting images is limited to no more than 4.1 mm × 2.3 mm (Fig. 5(f)). Subsequently, employing the same imaging range as the camera utilized in Fig. 5(f), a fine scanning process (sampling interval: 0x0020 counts) is conducted using the proposed system, resulting in the image depicted in Fig. 5(g). The small-ranged zoomed-in image by the proposed system also exhibits commendable imaging quality.
3.3.2. SR
To demonstrate the SR of the proposed system, which represents the graphic quality of the obtained image as mentioned in Section 2.3, an image capturing the sharp edge of the black-to-white boundary region is acquired. Subsequently, the SR is determined by analyzing the edge spread function (ESF) and line-spread function (LSF) derived from the image of the aforementioned boundary.
Fig. 5(h) displays the image of the previously mentioned boundary with a 12.8 µm interval. Fig. 5(i) illustrates the ESF and LSF of the proposed system. The ESF is obtained by fitting the normalized intensity to an error function, following which the LSF is calculated through the derivative of the ESF. Then the SR, measured as 93 µm, is determined by evaluating the full width at half maximum (FWHM) of the LSF.
Various combinations of measurement focusing lenses and sensors were also tested, and their SR was evaluated, as presented in Table 1. The BH1750 sensor being discussed here is a sensor with a measurement region area of approximately 500 µm × 500 µm. Its performance is inferior to that of the previously employed TCD1304 based sensor. The results of the evaluation further validated the notion mentioned in Section 2.3, that longer focal lengths combined with more precise sensors yield lower SR values, thereby resulting in sharper image.
The critical feature size of the cases demonstrated in this study are 2 mm (minimum separation between the corners) and 150 µm (minimal distance between the chips), respectively, both larger than the SR of the current prototype. Hence, the current SR of 93 µm is enough for these processing cases and will not limit the overall performance. In other situations requiring a higher SR, it can be achieved by employing a lens with an elongated focal length and using a sensor with reduced measurement areas, allowing for reduced convolution effect and improved imaging capabilities. This approach is expected to lead to a significant increase in the scale of the prototype, thus is not attempted in this work.
It is worth noticing that the SR is determined by the optical path setup, representing the sharpness and clarity of the captured image, consequently influencing the capacity for feature identification and the following path planning step. The processing resolution equals to the galvanometer’s minimum scanning step, which is 0.412 µm as shown in Section 3.1, and determines the system’s capability to execute the planned processing path. Notably, these two elements operate independently: Enhancing SR does not influence processing resolution, and conversely, refining processing resolution does not enhance SR. Low SR and processing resolution values will restrict the system’s ability of recognition and processing, respectively, thus together constraining the overall performance of the system. Typically, the processing resolution is significantly smaller than the SR, with values in this manuscript being 0.4 and 93 μm, thus the performance of the system is mainly constrained by the SR. For the system described in this study, given the imaging method of point-by-point scanning, the processing resolution also determines the pixel spacing of digital imaging. In extreme cases where the processing resolution approaches or even exceeds the SR, the pixel spacing in the image becomes too large and the image displays aliasing and grains. As a result, the image quality becomes determined by the processing resolution rather than the SR. However, this scenario is unlikely to occur because, as explained before, processing resolution is usually significantly smaller than SR.
4. Case studies and experiments
To validate the system’s processing capabilities, two distinct case studies are designed and conducted using the above prototype.
4.1. Special-shaped hole cutting of FPCs
FPCs are extensively employed in contemporary electronic devices due to their capacity to adapt to intricate and irregular geometries. To enable component placement, interconnectivity, and the downsizing of electronic devices, the processing of special-shaped holes is necessary for FPCs.
The predominant approach at present involves mechanical punching and milling, which provides efficient production but is restricted in its ability to handle complex hole shapes. These traditional machining methods are constrained by mold costs and precision, rendering them inadequate to meet the evolving requirements of highly complex and precise patterns. Laser stands out with its distinctive advantages, offering exceptional precision, enabling flexible design of hole shapes, demonstrating compatibility with a wide range of materials, minimizing mechanical stress on FPCs, and ensuring a clean and residue-free process. Nevertheless, the alignment of the laser with the target cutting location remains a persistent challenge.
In this experiment, the proposed laser full-in-situ processing system is employed, as detailed in this article, to conduct feature identification and in-situ processing on the FPC. Several FPC samples were designed and fabricated, with the locations of the special-shaped holes to be cut marked by printed white patterns. Fig. 6 compares the alignment performance of the proposed method and a conventional vision-based approach in cutting these holes. As an example, a pentagram-shaped hole on one of these FPC samples is attempted, and the whole procedure of the proposed method is shown in Fig. 6(a).
A large-ranged (27 mm × 27 mm) coarse scan is first conducted at a low resolution (approximately only 250 μm) to swiftly overview, identify, and locate the pentagram pattern, taking only 2 s. After identifying the approximate location of the pattern, a fine scan near the pattern at a reduced range and high resolution is conducted to give exact information with a high-quality image, taking around 10 min. The angular coordinates of the pattern’s edge points are extracted to generate a continuous trajectory, which is then transmitted to the galvanometer scanner for laser processing, enabling precise in-situ laser cutting. The laser accurately scans and cuts along the pattern’s edges, with no offset, deformation, or rotation observed, as shown in Fig. 6(b). Given that the laser cutting line width is 30 μm, it can be deduced that the precision of the proposed method is below 15 μm.
As a comparison, another laser cutting experiment of the pentagram pattern is also conducted using the visual coaxial camera approach (Fig. 6(d)). The same prototyping system as described in Section 3.1 is employed for this experiment, except that the luminance sensor is replaced with an industrial camera module featuring a resolution of 3840 × 2160. With the camera's scale calibrated to be at 1.060 μm⋅pixel-1, the resulting image frame size in this method is calculated to be 4.1 mm × 2.3 mm, which is inadequate to encompass the entirety of the pentagram feature. Thus, much effort must be made to locate only part of the feature due to the limited imaging range. Subsequently, more images are captured at intervals equivalent to the frame size and merged into a larger image to cover the pattern. Although capturing a single image requires less than one second, the whole process, including initial pattern localization, image capture, merging and corrections, and image field calibrations, results in an overall time consumption of approximately 5 min. Stitching errors and mismatches are visibly evident in the composite image; thus, corrections are applied on the image to eliminate these imperfections and achieve a more optimal image. The laser scanning path is determined based on the corrected image, and the laser processing result is shown in Fig. 6(c), with distortions notably visible at the vertices and significant displacements of approximately 200 μm (Fig. 6(c)). A comparison of the proposed method and the visual-based method is given in Table 2.
More details on both methods can be accessed in Figs. S6 and S7 in Appendix A. Additional patterns, including cross-shaped holes (Fig. S8 in Appendix A) and triangle-shaped holes (Fig. S9 in Appendix A), were also tested. The experiments conducted on cutting holes of different shapes demonstrated the excellent flexibility and accuracy of the system, along with its proficiency for full-in-situ imaging and processing.
4.2. Defect detection of Micro-LED chips: Misaligned die removal in laser induced mass transfer
The Micro-LED technology has attracted significant attention owing to its promising applications across various domains. The potential for delivering high brightness, energy efficiency, and miniaturization positions Micro-LEDs as a promising candidate for next-generation display and lighting technologies [56]. To realize the full potential of Micro-LEDs, precise and efficient mass transfer techniques are required. Laser induced forward transfer (LIFT) has emerged as a prominent method for the mass transfer of Micro-LEDs [43]. LIFT utilizes laser beams to selectively ablate and remove the polymer sacrificial layer between the Micro-LED chips and the source substrate. Subsequently, the Micro-LED chips are released and transferred from the source substrate to the target substrate, facilitating rapid and precise placement (Figs. 7(a) and (b)).
Prior to the LIFT process, the chips should be bonded to the polymer as a well-organized array on the source substrate, as depicted in Fig. 7(c). However, it is common for some chips to have misalignment and deflection during the bonding procedure, commonly known as "misaligned dies." The presence of these misaligned dies lowers the transfer yield, therefore, the removal of these misaligned dies before starting the LIFT process is of crucial importance [57].
In this experiment, the proposed full-in-situ imaging and laser processing system is employed to achieve the identification and removal of the misaligned dies. The source comprises a quartz plate coated with a layer of spin-coated polymer. An array of Micro-LEDs is bonded to the polymer sacrificial layer, as shown in Figs. 7(c) and (d). Initially, the system conducted a large-ranged coarse scan to locate the position of the array on the quartz plate and provide an overview, as shown in Fig. 7(e). Notably, certain chips in the upper left corner of the array are suspected to be misaligned. Subsequently, a more detailed scan on the suspected chips was performed, as the result presented in Fig. 7(f). Two misaligned dies have been identified and confirmed, while a third die is suspected to be misaligned. To obtain a more precise view of these three dies, a further detailed scan is conducted with a higher resolution, confirming that all three dies are misaligned dies, as shown in Fig. 7(g).
To remove these misaligned dies, full-in-situ laser processing is employed to selectively ablate the polymer layer between the misaligned dies and the quartz plate. The entire area is laser-scanned line-by-line in the red solid rectangular box which covered the misaligned dies, as shown in Figs. 7(g) and (h). All the materials within the rectangular area are ablated, resulting in the successful removal of the misaligned dies, as demonstrated in Fig. 7(i). It is also confirmed by microscope observation that these 3 removed misaligned dies are the only 3 misaligned dies among the array of 115 chips in total. With no misaligned dies remaining, the substrate is prepared for subsequent mass transfer processes.
5. Conclusion and future work
The inherent non-physical-contact nature of lasers has posed significant challenges in terms of observation and alignment during laser processing. In this study, these challenges have been overcome by incorporating the principle of reversible optical paths into the proposed galvanometer-based laser processing system, achieving the integration of imaging observation and in-situ processing, as well as eliminating the alignment errors. The conclusions are listed as follows:
(1) A full-in-situ imaging and laser processing system based on galvanometers has been proposed, which eliminates the alignment errors and is well suited for small-batched, highly customized, and complex processing tasks. This system specifically facilitates in-line inspection, detection, and measurement during laser fabrications.
(2) The proposed system supports variable ranges and resolutions for stitchless imaging and processing, simultaneously achieving a maximum range of 27 mm × 27 mm and a minimum resolution of 0.412 µm.
(3) Two experimental cases, which includes FPC cutting and Micro-LEDs defect detection, have been carried out to further demonstrate the system’s exceptional imaging and processing capabilities, with a maximum processing error of <15 µm.
It is also worth noticing that the overall accuracy for laser processing is determined by different factors from various areas. The proposed method eliminated the alignment errors, including the coordinate mapping errors, calibration errors, and so on, while the errors from other aspects (e.g., galvanometer repeatability, laser spot size) still exist and gave an overall error of <15 µm. These errors can be further eliminated by upgradation in galvanometer motor performance and beam shaping techniques to enhance the ultimate accuracy.
With future improvements in sensor and focal length in the future, this system is likely to achieve an even higher SR, thereby further improving the image quality. The point-by-point imaging method in this system helped eliminate the alignment error, but also made sacrifices in the imaging efficiency. Higher efficiency can be realized by upgrading the high-bandwidth communication protocols.
Furthermore, attempts will be made focusing on the integration of full-color imaging, 3D surface contouring, as well as the implementation of artificial intelligence-based feature identification and extraction, aiming at further enhancing the system’s functionality and enabling it to tackle a wider range of tasks with increased efficiency and accuracy.
MoonHI, KimS, ByunJE, KimYW, JungYW, JoSG, et al. Facile and rapid fabrication of wearable biosensors via femtosecond laser-directed micro-patterning with large-sized reduced graphene oxide for physiological monitoring. Chem Eng J2024; 479:147819.
[2]
WangX, MeiX, YinH, WangZ, HeX, CuiJ. Direct writing nanostructures on Au nano-film by laser irradiating cantilevered scanning near-field optical microscope tip. Engineering2024; 49:104-12.
[3]
ChongTC, HongMH, ShiLP. Laser precision engineering: from microfabrication to nanoprocessing. Laser Photonics Rev2010; 4(1):123-43.
[4]
ZhangJ, YangQ, MaQ, RenF, ChengY, GouX, et al. Fish skin inspired scale armored sliding surfaces. Engineering2024; 49:113-21.
[5]
BrownMS, ArnoldCB. Fundamentals of laser-material interaction and application to multiscale surface modification. In: Sugioka K, Meunier M, Piqué A, editors. Laser precision microfabrication. Berlin: Springer; 2010. p. 91-120.
[6]
LiaoJ, ZhangD, LiZ. Conceptional pure-tungsten metasurfaces based on femtosecond laser nanomanufacturing. Engineering2024; 49:61-80.
[7]
KumagaiM, UchiyamaN, OhmuraE, SugiuraR, AtsumiK, FukumitsuK. Advanced dicing technology for semiconductor wafer—stealth dicing. IEEE Trans Semicond Manuf2007; 20(3):259-65.
CuiM, LuL, ZhangZ, GuanY. A laser scanner-stage synchronized system supporting the large-area precision polishing of additive-manufactured metallic surfaces. Engineering2021; 7(12):1732-40.
[10]
JiaH, SunH, WangH, WuY, WangH. Scanning strategy in selective laser melting (SLM): a review. Int J Adv Manuf Technol2021; 113(9-10):2413-35.
[11]
NagarajanB, HuZ, SongX, ZhaiW, WeiJ. Development of micro selective laser melting: the state of the art and future perspectives. Engineering2019; 5 (4):702-20.
[12]
SchützeD, MüllerV, HeinzelG. Precision absolute measurement and alignment of laser beam direction and position. Appl Opt2014; 53(28):6503-7.
[13]
LiL. The challenges ahead for laser macro, micro and nano manufacturing. In: Lawrence J, editor. Advances in Laser Materials Processing. Amsterdam: Elsevier; 2018. p. 23-42.
[14]
MosallanejadMH, NiroumandB, AversaA, SabooriA. In-situ alloying in laser-based additive manufacturing processes: a critical review. J Alloy Compd2021; 872:159567.
[15]
McCannR, ObeidiMA, HughesC, McCarthyÉ, EganDS, VijayaraghavanRK, et al. In-situ sensing, process monitoring and machine control in laser powder bed fusion: a review. Addit Manuf2021; 45:102058.
[16]
EvertonSK, HirschM, StravroulakisP, LeachRK, ClareAT. Review of in-situ process monitoring and in-situ metrology for metal additive manufacturing. Mater Des2016; 95:431-45.
[17]
WeiC, ZhangZ, ChengD, SunZ, ZhuM, LiL. An overview of laser-based multiple metallic material additive manufacturing: from macro- to micro-scales. Int J Extreme Manuf2021; 3(1):012003.
[18]
LiuY, BluntL, GaoF, JiangX. High-dynamic-range 3D measurement for E- beam fusion additive manufacturing based on SVM intelligent fringe projection system. Surf Topogr2021; 9(3):034002.
[19]
KumarS, TongZ, JiangX. Advances in the design and manufacturing of novel freeform optics. Int J Extreme Manuf2022; 4(3):032004.
[20]
TongZ, ZhongW, ZengW, JiangX. Closed-loop form error measurement and compensation for FTS freeform machining. CIRP Ann2021; 70(1):455-8.
[21]
ZhongW, TongZ, JiangX. Integration of on-machine surface measurement into fast tool servo machining. Procedia CIRP2021; 101:238-41.
[22]
YangS, LiC, Zhang G. Measurement technology for precision machines. In: Jiang Z, Yang S, editors. Precision machines. Singapore: Springer; 2020. p. 511-61.
[23]
NeumannB, DämonA, HogenkampD, BeckmannE, KollmannJ. A laser-autofocus for automatic microscopy and metrology. Sens Actuators1989; 17 (1-2):267-72.
[24]
ArmbrusterO, NaghilouA, PöhlH, KautekW. In-situ and non-destructive focus determination device for high-precision laser applications. J Opt2016; 18 (9):095401.
[25]
CaoBX, HoangPL, AhnS, KimJ, NohJ. High-precision detection of focal position on a curved surface for laser processing. Precis Eng2017; 50:204-10.
[26]
KeaveneyJ. Automated translating beam profiler for in situ laser beam spot-size and focal position measurements. Rev Sci Instrum2018; 89(3):035114.
[27]
DuX, FlorianC, ArnoldCB. Single-lens dynamic z-scanning for simultaneous in situ position detection and laser processing focus control. Light Sci Appl2023; 12(1):274.
[28]
YangS, YangL, ZhangG, WangT, YangX. Modeling and calibration of the galvanometric laser scanning three-dimensional measurement system. Nanomanuf Metrol2018; 1(3):180-92.
[29]
DengH, HuP, ZhangG, XiaC, CaiY, YangS. Accurate and flexible calibration method for a 3D microscopic structured light system with telecentric imaging and Scheimpflug projection. Opt Express2023; 31(2):3092-113.
[30]
WangG, FengC, HuX, YangH. Temporal matrices mapping-based calibration method for event-driven structured light systems. IEEE Sens J2021; 21 (2):1799-808.
[31]
LiX, LiuB, MeiX, WangW, WangX, LiX. Development of an in-situ laser machining system using a three-dimensional galvanometer scanner. Engineering2020; 6(1):68-76.
[32]
HanZ, ZhangL. Modeling and calibration of a galvanometer-camera imaging system. IEEE Trans Instrum Meas2022; 71:1-9.
[33]
VanlanduitS, GuillaumeP, CaubergheB, VerbovenP. An automatic position calibration method for the scanning laser Doppler vibrometer. Meas Sci Technol2003; 14(8):1469-76.
[34]
SelsS, BogaertsB, VanlanduitS, PenneR. Extrinsic calibration of a laser galvanometric setup and a range camera. Sensors2018; 18(5):1478.
[35]
GodineauK, LavernheS, TournierC. Calibration of galvanometric scan heads for additive manufacturing with machine assembly defects consideration. Addit Manuf2019; 26:250-7.
[36]
YuanQ, ZhuD, GaoZ. Alignment analyses of a galvanometer-based scanner in free-space Fourier domain optical coherence tomography. Appl Opt2015; 54 (32):9554-62.
[37]
HuangJ, QinQ, WangJ, FangH. Two dimensional laser galvanometer scanning technology for additive manufacturing. Int J Mater Mech Manuf2018; 6 (5):332-6.
[38]
XieJ, HuangS, DuanZ, ShiY, WenS. Correction of the image distortion for laser galvanometric scanning system. Opt Laser Technol2005; 37(4):305-11.
[39]
YeungH, LaneBM, DonmezMA, MoylanS. In-situ calibration of laser/galvo scanning system using dimensional reference artefacts. CIRP Ann2020; 69 (1):441-4.
[40]
Ortega DelgadoMA, LasagniAF. Reducing field distortion for galvanometer scanning system using a vision system. Opt Lasers Eng2016; 86:106-14.
[41]
WisselT, WagnerB, StüberP, SchweikardA, ErnstF. Data-driven learning for calibrating galvanometric laser scanners. IEEE Sens J2015; 15 (10):5709-17.
[42]
ZhangF, DuanJ, ZengX, LiX. 355nm DPSS UV laser micro-processing for the semiconductor and electronics industry. In:Proceedings of the Laser Applications in Microelectronic and Optoelectronic Manufacturing XV; 2010 Jan 23-28; San Francisco, CA, USA. Bellingham: SPIE; 2010. p. 75840Z.
[43]
ChenF, BianJ, HuJ, SunN, YangB, LingH, et al. Mass transfer techniques for large-scale and high-density microLED arrays. Int J Extreme Manuf2022; 4 (4):042005.
[44]
GongY, GongZ. Laser-based micro/nano-processing techniques for microscale LEDs and full-color displays. Adv Mater Technol2023; 8(5): 2200949.
[45]
HuangC, Ching-RoaV, LiuY, GiacomelliMG. High-speed mosaic imaging using scanner-synchronized stage position sampling. J Biomed Opt2022; 27 (1):016502.
[46]
DumaVF, TankamP, HuangJ, WonJ, RollandJP. Optimization of galvanometer scanning for optical coherence tomography. Appl Opt2015; 54(17):5495-507.
[47]
ChoTS, ParisS, HornBKP, FreemanWT. Blur kernel estimation using the radon transform. In:Proceedings of the CVPR 2011; 2011 Jun 20-25; Colorado Springs, CO, USA. Piscataway: IEEE; 2011. p. 241-8.
[48]
VarelaLG, BoucheronLE, SandovalS, VoelzD, SiddikAB. Estimation of motion blur kernel parameters using regression convolutional neural networks. J Electron Imaging2024; 33(2):023062.
[49]
HuangL, XiaY. Joint blur kernel estimation and CNN for blind image restoration. Neurocomputing2020; 396:324-45.
[50]
ChangJ, KauvarI, HuX, WetzsteinG. Variable aperture light field photography:overcoming the diffraction-limited spatio-angular resolution tradeoff. In:Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2016 Jun 27-30; Las Vegas, NV, USA. Piscataway: IEEE; 2016. p. 3737-45.
[51]
LeeJ, SungS. Evaluating spatial resolution for quality assurance of UAV images. Spat Inf Res2016; 24(2):141-54.
[52]
SandersJ, HurwitzL, SameiE. Patient-specific quantification of image quality: an automated method for measuring spatial resolution in clinical CT images. Med Phys2016; 43(10):5330-8.
[53]
ZhangJ, LiJ. Spatial cognitive engine technology. New York City: Academic Press; 2023.
PDF
(3561KB)
