Automatic alignment coupling method and curing equipment

Abstract

The invention provides an automatic alignment coupling method and curing equipment, and relates to the technical field of photoelectron, and the method comprises the steps: obtaining a multi-mode visual image representing a coupling interface of an optical fiber array end surface and a silicon optical chip, executing space registration based on the image, and generating registration data; controlling the multi-degree-of-freedom micro-displacement platform to adjust the pose of the optical fiber array until a preset alignment condition is met; in response to meeting the alignment condition, driving a crimping mechanism to perform crimping and acquiring online detection data in real time; the coupling quality is judged according to online detection data, if the requirement is met, ultraviolet light is emitted for curing to form a connector, otherwise, the optical fiber array is released, and registration is conducted again; according to the method, the precision and stability of automatic alignment are improved, and the production yield is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optoelectronic technology, in particular, to an automatic alignment coupling method and a curing device. BACKGROUND

[0002] Silicon optical chips have become an important research direction in the field of modern optoelectronics due to their advantages in high-speed data transmission, integrated density and cost control. However, high-precision automatic alignment coupling between silicon optical chips and fiber arrays is still one of the important process problems that need to be solved in current mass production. In the existing technology of automatic alignment coupling between silicon optical chips and fiber arrays, a single waveband visual imaging system is usually used to identify the pose of the fiber end face and the chip surface. However, this method can only obtain surface or shallow structure features during imaging, and it is difficult to effectively obtain multi-dimensional information of buried waveguides inside the chip and surface structures at the same time, thereby limiting the spatial pose registration accuracy and making it difficult to meet the increasing requirements of optical communication coupling precision.

[0003] In addition, the existing multi-degree-of-freedom displacement control platform usually faces problems such as mechanical coupling between platform axes, limited control instruction resolution, and error accumulation caused by environmental factors when performing high-precision adjustment tasks, which can easily lead to misjudgment or inaccurate convergence of the control loop, reducing the long-term stability and reliability in batch automatic production.

[0004] Therefore, the existing automatic coupling alignment technology of silicon optical chips needs to be further improved to solve the above technical problems, so as to improve the coupling alignment precision, reliability and production efficiency under batch production conditions. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an automatic alignment coupling method and a curing device.

[0006] In a first aspect, the present application provides an automatic alignment coupling method, comprising:

[0007] Obtaining a multi-modal visual image for characterizing the coupling interface between the end face of the fiber array and the silicon optical chip;

[0008] Based on the multi-modal visual image, performing spatial registration to generate first registration data;

[0009] According to the first registration data, controlling a multi-degree-of-freedom micro-displacement platform to adjust the pose of the fiber array until a pose difference value is less than a first preset threshold, and outputting an alignment completion signal;

[0010] In response to receiving the alignment completion signal, a crimping mechanism is driven to crimp the fiber array to a coupling interface of the silicon optical chip at a preset crimping speed, and online detection data in a crimped state is acquired, the online detection data including an alignment error;

[0011] In response to the alignment error being less than a second preset threshold, a UV curing module is synchronously controlled to emit UV light to the coupling interface to cure the pre-positioned optical glue and form a cured connector, and a coupling completion instruction is output.

[0012] In response to the alignment error being greater than or equal to the second preset threshold, the crimping mechanism is controlled to release the fiber array and return to perform the spatial registration.

[0013] As an optional implementation, the generating first registration data includes:

[0014] Obtaining layout information of a metal interconnection layer of the silicon optical chip, and generating a mask corresponding to the metal interconnection layer;

[0015] Edge filtering is performed on the multi-modal visual image using the mask to remove edge information located in a metal interconnection layer region;

[0016] Based on the multi-modal visual image after edge filtering, a coupling reference feature point is extracted, and a coordinate transformation relationship between an end face of the fiber array and the coupling interface of the silicon optical chip is solved to generate the first registration data.

[0017] As an optional implementation, the removing edge information located in the metal interconnection layer region includes:

[0018] According to a near-infrared channel of the multi-modal visual image, a first edge set is extracted;

[0019] According to a visible light channel of the multi-modal visual image, a second edge set is extracted;

[0020] Consistency matching is performed on the first edge set and the second edge set to generate a common edge set;

[0021] Edge pixels not belonging to the common edge set are deleted within a metal interconnection layer region defined by the mask, and a multi-modal visual image after edge filtering is output.

[0022] As an optional implementation, the generating the first registration data includes:

[0023] At a first acquisition time, a first reference feature point set is extracted from the multi-modal visual image, and an initial coordinate transformation relationship is calculated;

[0024] At a second acquisition time point separated from the first acquisition time point by a preset synchronization delay, a second reference feature point set is extracted again from the multi-modal visual image, and a global tilt offset is calculated;

[0025] The initial coordinate transformation relationship and the global tilt offset are fused to generate a time-series compensated coordinate transformation relationship, and the first registration data is output.

[0026] As an optional implementation, the calculation of the global tilt offset comprises:

[0027] Two end feature points with a channel sequence number of 1 and a channel sequence number of M are determined in sequence according to the channel sequence number, where M is the total number of channels of the optical fiber array;

[0028] Coordinate data of the two end feature points in a preset three-dimensional Cartesian coordinate system is obtained, and a least square straight line fitting algorithm is used to calculate a first pitch angle of the optical fiber array relative to the X-Z plane based on the coordinate data;

[0029] In the second reference feature point set, two corresponding end feature points with a channel sequence number of 1 and a channel sequence number of M are determined with the same channel sequence number, coordinate data of the corresponding end feature points in the three-dimensional Cartesian coordinate system is obtained, and a least square straight line fitting algorithm is used to calculate a second pitch angle of the optical fiber array relative to the X-Z plane;

[0030] An absolute difference value between the first pitch angle and the second pitch angle is calculated, and the absolute difference value is set as the global tilt offset.

[0031] As an optional implementation, before the initial coordinate transformation relationship and the global tilt offset are fused, the method further comprises:

[0032] At least two temperature sensors are arranged at opposite corner positions of a silicon optical chip base, and real-time temperature data output by the temperature sensors is periodically collected;

[0033] The real-time temperature data is input into a preset thermal expansion model to calculate a thermal expansion prediction vector corresponding to the geometric center of the chip, wherein the thermal expansion model is established based on a linear expansion coefficient of a chip packaging material and a chip-clamp assembly size;

[0034] The thermal expansion prediction vector, a translation component of the initial coordinate transformation relationship, and a rotation component of the global tilt offset are vector superimposed to generate a heat-compensated coordinate transformation relationship.

[0035] As an optional implementation, the output of the alignment completion signal comprises:

[0036] inputting the first registration data, generating a first pose adjustment instruction according to a preset gain coefficient, and sending the first pose adjustment instruction to the multi-degree-of-freedom micro-displacement platform;

[0037] monitoring feedback displacements of each driving shaft of the multi-degree-of-freedom micro-displacement platform in real time, and if a coupling displacement amplitude between adjacent driving shafts is detected to exceed a preset inter-shaft coupling threshold, reducing the gain coefficient and generating a second pose adjustment instruction;

[0038] After executing the second pose adjustment instruction, the pose difference is monitored again, and if the pose difference falls within a preset threshold window and the pose difference of all channels satisfies a channel consistency determination condition, the alignment completion signal is output.

[0039] As an optional implementation, the output of the alignment completion signal further includes:

[0040] After the pose difference falls within the preset threshold window, a periodic micro-amplitude jitter instruction with an amplitude not greater than a minimum quantization step of the capacitive displacement sensor is injected to the multi-degree-of-freedom micro-displacement platform, and a mean value of feedback displacements of the capacitive displacement sensor is calculated in a continuous N sampling period.

[0041] In response to the mean value calculation result still remaining within the preset threshold window, the alignment completion signal is maintained; and in response to the mean value calculation result exceeding the preset threshold window, the multi-degree-of-freedom micro-displacement platform is iteratively adjusted according to the gain coefficient of the second pose adjustment instruction.

[0042] As an optional implementation, the consistency matching performed on the first edge set and the second edge set to generate a common edge set includes:

[0043] Based on a periodic channel arrangement of the optical fiber array in the V-groove clamp, V-groove edge reference marks corresponding to a plurality of adjacent channels are extracted;

[0044] Using the reference marks and pre-stored inter-channel spacing information, local distortion center offsets of image regions of each channel are determined;

[0045] For each channel region, a corresponding local distortion correction is applied respectively to map pixel coordinates of the first edge set and the second edge set to a unified correction coordinate system;

[0046] In the unified correction coordinate system, the first edge set and the second edge set are matched one by one according to a spatial nearest neighbor or a feature descriptor similarity determination to generate the common edge set.

[0047] In a second aspect, the present application provides a curing device, comprising:

[0048] a collection module, configured to acquire a multi-modal visual image for characterizing a coupling interface between an end face of a fiber array and a silicon optical chip;

[0049] a registration module, configured to perform spatial registration based on the multi-modal visual image to generate first registration data;

[0050] a first control module, configured to control a multi-degree-of-freedom micro-displacement platform to adjust a pose of the fiber array according to the first registration data until a pose difference value is less than a first preset threshold, and output an alignment completion signal;

[0051] a second control module, configured to drive a crimping mechanism to crimp the fiber array to the coupling interface of the silicon optical chip at a preset crimping speed and acquire online detection data in a crimped state in response to receiving the alignment completion signal, the online detection data including an alignment error;

[0052] a detection module, configured to control a ultraviolet curing module to emit ultraviolet light to the coupling interface to cure a pre-positioned optical glue and form a cured connector in response to the alignment error being less than a second preset threshold, and output a coupling completion instruction; and control the crimping mechanism to release the fiber array and return to perform the spatial registration in response to the alignment error being greater than or equal to the second preset threshold.

[0053] Compared with the prior art, the automatic alignment coupling method of the present application can effectively overcome the problem of insufficient traditional single-band imaging information by acquiring a multi-modal visual image and performing spatial registration based on multi-spectral information fusion, thereby helping to improve the accuracy and reliability of alignment between a silicon optical chip and a fiber array. In addition, the present application also utilizes the closed-loop feedback control mechanism of the multi-degree-of-freedom micro-displacement platform to ensure timely convergence of the pose adjustment process, reduce error accumulation and the adverse effects of inter-axis mechanical coupling, and effectively improve the precision and stability of the alignment process. Further, the present application uses a real-time online detection method during the crimping process, which can dynamically judge the coupling quality and timely discover and correct the alignment deviation, avoiding the scrapping problem caused by irreversible curing after crimping, thereby effectively reducing the production risk and improving the yield and process stability of batch production. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 a flowchart of the automatic alignment coupling method provided by the present application;

[0055] Figure 2 a flowchart of a method for generating first registration data provided by the present application;

[0056] Figure 3 a flowchart of a method for removing edge information located in the metal interconnection layer region provided by the present application;

[0057] Figure 4 A schematic diagram of a curing device provided by the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application.

[0059] Referring to Figure 1 FIG. 1 shows a flowchart of an automatic alignment coupling method provided by the present application, and the method comprises steps S101-S105, wherein:

[0060] S101: acquiring a multimodal vision image for characterizing a coupling interface between an end face of a fiber array and a silicon optical chip;

[0061] S102: performing spatial registration based on the multimodal vision image to generate first registration data;

[0062] S103: controlling a multi-degree-of-freedom micro-displacement platform to adjust a pose of the fiber array according to the first registration data until a pose difference value is detected to be less than a first preset threshold, and outputting an alignment completion signal;

[0063] S104: in response to receiving the alignment completion signal, driving a crimping mechanism to crimp the fiber array to the coupling interface of the silicon optical chip at a preset crimping speed, and acquiring online detection data in a crimping state, the online detection data comprising an alignment error;

[0064] S105: in response to the alignment error being less than a second preset threshold, synchronously controlling a ultraviolet curing module to emit ultraviolet light to the coupling interface to cure a pre-positioned optical glue and form a cured connector, and outputting a coupling completion instruction; in response to the alignment error being greater than or equal to the second preset threshold, controlling the crimping mechanism to release the fiber array and return to perform the spatial registration.

[0065] The system for performing the method described in the present application can comprise a silicon optical chip (Silicon Photonics, SiP), a fiber array (Fiber Array, FA), a multi-degree-of-freedom micro-displacement platform, a crimping mechanism, a multimodal vision imaging system, a ultraviolet curing module, and a central controller.

[0066] Exemplarily, the multimodal vision imaging system is usually located above the coupling interface for acquiring alignment images; the fiber array can be fixed on the multi-degree-of-freedom micro-displacement platform, which is usually used to drive the fiber array to realize precise alignment relative to the silicon optical chip; the crimping mechanism and the ultraviolet curing module can be coaxial or off-axis arranged for performing crimping and curing operations after alignment is completed.

[0067] The multi-degree-of-freedom micro-displacement platform can be a piezoelectric ceramic driven six-degree-of-freedom parallel mechanism (Hexapod Platform), and the displacement resolution can reach 1 nanometer, for example; the multi-modal visual imaging system can include coaxially arranged visible light cameras and near-infrared cameras, and share the light path through a beam splitter; the compression mechanism can exemplarily use a voice coil motor, and the precision can generally reach a sub-micron level; the ultraviolet curing module can use an ultraviolet light emitting diode (UV-LED) light source or a laser, and the typical wavelength can be 365 nm.

[0068] For the above S101:

[0069] The multi-modal visual image refers to a set of complementary images of the same coupling interface acquired synchronously under different spectral channels, wherein the visible light channel highlights the metal marks on the surface of the silicon optical chip and the geometric profile of the fiber end face, and the near-infrared channel can penetrate the silicon substrate to present the position of the waveguide inside the chip and the fiber core, avoiding the limitation that only one can be taken and the rest is abandoned when using traditional single-band imaging.

[0070] In a specific implementation, the silicon material significantly reflects light intensity in the visible light region, and the contrast between the surface metal structure and the substrate is extremely high; in the near-infrared region with a wavelength greater than about 1100 nm, the absorption coefficient of silicon to light is significantly reduced, and the imaging light can penetrate a silicon substrate hundreds of microns thick, so that the embedded waveguide and the profile of the butt joint fiber core can be presented. By spatially synchronizing and gray-scale registering the two-channel images, the algorithm can simultaneously locate the surface feature points and the internal optical center line in a unified coordinate system, thereby establishing a complete "chip-fiber" six-degree-of-freedom relative pose model.

[0071] Exemplarily, the silicon optical chip is first fixed on a vacuum suction base, and the fiber array is installed on a piezoelectric driven six-degree-of-freedom platform. The multi-modal imaging device is arranged directly above the interface, and distributes the reflected light to the two cameras through the coaxial beam splitter prism. The visible light channel uses a ring-shaped white light LED to provide diffuse reflection illumination, and the exposure time and aperture coefficient can be automatically adjusted according to the brightness on site to ensure that the gray scale of the metal mark is not overexposed; the near-infrared channel uses a 1310 nm line light source for coaxial illumination, and the exposure time is automatically searched to a region with a relatively high signal-to-noise ratio according to the thickness of the silicon substrate and the waveguide reflection intensity. The system can synchronously collect two images at one time, and then map to a unified three-dimensional reference system by a calibration matrix and output a feature point coordinate file, providing input for the next registration algorithm.

[0072] In addition, if the production line needs to adjust parameters due to changes in chip specifications, light source power or ambient light, the pre-set automatic exposure-automatic gain routine can quickly complete the recalibration without changing the imaging hardware architecture.

[0073] In this way, the acquired multi-modal visual images provide a reference benchmark containing both surface and internal information for subsequent spatial registration, which helps to reduce the alignment drift caused by the lack of single-band features and improves the robustness of pose solving. At the same time, synchronous acquisition avoids mechanical errors introduced by multiple movements of the workpiece, which positively contributes to improving the overall coupling accuracy and shortening the work station beat.

[0074] For the above S102:

[0075] After the acquisition of multi-modal visual images is completed, the system enters the spatial registration phase. Spatial registration, in this application, refers to mapping the surface feature points and internal optical feature points extracted from the visible light and near-infrared channels to the same three-dimensional reference system, and further calculating the relative pose relationship of the fiber array end face and the silicon optical chip coupling interface. The pose relationship is output in the form of rotation and translation, denoted as the first registration data.

[0076] The principle can be understood as conformal matching of two images with different information dimensions: the visible light channel provides a planar positioning reference, and the near-infrared channel provides a depth or refractive layer positioning reference. After conversion by the calibration matrix, the two images are overlapped in the calculation space, and the six-degree-of-freedom relative pose is obtained.

[0077] In specific implementation, the registration process can be divided into two levels from coarse to fine. In the coarse registration phase, the system first identifies the larger metal cross mark and the fiber end face ring in the visible light image, and extracts the brightness peak distribution center of the waveguide region in the near-infrared image. At this time, a fast matching strategy based on cross-correlation or normalized gray difference can be used to preliminarily align the chip and fiber array reference coordinates in the camera field of view, reducing the pose difference between the two to the micron level, providing a convergent initial value for subsequent high-precision algorithms.

[0078] In the fine registration phase, sub-pixel level center of gravity fitting is performed on each matched feature based on the coarse registration result, and the rotation and translation are optimized simultaneously by the least squares criterion. The feature point coordinates can be solved by rigid body transformation and iterative convergence until the reprojection error of all features is less than one-tenth of the visual resolution.

[0079] It is worth emphasizing that the above iteration number is not a fixed value, but is dynamically adjusted by the algorithm in real time according to the error convergence speed. When the image signal-to-noise ratio is low or the sample has local reflection, the iteration termination condition can be automatically relaxed and the registration result can be cross-validated to prevent falling into local optimum.

[0080] Exemplarily, first, a pre-calibrated dual-target homography is loaded to project both channels of the image onto a unified camera coordinate system; then, a feature detector based on differential Gaussian and fast linear classifier is run on each channel separately to filter out key points such as metal markers, waveguide bright lines, and fiber cores; next, a nearest neighbor search is performed on cross-channel features and false matches are removed by a random consistency check; finally, the remaining features are fed into a rigid pose solver based on gradient descent to output a three-dimensional rotation vector and a translation vector of the fiber array relative to the silicon photonic chip, and the confidence of each feature point is recorded as a quality assessment indicator. To balance processing speed and stability, the system can use an industrial control CPU or FPGA that supports parallel SIMD instruction sets on hardware to pipeline accelerate feature extraction and matching. If a high-reflectivity or low-contrast scene is encountered, the controller will automatically switch to a multi-scale template strategy to replace point feature matching with template matching, ensuring that the registration task is completed smoothly.

[0081] The generated first registration data not only retains the overall pose information but also contains feature confidence and re-projection residuals, providing quantifiable adjustment basis for the closed-loop control stage. Using this data, a multi-degree-of-freedom micro-displacement platform can place the coupling interface at the center of the capture window within a single closed loop, avoiding multiple backtracking operations due to error accumulation; at the same time, the residual distribution of the registration output can also be used as a process health indicator in post-analysis, which helps to predict optical path contamination or device defects.

[0082] In summary, through the above spatial registration steps, not only the accuracy and robustness of pose solving are improved, but also a repeatable and high-quality alignment foundation is laid for the subsequent crimping and curing processes.

[0083] For the above S103:

[0084] After obtaining the first registration data, the rotation vector and translation vector given therein are used to drive the multi-degree-of-freedom micro-displacement platform to complete fine pose adjustment in real time. The so-called pose difference is the residual error between the fiber array reference coordinate system and the target pose after registration in six degrees of freedom, which considers both translation and rotation.

[0085] The platform control follows the idea of "coarse entry-fine sweep-convergent locking": initially, the platform moves quickly according to the full compensation value given by the registration, compressing the residual error to sub-micron level; then it enters the micro-sweep stage, injecting a small amplitude sinusoidal jitter on each degree of freedom, evaluating the error gradient by real-time monitoring of the optical coupling power and capacitance displacement sensor feedback, and continuously iterating to reduce the jitter amplitude; when the pose difference of all degrees of freedom falls within the pre-set first threshold window, the central controller determines that the alignment is complete and issues a locking instruction.

[0086] For example, the six-degree-of-freedom platform is driven by piezoelectric ceramics, and the minimum step size of a single axis is preferably controlled to the order of tens of nanometers. The central controller reads the registration residual and sensor feedback every 2 ms, and runs parallel PID loops for three translational and three rotational degrees of freedom; for the possible inter-axis coupling, the proportional coefficient can be dynamically adjusted during operation to control the coupling displacement of adjacent drive axes within the resolution step of the platform.

[0087] Further, in order to avoid the platform from locking when approaching the target pose, the system temporarily reduces the integral gain when the error converges near the upper threshold, and simultaneously opens the power maximization judgment. If the optical power curve remains stable at the peak value for a plurality of consecutive sampling periods, the position is directly confirmed as the best coupling point.

[0088] It can be understood that the first preset threshold is not a fixed constant, but is set comprehensively with reference to the chip end face roughness, fiber array angle error budget and final coupling loss target. For example, on a common 4x4 channel array, the threshold can be set to an in-plane translation of no more than 100 nm, an out-of-plane height difference of no more than 200 nm, and a pitch and yaw angle of less than 0.01°; in higher density arrays or more sensitive coupling loss scenarios, the same algorithm can be reused by simply adjusting the threshold. At the end of the process, the central controller writes the locking instruction and the current pose record into the log for subsequent crimping and curing steps.

[0089] The closed-loop adjustment strategy thus designed can balance between the maximum stroke and the minimum step size of the platform, avoiding both long-time fine-grained search and overshooting due to one-time large compensation; compared with traditional pure power scanning or single-camera closed-loop methods, it uses spatial registration residual as an "absolute" reference and power variation as a "relative" tuning, achieving fast determination and adaptive convergence of pose error, which is helpful for improving coupling success rate and shortening equipment cycle.

[0090] For the above S104:

[0091] After detecting the alignment completion signal, the central controller triggers the crimping mechanism to start uniform speed pressing. The crimping mechanism can use a stroke programmable voice coil motor, and a small range force sensor is coaxially installed. The platform continuously outputs the force displacement curve during the pressing process; the pressing speed can be adaptively set according to the chip passivation layer thickness and the optical adhesive viscosity, generally in the range of several microns per second to tens of microns per second, and gradually increased to ensure that the optical adhesive spreads sufficiently without excessive overflow. At the same time, the system continuously obtains online detection data to determine whether there is new pose drift or coupling failure during the pressing process.

[0092] For example, the controller injects a stable power of probe light from one end of the fiber array, and a multi-channel photodiode array is configured at the output end of the chip to collect the coupling power curve in real time. During crimping, the optical power changes with the axial displacement in a typical "bell-shaped" manner; the controller performs sliding average and derivation on the power curve, and when the first derivative is stably close to zero within a preset window and the power value enters and remains within the peak fluctuation range, the current position is recorded as the coupling peak point, and the corresponding residual pose difference is defined as the alignment error. If the error is always lower than the second threshold, the system is considered to be successfully crimped.

[0093] In addition, in order to cover the scenarios of possible light path obstruction or power drift, the system can also enable a lateral visual monitoring channel in parallel. The lateral camera observes the chip and the array sidewall through a long working distance microscope lens, and a pair of high-contrast reflective markers are attached to the outer edges of the two; during crimping, the software tracks the relative distance and angle changes of these markers in real time, and maps the results as the out-of-plane tilt and height change. If there is a significant deviation between the lateral measurement and the optical power evaluation, the controller will temporarily stop pressing, trigger the platform fine-tuning, and then resume the uniform crimping. The two measurement links are independent of each other and can cross-verify each other, making the system have strong fault tolerance.

[0094] Within the normal process window, the force-displacement curve will eventually appear a platform state with rapidly rising force value and high light power, at which the central controller determines that the alignment error has met the set index, immediately locks the voice coil motor position and enters the ultraviolet curing process. If a sharp drop in optical power or lateral pose is observed at any stage, the controller releases the voice coil motor back to a safe height and calls the registration-fine-tuning subroutine to re-establish coupling. Through this online detection and dynamic rollback mechanism, the device can continuously obtain stable crimping quality without relying on the subjective judgment of the operator, and significantly reduce the yield loss caused by instantaneous drift.

[0095] For the above S105:

[0096] The second preset threshold in this embodiment is used to determine whether the coupling quality after crimping is sufficient to enter the qualified window of irreversible curing stage, which belongs to the same evaluation system as the first preset threshold but has different levels of action. The first preset threshold focuses on the geometric residual error in the platform closed-loop adjustment process, and focuses on whether it has entered the operable range; while the second preset threshold integrates the geometric error, coupling optical power and crimping force change, and focuses on whether it has met the final delivery requirements, which is wide in front and strict in back, ensuring both adjustment efficiency and product quality.

[0097] In a specific implementation, the central controller continuously collects the light power, displacement and force value curves during the crimping process, and when the real-time calculated alignment error falls within the second threshold window, it is determined that the coupling state reaches the curing condition. The threshold itself is not a fixed constant, and can be dynamically set according to the flatness of the chip end face, the flowability of the optical glue, and the target insertion loss budget. For example, in a four-channel array scenario, the in-plane residual displacement can be limited to tens of nanometers, and the coupling power fluctuation can be maintained within a small range; if the array density or optical power requirement is higher, the same control logic can be reused by tightening the window accordingly.

[0098] In a specific implementation, after the crimping enters a steady state, the system first locks the multi-degree-of-freedom platform position and maintains the constant crimping force output of the voice coil motor. The ultraviolet curing module is then triggered by the central controller, and the ultraviolet light beam irradiates the coupling interface along the array-chip coaxial path, with the irradiation intensity and duration automatically matched according to the optical glue curing curve. The light power and displacement curves are still recorded in real time during the curing process; if the curves remain stable, the curing process automatically outputs a coupling completion instruction after a preset time length, and submits the process log to the upper database. If the error rises to the upper limit of the second threshold at any time, the controller immediately turns off the ultraviolet light, releases the crimping force, separates the fiber array and the chip, and returns to the spatial registration sub-process to re-establish the alignment.

[0099] By introducing a more stringent second threshold before curing and adopting a real-time rollback mechanism, the system can ensure efficiency while avoiding irreversible fixing in unqualified poses, which helps to improve the yield and stability of mass production, and reduce the cost and downtime risk caused by repair or scrap.

[0100] As an optional implementation, refer to Figure 2 A flowchart of a method for generating first registration data provided by the present application includes steps S201-S203, wherein:

[0101] S201: Obtain the layout information of the metal interconnection layer of the silicon optical chip, and generate a mask corresponding to the metal interconnection layer;

[0102] S202: Use the mask to perform edge filtering on the multi-modal visual image, and remove the edge information located in the metal interconnection layer region;

[0103] S203: Based on the multi-modal visual image after edge filtering, extract the coupling reference feature points, and solve the coordinate transformation relationship between the fiber array end face and the coupling interface of the silicon optical chip, to generate the first registration data.

[0104] In practical applications, for silicon photonic chips with complex metal interconnection structures, the images collected under the visible light channel often contain a large number of regular and strong highlight edges. These edges mainly come from the reflection signals of the metal wiring layer, have good structural continuity and contrast, and are easily misidentified as effective features by traditional edge detection algorithms, thereby causing subsequent feature point mismatch, coordinate fitting drift, or unstable registration results. Especially in the case of weak waveguide, fiber array geometry and low signal-to-noise ratio, this interference may cause the registration result to deviate completely from the coupling center, thereby affecting the stability of the entire coupling link.

[0105] To reduce the problem of false feature extraction caused by such metal interference, the application introduces a layout mask strategy, that is, a shielding information consistent with the spatial distribution of the chip metal interconnection layer is introduced in the visual processing link to filter the edge response of the corresponding region in the image and eliminate potential interference from the source. The layout mask refers to the metal wiring layer derived from the chip process design file (GDSII or OASIS format), which is generated by spatial mapping and rasterization processing to generate a binary mask image corresponding to the image field. In the mask image, the part marked as the shielding region will not participate in the subsequent image feature extraction process.

[0106] The principle is based on the spatial mapping between the layout and the image coordinates. In specific implementation, first, in the system calibration stage, a two-dimensional affine transformation model between the chip design coordinate system and the visual image coordinate system is established through three standard Fiducial markers (three-point method) on the silicon photonic chip. Taking the "Metal3" and "Metal4" two metal layers in the chip GDS file as an example, the polygon boundaries can be exported as a vector graph by a layout processing tool such as KLayout, and the spatial transformation can be completed by cooperating with the above affine matrix, and then the cv2.fillPoly() function in the OpenCV library is used to rasterize it into a bitmap mask. The bitmap resolution should be consistent with the actual captured image, such as 2048x2048, so as to align pixel by pixel.

[0107] In the visual image processing stage, the mask will be used as a weight filtering matrix before edge detection to participate in the image convolution process. Taking Canny edge detection as an example, before calculating the image gradient, the mask is used to perform pixel-by-pixel multiplication operation, and the gray value of the shielding region in the mask is forcibly set to zero, or the gradient response of this region is set to be infinitesimal. Then the Sobel operator is used to calculate the image gradient, and then the double-threshold connection algorithm is used to generate the edge map. In this way, even if the metal interconnection layer reflects strongly, its edge will no longer participate in the feature point extraction and matching process in the visual image.

[0108] In a complete implementation process, after the system loads the mask image, it first performs mask shading on the multi-modal image, and then runs the pre-defined edge detection and feature extraction algorithm in the visible light and near-infrared images respectively. In this embodiment, a feature point detection algorithm based on LoG (Laplacian of Gaussian) weighted gradient can be used, and the implementation thereof includes:

[0109] The image is first filtered by a Gaussian filter to remove noise, and then a second-order Laplacian operator is used to extract edge centers to obtain sub-pixel level feature points. The feature point set then enters the matching module for spatial position registration to calculate the transformation relationship between the fiber array and the chip, and output the first registration data.

[0110] In this way, the interference of metal edges on chips with dense metal interconnection structures can be effectively avoided, and the stability of feature extraction can be improved, especially in coupling structures with limited or unevenly distributed edges. In addition, since the mask strategy is logically decoupled from the image processing module and does not depend on the specific image content, it can be applied to a variety of chip layouts or products with large changes in metal layers, which helps to reduce algorithm parameter adjustment costs, shorten registration time, and improve the overall environmental adaptability and yield performance of the system.

[0111] As an optional implementation, refer to Figure 3 A flowchart of a method for removing edge information located in the metal interconnection layer region provided by the present application includes steps S301-S304, wherein:

[0112] S301: Extract a first edge set according to the near-infrared channel of the multi-modal visual image;

[0113] S302: Extract a second edge set according to the visible light channel of the multi-modal visual image;

[0114] S303: Perform consistency matching on the first edge set and the second edge set to generate a common edge set;

[0115] S304: Delete edge pixels not belonging to the common edge set within the mask-defined metal interconnection layer region, and output the multi-modal visual image after edge filtering.

[0116] In actual production environment, the layout design of the metal interconnection layer of the silicon optical chip often has slight deviation from the actual manufacturing process, and the pure static mask scheme can only make theoretical shielding according to the layout data, and cannot adapt to the problems such as the metal layer offset caused by the layout distortion, illumination change or shooting angle change in the actual image. Therefore, only the edge shielding by the mask may miss some metal edges, or even mistakenly delete the real coupling edges. In order to further improve the accuracy of edge filtering, the application introduces a dynamic edge filtering method based on consistency matching between multi-modal image channels.

[0117] The consistency matching in the application specifically refers to that after the edge sets of the near-infrared and visible light channels are independently extracted from the multi-modal visual images of the two channels respectively, the real edges which are consistent in the two channels and have the same spatial position are screened out by comparing the position, length, contour shape and direction of the edges between the two channels, that is, the common edge set, and the edges which only appear in a single channel are considered as false edges and are removed.

[0118] The principle of the method is that the metal interconnection layer of the silicon optical chip usually only reflects obviously in the visible light band, and transmits or reflects weakly in the near-infrared band; and the waveguide structure or fiber end face which is really used for optical coupling alignment can form clear and position-stable edge features in the visible light and near-infrared bands. Therefore, by cross-channel consistency matching, the difference between the image and the layout in the actual production process can be dynamically adapted, and the robustness of edge filtering is enhanced.

[0119] In specific implementation, first, multi-modal visual images of the coupling interface are synchronously collected, specifically including one visible light channel image and one near-infrared channel image, and it is ensured that the two channel images have completed pixel-level registration. Then, edge detection is independently performed on each channel image to generate the corresponding edge set. In the embodiment, the exemplary edge detection algorithm is Canny algorithm, and the specific implementation can use the cv2.Canny() function in the OpenCV function library, and the typical parameter settings are, for example, low threshold 50, high threshold 150, and the edge connection pixel length is greater than 5 pixels to be reserved to filter out noise and isolated points. In this way, the first edge set of the near-infrared channel and the second edge set of the visible light channel are obtained.

[0120] Then, consistency matching is performed on the first edge set and the second edge set to determine the common edge set which has the same spatial position between the two sets.

[0121] For example, step one: the center point coordinates, length, circumscribed rectangle size and main direction angle of each edge curve in the visible light channel are calculated;

[0122] Step two: For each edge curve in the near-infrared channel, perform the same feature calculation as step one.

[0123] Step three: Match the edge curve features of the visible light channel one by one with those of the near-infrared channel. The specific matching criteria include but are not limited to: the Euclidean distance between the centers of the two curves is less than a preset threshold, such as 5 pixels; the length ratio is between 0.8 and 1.2; the intersection over union (IoU) of the circumscribed rectangle exceeds 0.6; the difference in the main direction angle is less than 10 degrees.

[0124] Step four: For edge pairs that meet the above matching criteria, further call a shape similarity comparison function for accurate confirmation, such as cv2.matchShapes(), and return a value less than 0.2 to consider the shapes similar. The confirmed edge pairs are included in the common edge set.

[0125] Subsequently, the system performs edge screening on the initial edge images of the visible light and near-infrared channels based on the generated common edge set, i.e., deleting pixels within the metal interconnection layer region defined by the mask that do not belong to the common edge set.

[0126] For example, first, load the mask image generated by the layout data to clearly define the spatial region for edge screening; then, check each edge pixel in the mask region one by one. If the edge pixel does not belong to the common edge set, delete it from the original image using median filtering or directly set it to zero. Using median filtering as an example, the cv2.medianBlur() function of the OpenCV library can be used for filtering; if the zero method is used, the image gray value at the edge pixel position is directly assigned to the background gray value.

[0127] For example, a certain horizontal edge of about 50 pixels is detected in the visible light image of a chip, while no obvious edge or only a length of 10 pixels is detected in the corresponding position of the near-infrared image. After the above matching criteria, the two edges cannot meet the consistency condition, so it can be determined that the edge belongs to the pseudo-edge caused by the metal layer in the visible light channel and should be deleted.

[0128] On the other hand, if a certain edge appears in both channels with a position deviation of no more than 3 pixels, a length ratio of about 0.95, and a shape matching score less than 0.1, it is determined to belong to a real coupled structure edge and should be retained in the common edge set.

[0129] Ultimately, the image after consistency filtering contains fewer pseudo-feature edges and can more reliably extract true feature points for subsequent coupling reference. Based on this image, the system further performs feature extraction and spatial registration algorithms to generate first registration data after dynamic edge filtering. Feature extraction can include sub-pixel center positioning and edge contour fitting, for example, while the spatial registration algorithm can be a rigid body transformation fitting algorithm based on minimum mean square error.

[0130] In this way, through dynamic cross-channel consistency matching, pseudo-edge features in a single channel can be effectively identified and eliminated, compensating for the problems of missed filtering or accidental deletion caused by the deviation between the layout and the actual image of the static mask. In production scenarios with different chip batches and large fluctuations in layout accuracy, it can adaptively improve the stability of image processing and the reliability of spatial alignment, and ultimately help improve the overall accuracy of the coupling between the chip and the fiber array and the production line yield.

[0131] As an optional implementation, generating the first registration data includes:

[0132] At a first acquisition moment, extracting a first reference feature point set from the multimodal visual image and calculating an initial coordinate transformation relationship;

[0133] At a second acquisition moment separated from the first acquisition moment by a preset synchronization delay, extracting a second reference feature point set from the multimodal visual image again, and calculating a global tilt offset;

[0134] The initial coordinate transformation relationship and the global tilt offset are fused to generate a time-compensated coordinate transformation relationship, and the first registration data is output.

[0135] In actual production processes, the spatial alignment accuracy of silicon photonic chips and fiber arrays is often susceptible to environmental vibrations, temperature drift, or transient jitter in the production line. As a result, the coordinate transformation relationship obtained at a specific acquisition moment may only be applicable to the state at that moment, but cannot accurately represent the continuous coupling posture change trend throughout the entire crimping process. To further reduce the impact of such temporal disturbances on the coupling process, this application proposes a coordinate transformation fusion method based on timing compensation.

[0136] Specifically, the timing compensation described in this application involves extracting a set of reference feature points at two adjacent acquisition moments separated by a preset delay, analyzing the displacement changes between the feature points, and inferring the global tilt offset trend caused by environmental disturbances, thereby correcting the coordinate transformation relationship in real time. This method dynamically predicts and compensates for timing-related errors, ensuring that the generated first registration data more accurately represents the continuous true alignment state, rather than being limited to a single static moment.

[0137] In implementation, first, at the first acquisition moment, the system synchronously acquires the multi-modal visual image of the coupling interface, and performs feature extraction algorithm on the image to obtain a stable first reference feature point set.

[0138] For example, sub-pixel feature detection can be performed by LoG (Laplacian of Gaussian) operator, and the cross metal mark on the chip surface, the fiber array end face ring feature, and the brightness center feature point of the internal waveguide are clearly marked. Subsequently, the system uses these feature point coordinates to solve the initial coordinate transformation relationship between the chip and the fiber array, which is usually realized by using a rigid transformation fitting algorithm based on the least square method, and outputs the initial coordinate transformation matrix.

[0139] At the second acquisition moment separated from the first acquisition moment by a preset synchronization delay, for example, a delay of 50 milliseconds to 200 milliseconds to match the typical environmental disturbance period, the system acquires the multi-modal visual image of the coupling interface again, and re-executes the same feature extraction process to generate a second reference feature point set. At this time, in order to ensure the reliability of the timing compensation, the image at the second acquisition moment should use the same optical system, camera settings and exposure parameters as the first moment.

[0140] Subsequently, the system calculates the global tilt offset of the coupling interface in the delay interval based on the difference between the two reference feature point sets.

[0141] Exemplarily, the system can use the following method for accurate calculation:

[0142] Step one: establish a three-dimensional space coordinate list for the feature point sets obtained at the two moments respectively, and use ICP (Iterative Closest Point) algorithm for matching. In specific implementation, the rigid registration of coordinate point clouds can be performed by using the pcl::IterativeClosestPoint function in the open source point cloud library PCL to extract the overall rotation and translation change trend caused by environmental disturbance.

[0143] Step two: in the execution of the ICP algorithm, the rigid transformation matrix between the two point sets is solved by iteration until the residual sum of squares reaches the convergence condition, for example, the convergence accuracy is At this time, the output rigid transformation matrix is the global tilt offset in the timing interval. In order to ensure the stability of the operation, the system can set the maximum number of iterations to 100, and perform outlier filtering in each iteration, such as the filtering method based on mean and standard deviation, to improve the fitting robustness.

[0144] Step three: the offset matrix determined by the ICP algorithm is specifically a rotation matrix and a translation vector in a three-dimensional space, and the displacement and rotation trend expressed by the offset matrix is the global drift caused by the environment, which can be used to predict the coupling pose change trend in the subsequent short time.

[0145] Next, the system fuses the initial coordinate transformation relationship and the global tilt offset to generate a time-compensated coordinate transformation relationship, and the specific implementation process is as follows:

[0146] Firstly, the pose represented by the initial coordinate transformation matrix is defined as the base state, and then the offset matrix output by the ICP algorithm is used to compensate and fuse the base state, for example, matrix multiplication is performed to stack, so as to predict the coupling pose trend in the future short time. The system can further perform interpolation or filtering processing on the fused matrix, for example, exponential smoothing filtering or Kalman filtering, to avoid excessive amplification of instantaneous errors and smooth the output of the predicted future pose state.

[0147] For example, at the first acquisition time, the initial rotation angle of the chip and the optical fiber array is detected to be 0.005°, and the translation deviation is 80 nanometers; at the second time, the ICP matching indicates that the global rotation change after the delay time is an additional 0.002°, and the translation change is 10 nanometers, and then the time-compensated coordinate transformation relationship after fusion is rotation of 0.007° and translation of 90 nanometers. The fused data is used to drive the pre-compensation action of the multi-degree-of-freedom micro-displacement platform, which can significantly improve the initial alignment efficiency and success rate in the subsequent coupling stage.

[0148] In this way, the time-compensated coordinate transformation relationship is output as the first registration data in the embodiment. Compared with the static measurement at a single time, the fusion method provides a more robust prediction for environmental disturbances, which helps to prevent the platform from deviating from the instantaneous alignment caused by device vibration or thermal drift in advance, greatly reduces the accumulation of dynamic errors caused by environmental factors, thereby improving the overall coupling stability, precision of the production line, and shortening the closed-loop time of single alignment, and improving the overall yield during batch production.

[0149] As an optional implementation, the calculation of the global tilt offset includes:

[0150] The two end feature points with a channel number of 1 and a channel number of M are determined in sequence according to the channel number, where M is the total number of channels of the optical fiber array;

[0151] The coordinate data of the two end feature points in the preset three-dimensional Cartesian coordinate system is obtained, and the first pitch angle of the optical fiber array relative to the X-Z plane is calculated based on the coordinate data using a least squares straight line fitting algorithm.

[0152] In the second reference feature point set, two corresponding end feature points with channel number 1 and channel number M are determined with the same channel number, the coordinate data of the corresponding end feature points in the three-dimensional Cartesian coordinate system is obtained, and the second pitch angle of the fiber array relative to the X-Z plane is calculated by using the least square straight line fitting algorithm.

[0153] The absolute difference between the first pitch angle and the second pitch angle is calculated, and the absolute difference is set as the global tilt offset.

[0154] The present application proposes a more efficient, faster and more targeted implementation details for the foregoing timing compensation method. Although the complete six degrees of freedom (6-DOF) calculation based on the iterative closest point algorithm can comprehensively and accurately obtain the rotation and translation amount, in actual production scenarios, the change of the pitch angle usually has the greatest influence on the coupling performance between the silicon optical chip and the fiber array, and is the main error source causing coupling loss. In addition, the ICP algorithm needs to perform complex iterative calculation on all feature points, which consumes a lot of computing resources, and it may be difficult to achieve real-time response on a high-bpm production line. Therefore, the present embodiment proposes a fast calculation method specially for the pitch angle component, to preferentially and accurately compensate for the change of the pitch angle.

[0155] The specific implementation steps are as follows:

[0156] First, at the first collection time, the feature points at both ends of the fiber array are accurately extracted by using multi-modal visual images, specifically the end feature points with channel number 1 (leftmost) and channel number M (rightmost). The LoG or Sobel operator with sub-pixel level accuracy is recommended for feature point extraction to ensure the coordinate extraction accuracy.

[0157] Secondly, the obtained two-dimensional image coordinates of the end feature points are mapped to the preset three-dimensional Cartesian coordinate system through the camera calibration model to determine the three-dimensional space coordinate data thereof. Subsequently, the least square straight line fitting algorithm is used to calculate the space straight line passing through the two end feature points, and the initial pitch angle of the straight line relative to the X-Z plane of the device coordinate system is calculated.

[0158] Thirdly, at the second collection time separated from the first collection time by a preset synchronization delay, the end feature points with channel number 1 and channel number M are extracted again in the same way, and the second pitch angle is calculated by using the least square straight line fitting algorithm. The typical setting of the synchronization delay is 50 milliseconds to 200 milliseconds.

[0159] Next, by comparing the difference between the first pitch angle and the second pitch angle, the definite pitch angle change value is obtained. This value directly reflects the overall tilt change of the fiber array caused by environmental disturbance or device vibration during the delay period.

[0160] Finally, the system fuses the calculated pitch angle change value as a key component in the overall timing compensation process. Specifically, the calculated pitch angle change value is converted into a rotation transformation matrix around a specific axis (e.g., the X-axis), and then fused with the initial coordinate transformation matrix through matrix fusion, thereby efficiently and quickly compensating for the most critical pitch angle error.

[0161] In this way, by accurately calculating the pitch angle change of the fiber array, a lightweight and efficient implementation is provided, which effectively reduces the computational load and can be used as a quick replacement or priority supplement to the complete ICP algorithm. It is particularly suitable for practical application scenarios that are extremely sensitive to pitch angle changes and require high production cycles, thereby improving overall production efficiency and coupling accuracy and reducing the risk of coupling errors due to environmental disturbances.

[0162] As an optional implementation, before fusing the initial coordinate transformation relationship and the global tilt offset, the method further includes:

[0163] At least two temperature sensors are arranged at opposite corner positions of the silicon optical chip base, and real-time temperature data output by the temperature sensors is periodically collected;

[0164] The real-time temperature data is input into a preset thermal expansion model to calculate a thermal expansion prediction vector corresponding to the geometric center of the chip, wherein the thermal expansion model is established based on the linear expansion coefficient of the chip packaging material and the chip-clamp assembly size;

[0165] The thermal expansion prediction vector is vector superimposed with the translation component of the initial coordinate transformation relationship and the rotation component of the global tilt offset to generate a heat-compensated coordinate transformation relationship.

[0166] Further, the present application introduces a mechanism for monitoring the temperature of the chip base and predicting thermal expansion to effectively reduce coupling errors caused by changes in environmental temperature. In actual production environments, the difference in thermal expansion between silicon optical chips and clamps often causes small displacements in the position of the chip, especially under continuous production conditions. This thermal effect can gradually accumulate into an error source that cannot be ignored. Therefore, the present application provides a specific implementation method for thermal expansion prediction and compensation to further improve the alignment and coupling accuracy.

[0167] In specific implementation, first, at least two high-sensitivity temperature sensors are installed at the diagonal positions of the silicon optical chip base. The temperature sensor is recommended to be of the thermistor or thermocouple type to ensure the real-time and accuracy of temperature data collection. The specific installation position can be the upper left corner and the lower right corner of the chip base. The system periodically collects the real-time temperature data output by the temperature sensors at a certain sampling period, such as once per second.

[0168] Secondly, the system processes the collected temperature data in real time according to the pre-established thermal expansion model, and accurately calculates the thermal expansion prediction vector of the geometric center of the chip. Specifically, the thermal expansion model is a mathematical model determined in advance based on the linear expansion coefficients of the chip and the fixture materials, and the actual chip-fixture assembly size. Through the model, the thermal expansion displacement trend of the chip in the X, Y and Z directions can be calculated according to the temperature change of each measuring point of the base.

[0169] Exemplarily, the construction steps of the thermal expansion model include: calibrating the linear expansion coefficients of the chip and fixture materials in advance, and obtaining the specific material property values through experiments or manufacturer-provided data. Subsequently, a spatial displacement function relationship is constructed according to the geometric size of the chip and fixture, that is, the predicted displacement vector of the chip center point is directly derived from the temperature difference.

[0170] Thirdly, after obtaining the thermal expansion prediction vector, the system further fuses it with the initial coordinate transformation relationship and the global tilt offset calculated previously, and the specific fusion process includes two independent and explicit mathematical steps:

[0171] Step one: compensate the translation component. Perform vector addition between the thermal expansion prediction vector and the translation component of the initial coordinate transformation relationship to compensate for the translation drift of the physical position of the chip caused by thermal effects.

[0172] Step two: compensate the rotation component. Perform matrix multiplication between the rotation component represented by the rotation matrix or the matrix generated by the specific rotation angle of the global tilt offset and the rotation component of the initial coordinate transformation relationship to compensate for the rotation drift of the device posture.

[0173] Exemplarily, assuming that the translation component of the initial coordinate transformation relationship is 10.0 nanometers (X axis), 20.0 nanometers (Y axis) and 5000.0 nanometers (Z axis), and the rotation component is represented by the rotation matrix . If the real-time temperature monitoring shows that the temperature of the chip base increases by 2°C, according to the linear expansion coefficients of the fixture materials and the size parameters of the chip, it is calculated that the chip center generates a thermal expansion prediction displacement vector of about 100 nanometers upward in the Z axis direction. At the same time, the time sequence compensation module calculates the global tilt offset represented by the rotation matrix .

[0174] Subsequently, the system performs the specific fusion compensation process:

[0175] The new translation component is (10.0, 20.0, 5000.0) + (0.0, 0.0, 100.0) = (10.0, 20.0, 5100.0) nanometers.

[0176] The new rotation component is .

[0177] The translation vector and the rotation matrix obtained through the above operation jointly constitute a final coordinate transformation relationship after thermal compensation and timing compensation, which is used to control the platform to realize accurate pre-compensation action.

[0178] In this way, by introducing temperature monitoring and thermal expansion prediction model, the spatial displacement error caused by environmental temperature change is effectively compensated. Compared with the traditional method which only relies on static or short-time scale prediction, the present application provides a dynamic and real-time thermal compensation strategy, which significantly improves the alignment accuracy and stability under long-term operation conditions, and is especially suitable for the automatic coupling production line of silicon optical chips and fiber array which are sensitive to thermal effects and have high precision requirements.

[0179] As an optional implementation, the output alignment completion signal includes:

[0180] Taking the first registration data as input, a first pose adjustment instruction is generated according to a preset gain coefficient, and the first pose adjustment instruction is sent to the multi-degree-of-freedom micro-displacement platform.

[0181] The feedback displacement of each driving shaft of the multi-degree-of-freedom micro-displacement platform is monitored in real time, and if the coupling displacement amplitude between adjacent driving shafts is detected to exceed a preset inter-axis coupling threshold, the gain coefficient is reduced and a second pose adjustment instruction is generated.

[0182] After executing the second pose adjustment instruction, the pose difference value is monitored again, and if the pose difference value falls within a preset threshold window and the pose difference values of all channels satisfy the channel consistency determination condition, the alignment completion signal is output.

[0183] Further, the present application proposes a multi-degree-of-freedom micro-displacement platform control strategy to effectively avoid the reduction of alignment accuracy or unstable adjustment caused by excessive coupling displacement between platform driving shafts.

[0184] In specific implementation, first, the system generates an initial pose adjustment instruction based on the first registration data obtained previously using a closed-loop PID control algorithm. The gain coefficient specifically represents the proportional gain, integral gain and differential gain parameters of the PID controller, which are initially set to be high to ensure fast response and drive the fiber array to the target position for preliminary adjustment.

[0185] Secondly, during the execution of the initial pose adjustment instruction by the multi-degree-of-freedom micro-displacement platform, the system monitors the feedback displacement of each driving shaft of the platform in real time. The inter-axis coupling displacement specifically refers to: when the system sends a displacement instruction to a certain driving shaft, due to the structural characteristics of the platform, an unintended displacement in the direction of other non-target shafts is generated. The system quantitatively calculates the inter-axis coupling ratio by comparing the difference between the instruction displacement and the actual feedback displacement. For example, when a displacement instruction of a certain amplitude is sent to the X-axis, if an unintended displacement change occurs in the Y-axis and the amplitude of the displacement change exceeds a preset threshold, such as 5%, it is determined that the inter-axis coupling is out of limit.

[0186] Subsequently, once it is detected that the inter-axis coupling is out of limit, the system reduces the gain coefficient of the currently used PID controller, for example, by a fixed percentage, such as 20%. Then, the pose adjustment instruction is regenerated according to the updated gain parameter, and is sent to the multi-degree-of-freedom micro-displacement platform again to achieve more stable pose adjustment.

[0187] After the multi-degree-of-freedom micro-displacement platform executes the pose instruction with gain adjustment, the system again monitors the pose difference of the platform in real time, i.e., the residual translation and rotation error between the fiber array and the target alignment position. In addition, the system also monitors the optical coupling power loss values of all channels at the same time, and determines the consistency of the channels by calculating the range of these loss values, i.e., the difference between the maximum loss value and the minimum loss value. For example, when the range is less than a preset power threshold (such as 0.5 dB), it is determined that the error distribution between the channels is uniform and no obvious abnormality occurs.

[0188] After the pose difference successfully falls within the preset threshold window and at the same time meets the channel consistency condition, the central controller formally outputs an alignment completion signal, locks the current pose state, and provides it for subsequent crimping and curing links.

[0189] Through the above multi-axis displacement monitoring and gain self-adaptive adjustment mechanism, the application effectively avoids the mutual interference between the driving shafts of the platform caused by excessively high gain, improves the stability and accuracy of automatic alignment adjustment, and is especially suitable for high-precision, low-fault-tolerant silicon optical chip and fiber array automatic coupling application scenarios.

[0190] As an optional implementation, the output alignment completion signal further includes:

[0191] After the pose difference falls within the preset threshold window, a periodic micro-amplitude jitter instruction with an amplitude not greater than the minimum quantization step of the capacitive displacement sensor is injected to the multi-degree-of-freedom micro-displacement platform, and the feedback displacement of the capacitive displacement sensor is averaged over N consecutive sampling periods.

[0192] In response to the mean value calculation result still remaining within the preset threshold window, the alignment completion signal is maintained; and in response to the mean value calculation result exceeding the preset threshold window, the gain coefficient of the second pose adjustment instruction is iteratively adjusted to adjust the multi-degree-of-freedom micro-displacement platform.

[0193] Further, the application proposes an accurate and effective closed-loop stability verification strategy to ensure that the actual coupled position after the completion of the pose adjustment can be stably maintained within the expected target range.

[0194] In a specific implementation, when the pose difference of the multi-degree-of-freedom micro-displacement platform successfully falls within the preset threshold window, the system continuously injects a set of micro-amplitude periodic jitter instructions into the platform. The amplitude of this set of jitter instructions is very small, generally set to be slightly lower than the minimum quantization step of the capacitive displacement sensor, and a specific typical value can be about 0.8 nanometers.

[0195] Although the single jitter amplitude is smaller than the single measurement accuracy of the sensor, through continuous sampling of multiple cycles, the system can utilize the inherent random measurement noise of the sensor to achieve higher effective resolution to accurately detect the micro response of the platform under the jitter excitation, and further verify the real stability of the pose.

[0196] Secondly, the system collects the displacement data fed back by the capacitive displacement sensor on each driving shaft within a plurality of sampling periods, for example, N sampling periods, and a typical value is 10 sampling periods. These displacement data are jointly affected by measurement noise and micro response, and through mean value calculation on the data within multiple cycles, the system can effectively improve the signal-to-noise ratio of the measurement result and realize resolution beyond.

[0197] The above-mentioned technology of utilizing sensor noise to realize resolution improvement is essentially oversampling and averaging. The principle of this method is that the random noise generated by the sensor in each measurement tends to cancel out in the averaging process of multiple samplings, while the regular response signal generated by the actual micro motion of the platform can be revealed through mean value calculation, thereby realizing accurate measurement of micro displacement below the quantization step of single measurement.

[0198] The implementation purpose of the micro-amplitude periodic jitter of the platform is to actively overcome the problems of possible static friction, controller integral saturation, and sensor quantization dead zone, etc. These factors may cause the platform to appear stable but have hidden instability. Once the platform is disturbed by external disturbances in actual operation, the hidden instability will cause the pose to jump unexpectedly. By applying periodic jitter instructions, the system can actively stimulate the micro response of the platform, and expose and eliminate these potential stability hazards in advance.

[0199] Exemplarily, the system continuously collects feedback displacement data for 10 cycles with a sampling period of 2 milliseconds after injecting a dithering instruction with an amplitude of 0.8 nanometers and a frequency of several to tens of hertz to the platform, and performs mean value calculation on the data.

[0200] Subsequently, the system evaluates the above mean value calculation result. If the calculation result is stably maintained within a preset pose difference threshold window, it indicates that the platform has good closed-loop stability and no significant pose drift occurs, and the system maintains the output alignment completion signal.

[0201] On the contrary, if the mean value calculation result exceeds the threshold window, it indicates that there is a non-negligible drift or stability problem in the actual pose of the platform. In this case, the system cancels the current alignment completion signal and re-enables the gain coefficient corresponding to the second pose adjustment instruction for iterative adjustment until the platform pose is stable again.

[0202] For example, when the mean value calculated displacement exceeds the allowed range, such as more than 5 nanometers, the system immediately performs fine adjustment of the pose again to ensure that the platform is finally stabilized at the expected target position.

[0203] In this way, by means of explicit super-resolution mean value calculation principle and clear dithering excitation purpose, the application provides a fully disclosed closed-loop stability verification mechanism, effectively improving the long-term stability of the automatic alignment coupling system, and is especially suitable for silicon optical chip and optical fiber array automatic coupling application scenarios with strict requirements for long-term stability.

[0204] As an optional implementation, the performing consistency matching on the first edge set and the second edge set to generate a common edge set comprises:

[0205] Based on the periodic channel arrangement of the optical fiber array in the V-groove clamp, extracting V-groove edge reference marks corresponding to a plurality of adjacent channels;

[0206] Using the reference marks and pre-stored channel spacing information, determining the local distortion center offset of the image region where each channel is located;

[0207] For each channel region, respectively applying the corresponding local distortion correction to map the pixel coordinates of the first edge set and the second edge set to a unified correction coordinate system;

[0208] In the unified correction coordinate system, the first edge set and the second edge set are matched one by one to generate the common edge set according to the spatial nearest neighbor or feature descriptor similarity judgment.

[0209] Further, the application also provides a multi-channel consistency matching method based on local distortion correction, to ensure that the edge information of visible light and near-infrared channel images can be accurately matched, thereby improving the accuracy of the final registration data.

[0210] In a specific implementation, considering that the fiber array presents a periodic channel arrangement structure in the V-groove fixture, the system extracts V-groove edge reference marks corresponding to a plurality of adjacent channels from the collected multi-modal visual images. The V-groove edge reference marks specifically refer to the slot edge feature lines of the V-groove, such as the left and right edge lines of each slot. The system can accurately detect these slot edge lines by using a Difference of Gaussians (DoG) or a Sobel operator, so as to accurately determine the position and periodic structure of the V-groove.

[0211] Secondly, the system measures the spacing between adjacent V-grooves in the actual image by using the extracted V-groove reference marks. Specifically, the system accurately measures the actual interval distance between each slot edge line in the image coordinate system, and compares it with the ideal channel spacing pre-stored in the chip design one by one. The difference between the two is the spacing deviation, which represents the degree of local imaging distortion.

[0212] Subsequently, the system calculates the local distortion center offset of each channel region according to the measured spacing deviation.

[0213] For example, the specific calculation method is as follows:

[0214] First, for the spacing deviation values of a plurality of continuous measurement points, a linear or quadratic polynomial fitting model is used to derive the distortion offset trend of the entire channel region; then, the distortion center point in each channel image region is accurately located according to the trend function, which is used as the reference point of the correction transformation.

[0215] Next, the system establishes a local distortion correction model for each channel region based on the calculated distortion center offset. Common models include local affine transformation or quadratic polynomial distortion correction. In specific implementation, the system takes the distortion center offset as the reference, and solves the affine transformation parameters between the matching point pairs in the local image region by using the least square method, or directly determines the polynomial coefficients by substituting the offset into the pre-set polynomial model, to accurately compensate for the local image geometric distortion. The corrected image coordinates are uniformly mapped into the same correction coordinate system through the transformation model.

[0216] Finally, in the unified correction coordinate system, the system performs accurate consistency matching on the corrected first edge set and the second edge set. The matching method can be selected as a spatial nearest neighbor algorithm or a feature descriptor similarity judgment to ensure that the edge points from the two channels correspond one by one. For example, the system sets a spatial distance threshold of 3 pixels. When the distance between a pair of edge points from different channels is less than the threshold, it is determined that the edge point pair is matched, and is included in the common edge set.

[0217] Through the above local distortion correction and accurate matching steps, the system finally realizes accurate multi-channel consistency matching, effectively eliminating the image mismatching problem caused by local distortion. This method is particularly suitable for automatic alignment and coupling applications of high-density and high-precision silicon optical chips and fiber arrays.

[0218] Exemplarily, after extracting a plurality of V-groove edge reference marks in succession and completing the pitch deviation measurement, the system derives the distortion center distribution trend of each channel region through a second-order polynomial fitting, and then establishes a local distortion correction model for each channel. After these correction models are applied, the original local geometric distortion existing in the two-channel images is greatly weakened, and the edge coincidence degree of each channel region is significantly improved. Through matching verification in the unified coordinate system, it can be seen that the corresponding edges of the visible light and near-infrared channels remain closely consistent in space, laying a stable foundation for the construction of the common edge set and providing more reliable reference information for subsequent pose adjustment.

[0219] Based on the same inventive concept, the embodiments of the present application also provide a solidification device corresponding to the automatic alignment and coupling method. Since the principle of solving the problem in the device of the embodiments of the present application is similar to the above-mentioned automatic alignment and coupling method of the embodiments of the present application, the implementation of the device can be referred to the implementation of the method, and the repeated parts will not be described again.

[0220] Referring to Figure 4 FIG. 1 shows a schematic diagram of a solidification device provided by the present application, which comprises:

[0221] The acquisition module 10 acquires a multi-modal visual image for representing the coupling interface between the fiber array end face and the silicon optical chip;

[0222] The registration module 20 performs spatial registration based on the multi-modal visual image to generate first registration data;

[0223] The first control module 30 controls the multi-degree-of-freedom micro-displacement platform to adjust the pose of the fiber array according to the first registration data until the detected pose difference value is less than a first preset threshold, and outputs an alignment completion signal;

[0224] The second control module 40 drives the crimping mechanism to crimp the fiber array to the coupling interface of the silicon optical chip at a preset crimping speed in response to receiving the alignment completion signal, and acquires online detection data in the crimping state, the online detection data including an alignment error;

[0225] The detection module 50 controls the ultraviolet curing module to emit ultraviolet light to the coupling interface to cure the preset optical glue and form a cured connector in response to the alignment error being less than a second preset threshold, and outputs a coupling completion instruction; controls the crimping mechanism to release the fiber array and returns to perform the spatial registration in response to the alignment error being greater than or equal to the second preset threshold.

[0226] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed in the present application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

Claims

1. An automatic alignment coupling method, characterized in that: include: Acquire multimodal visual images for characterizing the coupling interface between the fiber array end face and the silicon photonic chip; Based on the multimodal visual image, perform spatial registration to generate first registration data; Controlling the multi-degree-of-freedom micro-displacement platform to adjust the posture of the optical fiber array according to the first registration data until a posture difference is detected to be less than a first preset threshold, and outputting an alignment completion signal; In response to receiving the alignment completion signal, driving a crimping mechanism to crimp the optical fiber array to the coupling interface of the silicon photonic chip at a preset crimping speed, and acquiring online detection data in the crimping state, the online detection data including alignment error; In response to the alignment error being less than a second preset threshold, the UV curing module is synchronously controlled to emit UV light to the coupling interface to cure the preset optical glue and form a cured connector, and a coupling completion instruction is output; in response to the alignment error being greater than or equal to the second preset threshold, the crimping mechanism is controlled to release the optical fiber array and return to perform the spatial alignment.

2. The automatic alignment coupling method according to claim 1, characterized in that: Generating the first registration data includes: Acquire layout information of a metal interconnection layer of a silicon photonic chip and generate a mask corresponding to the metal interconnection layer; Performing edge filtering on the multimodal visual image using the mask to remove edge information located in the metal interconnection layer region; Based on the edge-filtered multimodal visual image, coupling reference feature points are extracted, and the coordinate transformation relationship between the optical fiber array end face and the coupling interface of the silicon photonic chip is solved to generate the first registration data.

3. The automatic alignment coupling method according to claim 2, characterized in that: The removing of edge information in the metal interconnection layer region comprises: extracting a first edge set according to a near-infrared channel of the multimodal visual image; extracting a second edge set according to a visible light channel of the multimodal visual image; Performing consistency matching on the first edge set and the second edge set to generate a common edge set; Edge pixels that do not belong to the common edge set are deleted in the metal interconnection layer area defined by the mask, and a multimodal visual image after edge filtering is output.

4. The automatic alignment coupling method according to claim 2, characterized in that: Generating the first registration data includes: At a first acquisition moment, extracting a first reference feature point set from the multimodal visual image and calculating an initial coordinate transformation relationship; At a second acquisition moment separated from the first acquisition moment by a preset synchronization delay, extracting a second reference feature point set from the multimodal visual image again, and calculating a global tilt offset; The initial coordinate transformation relationship and the global tilt offset are fused to generate a time-compensated coordinate transformation relationship, and the first registration data is output.

5. The automatic alignment coupling method according to claim 4, characterized in that: Calculating the global tilt offset includes: Determine two end feature points with channel numbers 1 and M in sequence according to the channel numbers, where M is the total number of channels in the optical fiber array; Obtaining coordinate data of the two end feature points in a preset three-dimensional Cartesian coordinate system, and calculating a first pitch angle of the optical fiber array relative to the XZ plane using a least squares straight line fitting algorithm based on the coordinate data; In the second reference feature point set, two corresponding end feature points with channel numbers 1 and M are determined with the same channel number, coordinate data of the corresponding end feature points in the three-dimensional Cartesian coordinate system are obtained, and a least squares straight line fitting algorithm is used to calculate a second pitch angle of the optical fiber array relative to the XZ plane; An absolute difference between the first pitch angle and the second pitch angle is calculated, and the absolute difference is set as a global tilt offset.

6. The automatic alignment coupling method according to claim 4 or 5, characterized in that: Before fusing the initial coordinate transformation relationship with the global tilt offset, the method further includes: At least two temperature sensors are arranged at opposite corners of the silicon photonic chip base, and real-time temperature data output by the temperature sensors are periodically collected; Inputting the real-time temperature data into a preset thermal expansion model to calculate a thermal expansion prediction vector corresponding to the chip geometric center, wherein the thermal expansion model is established based on the linear expansion coefficient of the chip packaging material and the chip-fixture assembly dimensions; The thermal expansion prediction vector is vector-superimposed with the translation component of the initial coordinate transformation relationship and the rotation component of the global tilt offset to generate a thermally compensated coordinate transformation relationship.

7. The automatic alignment coupling method according to claim 6, characterized in that: The output alignment completion signal includes: Taking the first registration data as input, generating a first posture adjustment instruction according to a preset gain coefficient, and sending the first posture adjustment instruction to the multi-degree-of-freedom micro-displacement platform; monitoring the feedback displacement of each drive axis of the multi-degree-of-freedom micro-displacement platform in real time, and if it is detected that the coupling displacement amplitude between adjacent drive axes exceeds a preset inter-axis coupling threshold, reducing the gain coefficient and generating a second posture adjustment instruction; After executing the second posture adjustment instruction, the posture difference is monitored again; if the posture difference falls into a preset threshold window and the posture differences of all channels meet the channel consistency judgment condition, the alignment completion signal is output.

8. The automatic alignment coupling method according to claim 7, characterized in that: The output alignment completion signal further includes: After the posture difference falls within the preset threshold window, injecting a periodic micro-dithering instruction with an amplitude no greater than a minimum quantization step of the capacitive displacement sensor into the multi-degree-of-freedom micro-displacement platform, and performing an average calculation on the feedback displacement of the capacitive displacement sensor over N consecutive sampling periods; In response to the mean calculation result still falling within the preset threshold window, the alignment completion signal is maintained; in response to the mean calculation result exceeding the preset threshold window, the iterative adjustment of the multi-degree-of-freedom micro-displacement platform according to the gain coefficient of the second posture adjustment instruction is resumed.

9. The automatic alignment coupling method according to claim 3, characterized in that: The performing consistency matching on the first edge set and the second edge set to generate a common edge set includes: Extracting V-groove edge reference marks corresponding to a plurality of adjacent channels based on the periodic channel arrangement of the optical fiber array in the V-groove fixture; Determining the local distortion center offset of the image region where each channel is located using the reference marker and pre-stored channel spacing information; For each channel region, applying corresponding local distortion correction respectively, and mapping the pixel coordinates of the first edge set and the second edge set to a unified correction coordinate system; In the unified rectified coordinate system, the first edge set and the second edge set are matched one-to-one according to spatial nearest neighbor or feature descriptor similarity determination to generate the common edge set.

10. A curing device, characterized in that: include: An acquisition module, which acquires multimodal visual images used to characterize the coupling interface between the fiber array end face and the silicon photonic chip; a registration module, performing spatial registration based on the multimodal visual image to generate first registration data; a first control module, controlling a multi-degree-of-freedom micro-displacement platform to adjust the posture of the optical fiber array according to the first registration data until a posture difference is detected to be less than a first preset threshold, and outputting an alignment completion signal; a second control module, in response to receiving the alignment completion signal, driving a crimping mechanism to crimp the optical fiber array to the coupling interface of the silicon photonic chip at a preset crimping speed, and acquiring online detection data in a crimped state, the online detection data including an alignment error; The detection module, in response to the alignment error being less than a second preset threshold, synchronously controls the UV curing module to emit UV light to the coupling interface to cure the preset optical glue and form a cured connector, and outputs a coupling completion instruction; in response to the alignment error being greater than or equal to the second preset threshold, controls the crimping mechanism to release the optical fiber array and return to perform the spatial alignment.

Images