Automatic method for universal wafer needle alignment identification
By constructing a standard gradient field and a deep learning model to automatically identify the tip position, the automation problem of pin alignment on wafer testing equipment was solved, realizing an efficient and accurate automatic pin alignment process and improving the automation level of wafer testing equipment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI GLORYSOFT CO LTD
- Filing Date
- 2024-11-21
- Publication Date
- 2026-05-26
AI Technical Summary
Existing wafer testing equipment relies on manual operation for pin alignment, resulting in low automation levels, cumbersome and complex operation, and a tendency for mismatches, which affects production efficiency and accuracy.
A standard gradient field is constructed using a gradient diffusion algorithm. Combined with a deep learning key point detection model, the needle tip position is automatically identified, and the optimal needle alignment position is found through multiple verifications, thus achieving full-process automation.
It significantly improves the automation level and accuracy of the needle matching process, reduces human interference, enhances work efficiency and accuracy, and reduces mismatch phenomena.
Smart Images
Figure CN122089628A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor manufacturing technology, and more specifically to an automated method for general-purpose wafer pin identification. Background Technology
[0002] Wafer testing machines are widely used in semiconductor chip manufacturing plants and packaging and testing facilities. Currently, the processes of recipe loading, batch number entry, and pin alignment (pin card position calibration) on existing wafer testing machines are mostly performed manually, resulting in low automation levels and a pressing need to improve work efficiency.
[0003] There are two existing needle alignment modes. One is manual alignment, where the operator uses the alignment software interface to switch between different lighting conditions and focus positions until a clear needle tip image appears on the screen. The operator visually identifies the arrangement of each needle tip and the overall shape of the needle holder; simultaneously, the operator identifies the crosshair pattern arranged in a fixed shape on the alignment screen. The operator visually matches the fixed crosshair pattern and the needle tip array, and after identifying a set that is considered the best match, manually clicks on the needle tip position to move the alignment position onto the needle tip. After a second verification, the alignment task is completed. This needle alignment method is cumbersome, inefficient, and disruptive to overall automated processes, interfering with production control. Under certain conditions, manual alignment can fail due to operator error or fatigue. Changing products or needle cards requires manual re-entry into the interface and re-marking the white needle tip position on the machine screen, which is complex. For multiple duplicate needle cards, alignment may be misaligned, which is difficult for operators to observe and prone to mismatches. The operation relies excessively on operator experience, hindering the development of a self-sustaining, efficient knowledge base within the factory.
[0004] The second method is semi-automatic needle alignment, which most needle alignment machine software offers. Before semi-automatic alignment, personnel need to observe and adjust the screen, adjusting the brightness and focus of different light sources until the needle tip is clearly visible. Then, the operator clicks the automatic alignment process, and the automatic alignment algorithm identifies the optimal alignment position based on the target area in the pre-configured needle card file and adjusts automatically. This method also suffers from drawbacks such as cumbersome operation, low efficiency, and susceptibility to mismatches. Before using semi-automatic alignment, personnel need to manually configure the needle card file for the product, which also relies on experience. This configuration is also cumbersome, and the alignment effect requires repeated adjustments to the target position in the needle card. In actual production, this method does not significantly improve factory efficiency. Summary of the Invention
[0005] To address the technical problems existing in the prior art, this invention provides a general-purpose automated method for wafer pin identification.
[0006] To achieve the above objectives, the technical solution of the present invention is as follows:
[0007] An automated method for general-purpose wafer pin identification includes the following steps:
[0008] Step 1: Read the pin card and construct a standard gradient field;
[0009] Step 2: Pin image recognition;
[0010] Step 3, optimal position matching;
[0011] Step 4: Automatic needle alignment.
[0012] As a preferred technical solution, in step one, reading the needle card means reading the needle card file corresponding to the product and drawing the needle tip image of the needle card file. Constructing the standard gradient field is achieved by using a gradient diffusion algorithm to convert the needle tip position drawn in the standard needle card file into a gradient.
[0013] As preferred technical solutions, gradient diffusion algorithms include Gaussian filtering, linear filtering, and cosine roll-off filtering.
[0014] As a preferred technical solution, the steps for constructing a standard gradient field are as follows: first, diffuse grayscale pixels to the entire field, and ensure that the positive gradient direction at any point points to the final position of the needle tip. When gradient overlap occurs at two adjacent needle tip positions, use a needle tip weight assignment algorithm to assign different weights to each needle tip at the optimal position.
[0015] As a preferred technical solution, in step two, while constructing the gradient field of the needle card file, the machine screen is moved by the automated control platform, and the lens is sequentially traversed to all preset positions to capture the needle tip image at each actual needle tip position. The key point position AI recognition model is established and trained, and the needle tip key point position is identified through the model.
[0016] As a preferred technical solution, in step three, after obtaining the position information of the tip of each single frame, a full-field mapping is performed on all pin cards according to the relevant alignment information, and the pin tip key point recognition results are used to perform optimal matching with the standard gradient field of the full-field pin cards.
[0017] As a preferred technical solution, the optimal position matching strategy is as follows: first, place the key point position at a certain initial position in the gradient field, calculate the positive gradient direction of each needle tip key point at the initial position, and then move the key point position in space along the positive gradient direction by a fixed step size. Then, repeat the above operation continuously until the number of iterations is reached.
[0018] As a preferred technical solution, in step four, after matching is completed, the translation relationship between the frame tip position of the captured image and the standard gradient field is obtained. The translation relationship is the offset required for automatic frame alignment. The automatic frame alignment step is executed to map the offset to the forming mechanism of the operating machine to complete the alignment process.
[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0020] This invention provides a universal automated method for wafer pin alignment, automating the entire process from pin card file reading to pin tip identification and final second- and third-order comparisons. This significantly improves work efficiency and enhances the automation level of the pin alignment process. Its core advantage lies in the fact that, using this invention's automated algorithm and process, users no longer need to configure pin cards and target areas separately for each product. Instead, the algorithm automatically compares the standard pin card file with actual pin tip images captured by a camera. This involves a deep learning-based keypoint detection model, which can identify pin tip positions while removing noise and resisting lighting interference. This invention includes a multi-verification function, performing three or more comparisons on the beginning, middle, and end of the entire pin card file. The optimal alignment position is found through statistical averaging, offering higher accuracy compared to manual single-position matching and further improving work efficiency. Attached Figure Description
[0021] Figure 1 This is a flowchart of the automated method for general-purpose wafer pin identification according to the present invention;
[0022] Figure 2 This is the alignment screen in step two of the general-purpose automated wafer alignment identification method of the present invention;
[0023] Figure 3 This is an interface diagram of the training of the AI recognition model for key point positions in the general-purpose automated method for wafer pin identification of the present invention.
[0024] Figure 4 This is a schematic diagram of the optimal position matching in step three of the automated method for general-purpose wafer pin identification of the present invention. Detailed Implementation
[0025] The technical solution of the present invention will be further described below with reference to specific embodiments:
[0026] like Figure 1 As shown, an automated method for general-purpose wafer pin identification includes the following steps:
[0027] Step 1: Read the pin card and construct a standard gradient field;
[0028] Specifically, this method involves a general-purpose automated system for wafer pin identification. Before entering the automatic frame matching process for a particular product, the system first reads the corresponding SIM card file for that product and draws the pin tips of the SIM card file in the image. Key positions are represented by white dots, such as... Figure 2 The left image is shown in the figure. The invention proposes a standard gradient field construction algorithm, which converts the needle tip position drawn in the standard needle card file into a gradient. First, its grayscale pixels are diffused across the entire field, ensuring that the positive gradient direction at any point points towards the final position of the needle tip. A visualization of the standard gradient field is shown below. Figure 2 The image on the right is shown in the diagram. Gradient diffusion algorithms can use different filtering methods such as Gaussian filtering, linear filtering, and cosine roll-off filtering. Gradient overlap can occur at two closely spaced needle tips. To address this overlap, this invention uses a needle tip weighting algorithm, assigning different weights to each needle tip at its optimal position. This prevents the gradient direction from being perpendicular to the ideal direction in equipotential spatial locations between the needles, thus avoiding getting trapped in a submaximum when calculating the optimal gradient.
[0029] Step 2: Pin image recognition;
[0030] Specifically, while constructing the gradient field of the needle card file, an automated control platform controls the movement of the machine's screen, sequentially traversing all preset positions, such as the head, middle, and tail positions, capturing images of the needle tip at each actual needle tip location. In each needle tip image, the system calls a keypoint location AI recognition model to identify the needle tip's keypoint locations. This keypoint location AI recognition model is a simplified pixel segmentation model. Its core function is to determine the classification of each pixel. Compared to traditional segmentation models, the keypoint recognition model proposed in this invention does not classify dense pixels across the entire field; it only identifies discrete keypoint locations for subsequent matching.
[0031] like Figure 3 As shown, the model's establishment and training process relies on manually labeled prior data. Before functional deployment, a batch of images needs to be provided manually, and the locations of key points need to be labeled. These images are then trained using artificial intelligence algorithms, and the trained model can then possess key point recognition capabilities. During training, the training set can include various images under different lighting, noise, and focusing conditions. When the training set accumulates a sufficient number of images, the model will possess corresponding anti-interference capabilities.
[0032] Step 3, optimal position matching;
[0033] After obtaining the pin tip position information for each individual frame, the system performs full-field mapping on all pin cards based on relevant alignment information such as camera movement. Figure 4As shown in the image on the left. The fewer sub-screens identified during the traversal, the less information is available for matching, but the higher the efficiency. Conversely, traversing all screens of the pin card to identify all pin tip positions for matching yields the highest accuracy, but at the cost of some efficiency. The optimal position matching algorithm uses... Figure 4 The needle tip keypoint recognition results in the left image are compared with... Figure 4 The full-field standard gradient field of the right-hand image is optimized for matching.
[0034] The matching strategy proceeds as follows: First, the keypoints are placed at an initial position in the gradient field. Then, the positive gradient direction of each keypoint is calculated from this initial position. Next, the keypoints are moved in space along their positive gradient direction with a fixed step size. This process is repeated continuously. The preset step size can be adjusted in real-time based on the number of iterations. The optimization objective function includes minimizing the overall gradient direction, minimizing the gradient change between two iterations, maximizing the sum of the current tip weights for each point, and ensuring that the number of iterations reaches the termination condition.
[0035] After matching is completed, the translation relationship between the frame tip position of the captured image and the standard gradient field can be obtained. This translation relationship is the offset required for automatic frame alignment. Subsequently, an automatic frame alignment step can be performed to map this offset into the forming mechanism of the operating machine. At this point, the alignment process of the present invention is completed.
[0036] This embodiment is merely a further explanation of the present invention and is not intended to limit the present invention. Those skilled in the art can make non-creative modifications to this embodiment as needed after reading this specification, but as long as they are within the scope of the claims of the present invention, they are protected by patent law.
Claims
1. An automated method for general-purpose wafer pin identification, characterized in that, The method includes the following steps: Step 1: Read the pin card and construct a standard gradient field; Step 2: Pin image recognition; Step 3, optimal position matching; Step 4: Automatic needle alignment.
2. The automated method for general-purpose wafer pin identification according to claim 1, characterized in that, In step one, reading the needle card means reading the needle card file corresponding to the product and drawing the needle tip image from the needle card file. Constructing the standard gradient field involves using a gradient diffusion algorithm to convert the needle tip position drawn in the standard needle card file into a gradient.
3. The automated method for general-purpose wafer pin identification according to claim 2, characterized in that, The gradient diffusion algorithm includes Gaussian filtering, linear filtering, and cosine roll-off filtering.
4. The automated method for general-purpose wafer pin identification according to claim 2, characterized in that, The steps to construct a standard gradient field are as follows: First, diffuse the grayscale pixels to the entire field and ensure that the positive gradient direction at any point points to the final position of the needle tip. When gradient overlap occurs at two adjacent needle tip positions, use the needle tip weight assignment algorithm to assign different weights to each needle tip at the optimal position.
5. The automated method for general-purpose wafer pin identification according to claim 1, characterized in that, In step two, while constructing the needle card file gradient field, the machine screen is moved by the automated control platform, and the lens is sequentially traversed to all preset positions. The needle tip image is captured at each actual needle tip position, and a key point position AI recognition model is established and trained. The needle tip key point position is identified through the model.
6. The automated method for general-purpose wafer pin identification according to claim 5, characterized in that, In step three, after obtaining the position information of the tip of each single frame, a full-field mapping is performed on all pin cards according to the relevant alignment information, and the pin tip key point recognition results are used to perform optimal matching with the standard gradient field of the full-field pin cards.
7. The automated method for general-purpose wafer pin identification according to claim 6, characterized in that, The optimal position matching strategy is as follows: first, place the key point position at an initial position in the gradient field, calculate the positive gradient direction of each needle tip key point at the initial position, and then move the key point position in space along the positive gradient direction by a fixed step size. Then, repeat the above operation until the number of iterations is reached.
8. The automated method for general-purpose wafer pin identification according to claim 1, characterized in that, In step four, after matching is completed, the translation relationship between the frame tip position of the captured image and the standard gradient field is obtained. The translation relationship is the offset required for automatic frame alignment. The automatic frame alignment step is executed to map the offset to the forming mechanism of the operating machine to complete the alignment process.