Method for detecting fixed position defect clusters in chemical vapor deposition products

Abstract

The present application relates to a kind of chemical vapor deposition product fixed position defect cluster detection method, the semiconductor wafer detection technical field, the method includes: establishing the statistical process control (SPC) control chart of specific name, and set corresponding control limit;Monitoring Cluster defect data, when data exceeds control limit and produces OOC (out of control) alarm, lock batch;Retrieve the SPX detection Map of all silicon waits in OOC batch;The map is carried out overlay processing;According to the overlay result, manual judgment is carried out, if Cluster defect points are found to coincide in the same position after overlay, it is judged as fixed position Cluster defect and is processed with rework.By SPC monitoring early warning combined with artificial overlay confirmation, Cluster defects located in fixed position but not exceeding quantity specification can be effectively identified, solving the problem of missed detection caused by relying only on Cluster quantity control in the prior art and the problem of good product overkill caused by simply tightening specification, without affecting yield output efficiency, accurately intercepting defective product outflow.

Description

Technical Field

[0001] This invention relates to the field of semiconductor wafer inspection technology, and more specifically to a method for detecting fixed-position defect clusters in chemical vapor deposition products. Background Technology

[0002] Since the "12-inch Particle Film Thickness Resistivity Detection Recipe Setting Management Regulations" already specifies the default setting specifications for cluster count detection, the cluster count is judged based on the specifications in the specification sheet or the default value. For a batch of silicon wafers that meet the cluster count specifications, if the cluster count positions are located at the same fixed point (the defect overlay positions are the same), it is also a defect that needs to be detected and reworked. However, the current SM system cannot accurately identify and intercept it, and there is no relevant patent technology in the industry. Therefore, there is an urgent need for a detection method that can accurately detect cluster defects at fixed positions in CVD products.

[0003] The existing technology, "A Method for Repairing Exposure Fixing Point Defects in the Production of Circuit Boards for Aerospace Use", with application number CN201611055936.X, discloses that when an exposure fixing point defect is detected in a product, CAM data is first created, then a new film is drawn according to the actual film ratio during the product's exposure, and then the film is placed for a second exposure to remove the exposure fixing point defect. After development, etching, and film removal processes, AOI inspection is performed, and the process is completed after passing the inspection.

[0004] However, in actual production and client applications, the above-mentioned conventional detection methods have obvious limitations: I. Risk of Missed Detection of Fixed-Position Defects: Some particulate defects introduced by the substrate itself will be mapped as cluster defects after the CVD process. These defects are often few in number and do not exceed the conventional cluster quantity specification limit, so the equipment will automatically judge them as OK. However, if these defects occur in the same fixed position in the same cassette or multiple wafers (i.e., the defect overlay positions are the same), although the cluster quantity of a single wafer does not exceed the limit, the concentrated occurrence of fixed-position defects will still lead to customer complaints (e.g., feedback of cross-shaped abnormalities).

[0005] II. Yield Loss Due to Simply Tightening Specifications: In theory, to intercept the cluster defects at the aforementioned fixed locations, this can be achieved by tightening the upper limit of cluster defects in the SPX equipment. However, technical verification has shown that cluster defects usually have a random distribution. Tightening the upper limit will cause a large number of good products that only have a small number of clusters randomly appearing to be misjudged as defective products (over-crippling), which seriously affects the product yield and production efficiency.

[0006] Third, there is a lack of targeted detection methods: Currently used detection systems (such as the SM system) can only make judgments based on quantitative logic and cannot identify the characteristic of "fixed location". At the same time, there are no patented technologies or mature detection solutions in the industry for the automated detection of such fixed-location cluster defects. Summary of the Invention

[0007] This invention addresses the shortcomings of existing technologies by providing a method for detecting fixed-position defect clusters in chemical vapor deposition products. By combining SPC monitoring and early warning with manual image overlay confirmation, it can effectively identify cluster defects located in fixed positions that do not exceed the quantity specifications. This solves the problems of missed detection caused by relying solely on cluster quantity control and over-criteria of good products caused by simply tightening specifications in existing technologies. It accurately intercepts defective products from leaving the market without affecting yield and production efficiency.

[0008] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions: A method for detecting fixed-location defect clusters in chemical vapor deposition products includes the following steps: Step 1: Establish statistical process control charts for silicon wafers with specific product names.

[0009] Step 2: Monitor the cluster defect data of the silicon wafer. When the cluster defect data exceeds the control limit of the SPC control chart and generates an OOC alarm, lock the batch that generated the OOC.

[0010] Step 3: Retrieve the SPX detection map of all silicon wafers in the OOC batch.

[0011] Step 4: Overlay multiple SPX detection Map images from the batch.

[0012] Step 5: Perform manual judgment based on the overlay results. If cluster defect points are found to overlap in the same position after overlaying, it is judged as a fixed-position cluster defect; if cluster defect points are scattered, it is processed according to the system's set logic.

[0013] Preferably, the SPC control chart is an XR chart; the upper control limit UCL of the SPC control chart is set to 0.01. This value is set based on the theoretical maximum proportion of cluster defects generated by a single silicon wafer in a single silicon wafer box, ensuring that OOC interception can be triggered as long as a cluster defect occurs in a single box.

[0014] As a preferred embodiment, the calculation basis for the theoretical maximum ratio value is as follows: there are a maximum of 25 silicon wafers in a single box, and the minimum number of clusters is 1 wafer. The maximum ratio when a single wafer generates a cluster defect is 1 / 25 = 0.04. The UCL = 0.01 is less than 0.04, which meets the interception setting.

[0015] Preferably, the silicon wafers of the specific product name are inspected at SP3, SP5 and SP7 stations. In the first step, SPC control chart data IDs corresponding to the inspection values ​​of SP3, SP5 and SP7 need to be established respectively.

[0016] As a preferred approach, the batch name information that generated OOC is retrieved from the data decision system, and the batch information of Cluster OOC is exported from the SPX detection basic data report. The SPX detection map of the Cluster OOC silicon wafer is then retrieved using a map retrieval tool.

[0017] As a preferred option, batches with overlapping cluster defect points confirmed by manual overlay are reworked, and then SPX testing is performed again after rework.

[0018] Preferably, the silicon wafer is a 12-inch CVD type silicon wafer.

[0019] As a preferred approach, existing SPC monitoring resources can be combined with manual image overlay judgment to identify cluster defects that are not out of specification but located in fixed positions without tightening the upper limit of cluster defect control.

[0020] The present invention can achieve the following effects: This invention provides a method for detecting fixed-position defect clusters in chemical vapor deposition products, which has the following advantages compared with existing technologies: I. Precisely Identify Specific Defects and Avoid Missed Detections: Overcoming the limitations of existing technologies that rely solely on the number of clusters for control, this method effectively identifies cluster defects that repeatedly appear in fixed positions but do not exceed specification limits by establishing SPC control charts and combining them with manual overlay judgment. In particular, it can precisely intercept defects caused by substrate particles that appear in fixed positions on the CVD layer, effectively preventing defective products from reaching customers and reducing customer complaints.

[0021] II. Avoiding Over-Defective Products and Ensuring Production Efficiency: Unlike traditional improvement approaches that tighten the upper limit on the number of cluster defects, this method does not require adjusting the criteria for determining the number of cluster defects. This means that good products with a small number of clusters and a random distribution can still pass inspection normally. This method intercepts defective products in fixed locations while maximizing the pass rate for good products with random cluster defects, thereby ensuring product yield and production efficiency.

[0022] Third, low implementation cost and high operability: Utilizing existing SPC monitoring systems and SPX inspection equipment-generated map resources, there is no need to purchase expensive inspection equipment or undertake complex system reconstruction. Implementation can be achieved simply by creating specific SPC charts for early warning and manually verifying the maps. This method has been validated in actual production and recognized by customers, demonstrating extremely high practical application value.

[0023] IV. Optimized Quality Control Process: By incorporating the occurrence rate of cluster defects within a single cell into SPC management (e.g., setting UCL=0.01), the system transforms the monitoring of cluster defects from "post-event quantity statistics" to "process trend monitoring." Once a trend of defects in a fixed location appears, the system can trigger an OOC alarm, facilitating timely intervention by engineers to investigate the cause (e.g., substrate quality) and optimize the process from the source.

[0024] V. Flexible adaptation to different process nodes: Data IDs and SPC Charts are established for different inspection stations such as SP3, SP5, and SP7, which can meet the quality monitoring needs of 12-inch CVD silicon wafers at different manufacturing stages and have a wide range of applications. Detailed Implementation

[0025] The technical solution of the invention will be further described in detail below through examples.

[0026] Example 1: A method for detecting fixed-position defect clusters in chemical vapor deposition (CVD) products, primarily applied in the production inspection of 12-inch CVD silicon wafers. This method aims to address the problems in existing technologies that rely solely on cluster quantity specifications for control, leading to missed detection of cluster defects (such as fixed-position particle defects caused by the substrate) that are within specifications but located in fixed positions, and the over-criteria rejection of good products due to simply tightening specifications. The method includes the following steps: Step S1: Establish a Product Name SPC Chart (Statistical Process Control Chart). For the specific CVD product name requiring control, establish an SPC control chart in the inspection system. Since 12-inch silicon wafers typically undergo inspection at multiple stations such as SP3, SP5, and SP7, separate SPC charts need to be established for the three data IDs corresponding to the SP3, SP5, and SP7 inspection values.

[0027] In this embodiment, the SPC Chart type used is an XR chart.

[0028] Regarding the setting of the control limit: Considering that a box of silicon wafers contains a maximum of 25 wafers, and assuming a minimum detection limit of 1 cluster defect, when only one wafer in the box has a cluster defect, the occurrence rate is 1 / 25 = 0.04. To sensitively capture cluster anomaly signals, this embodiment sets the upper control limit (UCL) to 0.01. This setting value is less than 0.04, meaning that once a cluster defect appears in a batch of products, even if only a small number of wafers are detected, an SPC alarm can be triggered, thus including that batch in the investigation scope and avoiding missed detections.

[0029] Step S2: Cluster OOC (Out of Control) Determination and Batch Locking. The established SPCChart is used to monitor CVD products on the production line in real time. When the system detects that the Cluster defect data of a batch of products exceeds the set UCL (i.e., 0.01), the batch is determined to be an OOC batch, and the product name information and the specific location information where the OOC occurred (such as the slot number) are recorded.

[0030] Step S3: Retrieve all MAP maps for the Cluster OOC batch. Export the batch information identified as OOC from the data decision system and confirm the detailed information in the SPX inspection basic data report. Subsequently, use the map retrieval tool recommended by the IT department to quickly retrieve the SPX inspection map maps of all silicon wafers in the OOC batch, especially the map maps containing Cluster defect signals, to improve the efficiency of subsequent analysis.

[0031] Step S4: Perform manual image overlay verification. Multiple SPX detection maps from the same batch (especially the same slot) are manually overlaid. The purpose of overlaying is to observe the distribution characteristics of cluster defect points.

[0032] The specific judgment logic is as follows: If, after overlaying the images, it is found that the cluster defect points in multiple Map images coincide at the same coordinate position on the wafer (i.e., the defect position is fixed and the shape presents a specific shape such as a cross), then it is determined that the batch has a fixed-position cluster defect. This defect is a mapping of the substrate defect in the CVD process and is considered a defective product.

[0033] If, after overlaying the image, the cluster defect points are scattered and have no fixed overlapping features, then the defect is determined to be a random particle and processed according to the system's set logic (i.e., conventional quantity control), and is usually judged as a good product.

[0034] Step S5: Handling and Re-inspection. Batches with overlapping defects in fixed locations, confirmed by manual image overlay, are deemed NG and reworked. Reworked products must undergo SPX testing again, and the above testing process is repeated until they pass. Batches with dispersed defects as shown in the overlay images proceed normally to the next process.

[0035] Example 2: Specific Application Case. Taking a 12-inch CVD silicon wafer with batch number 001VQ4019A29 as an example, the implementation effect of the present invention will be explained.

[0036] I. Problem Discovery: Customers reported that during the SPX scanning process, there was an abnormal cross shape at a fixed position in the middle of this batch of products.

[0037] II. Initial Investigation and Reproduction: During our internal investigation, we retrieved the pre-shipment substrate map of the wafer and compared it with the abnormal map reported by the customer. We found that both maps contained particle dots at the same locations, confirming it as a substrate defect. Further confirmation indicated that these particle dots were identified as cluster counts during SPX testing.

[0038] III. System Limitations: Upon reviewing the corresponding Map on the SPX equipment, it was found that although the equipment detected a Cluster count defect, the system automatically classified it as OK because the set upper limit of the control specification exceeded the Cluster count reported by the customer. After evaluation by the technical department, it was determined that simply tightening the control upper limit of the SPX cluster count would lead to a large number of good products being misclassified due to random particle counts, severely impacting product yield and production efficiency; therefore, this approach is not feasible.

[0039] IV. Application of the Invention Method: SPC monitoring: Based on the SPC Chart established in Example 1, this batch triggered a Cluster OOC alarm (UCL=0.01) during inspection at a certain workstation in SP3 / SP5 / SP7.

[0040] Map retrieval: The system automatically locks and exports the SPX detection map for this batch (especially Slot 07 to Slot 15).

[0041] Haze Map Analysis: Engineers performed haze map processing on the Slots 07 to 15 of this batch. The results showed that cluster defect points overlapped at the same location at the center point of these wafers, exhibiting a distinct cross-shaped abnormality, corresponding to the location of the substrate particles.

[0042] V. Final Judgment: Based on the judgment rules of this invention, it was confirmed that this batch had a fixed-location cluster defect. Although the number of clusters did not exceed the upper limit of the conventional specification, it was still accurately detected.

[0043] VI. Result Verification: After rework, the batch passed the map test again, indicating the defect was eliminated. This method successfully prevented defective products from reaching customers and avoided the risk of misjudgment that might arise from strictly controlling quantity and specifications.

[0044] In summary, this method for detecting fixed-location defect clusters in chemical vapor deposition products, by establishing an early warning mechanism through SPC charts and combining it with manual image overlay confirmation, can effectively identify cluster defects that are within specifications but located in fixed positions. This method utilizes existing SPC resources and manual assistance, without requiring high equipment modification costs. It prevents missed detection of defective products, avoiding customer complaints, and also avoids over-discarding of good products, thus ensuring production efficiency.

[0045] The above description is only a specific embodiment of the present invention, but the structural features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art within the scope of the present invention are covered by the patent scope of the present invention.

Claims

1. A method for detecting fixed-position defect clusters in chemical vapor deposition products, characterized in that... The following steps are included: Step 1: Establish statistical process control charts for silicon wafers with specific product names; Step 2: Monitor the cluster defect data of the silicon wafer. When the cluster defect data exceeds the control limit of the SPC control chart and generates an OOC alarm, lock the batch that generated the OOC. Step 3: Retrieve the SPX detection map of all silicon wafers in the OOC batch; Step 4: Overlay multiple SPX detection Map images from the batch; Step 5: Perform manual judgment based on the overlay results. If cluster defect points are found to overlap in the same position after overlaying, it is judged as a fixed-position cluster defect; if cluster defect points are scattered, it is processed according to the system's set logic.

2. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 1, characterized in that: The SPC control chart is of type XR chart; the upper control limit UCL of the SPC control chart is set to 0.

01. This value is set based on the theoretical maximum proportion of cluster defects generated by a single silicon wafer in a single silicon wafer box, ensuring that OOC interception can be triggered as long as a cluster defect occurs in a single box.

3. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 2, characterized in that: The calculation basis for the theoretical maximum ratio is as follows: there are a maximum of 25 silicon wafers in a single box. Assuming the minimum number of clusters is 1 wafer, the maximum ratio of a single wafer to generate a cluster defect is 1 / 25 = 0.

04. The UCL = 0.01 is less than 0.04, which meets the interception setting.

4. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 1, characterized in that: The silicon wafers of the specific product name are tested at SP3, SP5 and SP7 stations. In the first step, SPC control chart data IDs corresponding to the test values ​​of SP3, SP5 and SP7 need to be established respectively.

5. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 1, characterized in that: Extract the batch name information that generated OOC from the data decision system, export the batch information of Cluster OOC from the SPX detection basic data report, and use the map extraction tool to extract the SPX detection map of the Cluster OOC silicon wafer.

6. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 1, characterized in that: Batches with overlapping cluster defect points, confirmed by manual overlay mapping, are reworked and then re-tested using SPX.

7. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 1, characterized in that: The silicon wafer is a 12-inch CVD type silicon wafer.

8. The method for detecting fixed-position defect clusters in chemical vapor deposition products according to claim 1, characterized in that: By combining existing SPC monitoring resources with manual image overlay judgment, cluster defects that are within specifications but located in fixed positions can be identified without tightening the upper limit of cluster defect control.