A method for generating knowledge points based on SPC numerical sequence data

Abstract

The application relates to a method for generating a knowledge point based on SPC numerical sequence data, which comprises the following steps: obtaining a target numerical sequence from statistical process control (SPC) data generated in semiconductor manufacturing; performing feature extraction on the target numerical sequence to obtain a feature vector sequence; inputting the feature vector sequence into a trained machine learning model to obtain a plurality of abnormality detection results, each of which at least includes a data abnormality type of a sequence segment in the target numerical sequence and a corresponding wafer number; and generating a target knowledge point according to the plurality of abnormality detection results, wherein the target knowledge point is used for generating or updating a knowledge graph in the semiconductor field.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing, and in particular to a method for generating knowledge points based on SPC numerical sequence data. Background Technology

[0002] The semiconductor manufacturing process generates various types of data, which can be broadly categorized into two types based on their source: one type comes from various equipment, such as equipment operating status, equipment parameters, equipment manuals, and operating logs; the other type comes from the testing data of the produced chips, such as the chip's electrical characteristics and defects. In current production processes, often only a small portion of this data is analyzed to guide production, and the majority of the data remains underutilized.

[0003] In knowledge graph generation, the conventional approach involves extracting nodes and causal / sequential relationships between them from open-domain text documents using Natural Language Processing (NLP) to construct the knowledge graph. However, due to the large amount of structured data in semiconductor manufacturing, such as SPC numerical sequence data, the conventional knowledge graph generation method requires first generating event description text using a language generation model, and then extracting knowledge from the event description text to construct the knowledge graph. This data processing involves language generation and deconstruction, which is cumbersome and inefficient. Summary of the Invention

[0004] This specification describes one or more embodiments of a method for generating knowledge points based on SPC numerical sequence data. Instead of generating event description text from the data, it directly generates knowledge points from the data. These knowledge points can be in triplet form, and a knowledge graph can then be generated based on these knowledge points. This simplifies the process of generating a knowledge graph throughout the semiconductor manufacturing process. After generating the knowledge graph, it can guide the entire semiconductor production process.

[0005] Firstly, a method for generating knowledge points based on SPC numerical sequence data is provided, including:

[0006] Obtain the target numerical sequence from the statistical process control (SPC) data produced in semiconductor manufacturing;

[0007] Feature extraction is performed on the target numerical sequence to obtain a feature vector sequence;

[0008] The feature vector sequence is input into the trained machine learning model to obtain several anomaly detection results. Each anomaly detection result includes at least the data anomaly type of a sequence segment in the target numerical sequence and the corresponding wafer number.

[0009] Based on the aforementioned anomaly detection results, target knowledge points are generated, which are used to generate or update a knowledge graph in the semiconductor field.

[0010] In one possible implementation, the statistical process control (SPC) data includes a dictionary sequence of key-value pairs ordered by key; obtaining the target numerical sequence from the SPC data produced by equipment in semiconductor manufacturing includes:

[0011] Extract subsequences from the dictionary sequence, and extract the values ​​from the key-value pairs contained in the subsequences to form the target numerical sequence.

[0012] In one possible implementation, feature extraction is performed on the target numerical sequence to obtain a feature vector sequence, including:

[0013] The first vector sequence is obtained by processing each value in the target numerical sequence using a feature extraction model.

[0014] Based on the position of each value in the target value sequence, a second vector sequence is obtained by embedding the values.

[0015] The feature vector sequence is determined based on the first vector sequence and the second vector sequence.

[0016] In one possible implementation, the feature extraction model is a convolutional neural network or a recurrent neural network.

[0017] In one possible implementation, the machine learning model includes a Transformer-based encoder and decoder, and a feedforward prediction network; the feature vector sequence is input into the trained machine learning model to obtain several anomaly detection results, including:

[0018] The encoder is used to perform attention-based encoding on the feature vector sequence to obtain an encoded vector sequence.

[0019] The decoder is used to decode the encoded vector sequence to obtain a decoded vector sequence.

[0020] The feedforward prediction network is used to process the decoded vector sequence and output the several anomaly detection results.

[0021] In one possible implementation, the anomaly detection result is in the form of a triple, which includes the data anomaly type, and the start and end position indices of the corresponding sequence segment in the target numerical sequence.

[0022] In one possible implementation, target knowledge points are generated based on the aforementioned anomaly detection results, including:

[0023] For any anomaly detection result, the corresponding anomaly location is determined based on the start and end position indices in its corresponding triplet, combined with the relative position of the target numerical sequence in the SPC data.

[0024] Based on the data anomaly type and the location where the anomaly occurred, target knowledge points are generated.

[0025] In one possible implementation, it also includes:

[0026] The target knowledge points are added to a temporary knowledge base, and the knowledge points in the temporary knowledge base are deduplicated.

[0027] The knowledge graph is generated or updated using the knowledge points in the temporary knowledge base after deduplication.

[0028] In one possible implementation, the machine learning model is trained on a training set, which includes multiple training samples, including sample numerical sequence data and labeled anomaly type tags.

[0029] In one possible implementation, the data anomaly types include at least: several data points above the UCL line, several data points below the LCL line, several data points continuously above the center line, several data points continuously below the center line, several data points continuously rising, several data points continuously falling, and several data points alternating between rising and falling.

[0030] Secondly, an apparatus for generating knowledge points based on SPC numerical sequence data is provided, including:

[0031] The data acquisition unit is configured to acquire a target numerical sequence from the statistical process control (SPC) data produced in semiconductor manufacturing; the feature extraction unit is configured to extract features from the target numerical sequence to obtain a feature vector sequence; the anomaly detection unit is configured to input the feature vector sequence into a trained machine learning model to obtain several anomaly detection results, each anomaly detection result including at least the data anomaly type of a sequence segment in the target numerical sequence and the corresponding wafer number; and the knowledge point generation unit is configured to generate target knowledge points based on the several anomaly detection results, the target knowledge points being used to generate or update a knowledge graph in the semiconductor field.

[0032] In one possible implementation, the apparatus further includes: a deduplication unit configured to add the target knowledge points to a temporary knowledge base and perform deduplication processing on the knowledge points in the temporary knowledge base; and a knowledge graph generation unit configured to generate or update the knowledge graph using the deduplicated knowledge points in the temporary knowledge base.

[0033] This invention proposes a method for generating knowledge points based on SPC numerical sequence data. Using an end-to-end approach, it does not generate event description text from the data, but directly generates knowledge points from the data. Then, a knowledge graph can be generated based on the knowledge points, thereby simplifying the process of generating knowledge points and knowledge graphs throughout the semiconductor manufacturing process. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the various embodiments disclosed in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only a few embodiments disclosed in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart of a method for generating knowledge points based on SPC numerical sequence data, as disclosed in an embodiment of the present invention.

[0036] Figure 2 This is an SPC control diagram disclosed in an embodiment of the present invention;

[0037] Figure 3 This is a flowchart of generating knowledge points and knowledge graphs based on SPC numerical sequence data, as disclosed in an embodiment of the present invention.

[0038] Figure 4 This is a schematic block diagram of an apparatus for generating knowledge points based on SPC numerical sequence data, as disclosed in an embodiment of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Statistical Process Control (SPC) is a process control tool that utilizes mathematical statistics. It analyzes and evaluates the production process, promptly identifies signs of systematic factors based on feedback information, and takes measures to eliminate their impact, maintaining the process in a controlled state affected only by random factors to achieve quality control. It uses statistical methods to monitor the process status, ensuring the production process is under control to reduce product quality variation. Semiconductor manufacturing generates different types of SPC data. Currently, the common practice is for engineers to analyze this data using SPC tools, i.e., statistical methods, and then generate reports on defect analysis and yield improvement. However, this approach requires manual intervention and, when dealing with large volumes of data, cannot efficiently and promptly identify problems.

[0041] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0042] Figure 1 This is a flowchart illustrating a method for generating knowledge points from SPC numerical sequence data, as disclosed in an embodiment of the present invention. Figure 1 As shown, the method includes at least the following steps: Step 101, obtaining a target numerical sequence from the statistical process control (SPC) data produced in semiconductor manufacturing; Step 102, extracting features from the target numerical sequence to obtain a feature vector sequence; Step 103, inputting the feature vector sequence into a trained machine learning model to obtain several anomaly detection results, each anomaly detection result including at least the data anomaly type of a sequence segment in the target numerical sequence and the corresponding wafer number; Step 104, generating target knowledge points based on the several anomaly detection results, the target knowledge points being used to generate or update a knowledge graph in the semiconductor field.

[0043] In step 101, the target numerical sequence is obtained from the statistical process control (SPC) data produced in semiconductor manufacturing.

[0044] Among them, the statistical process control (SPC) data is a numerical sequence related to wafer manufacturing.

[0045] The numerical sequence related to wafer manufacturing can include the numerical sequence related to wafer manufacturing equipment, such as data on the status of wafer manufacturing equipment (also known as a machine) obtained by sensors during production, including but not limited to: temperature, humidity, pressure, voltage, current, etc.; and the utilization rate of wafer manufacturing equipment, etc.

[0046] And / or, wafer-related data, such as data obtained through defect detection during the wafer manufacturing process (e.g., wafer defect data); data obtained through electrical testing during the wafer manufacturing process (e.g., wafer failure type data, including CPU zone failure, GPU zone failure, memory zone failure, etc.); wafer yield data and other purely numerical data.

[0047] There are various ways to acquire Statistical Process Control (SPC) data. For example, it can be acquired directly from the acquisition devices of each machine in the semiconductor manufacturing process, or it can be retrieved from a database that stores structured numerical data generated by each machine in the semiconductor manufacturing process, or it can be acquired by receiving user input, etc.

[0048] In one embodiment, the statistical process control (SPC) data includes a dictionary sequence consisting of key-value pairs ordered by keys; in this case, obtaining the target numerical sequence from the SPC data produced by the equipment in semiconductor manufacturing includes: extracting a subsequence from the dictionary sequence, extracting the values ​​from the key-value pairs contained in the subsequence, and constructing the target numerical sequence.

[0049] In step 102, feature extraction is performed on the target numerical sequence to obtain a feature vector sequence.

[0050] In one embodiment, a feature extraction model is used to process each value in the target numerical sequence to obtain a first vector sequence; position embedding is performed based on the position of each value in the target numerical sequence to obtain a second vector sequence; and the feature vector sequence is determined based on the first vector sequence and the second vector sequence.

[0051] In a more specific embodiment, the feature extraction model is a convolutional neural network (CNN) or a recurrent neural network (RNN). The step of processing each value in the target numerical sequence using the feature extraction model to obtain a first vector sequence includes: dividing the target numerical sequence into several segments of the same size, and inputting the segments into the feature extraction model to obtain the first vector sequence.

[0052] In step 103, the feature vector sequence is input into the trained machine learning model to obtain several anomaly detection results. Each anomaly detection result includes at least the data anomaly type of a sequence segment in the target numerical sequence and the corresponding wafer number. In step 104, based on the several anomaly detection results, target knowledge points are generated. These target knowledge points are used to generate or update a knowledge graph in the semiconductor field.

[0053] The machine learning model is trained on a training set, which includes multiple training samples, including sample numerical sequence data and labeled anomaly type tags.

[0054] Training sets can be obtained in various ways. For example, they can be extracted from engineer experience documents, which record historical SPC numerical sequence data generated by wafer manufacturing equipment, as well as anomaly analysis and identification of the SPC numerical sequence data. Multiple SPC numerical sequence data and their corresponding anomaly types can be extracted from these documents, and these multiple SPC numerical sequence data and their corresponding anomaly types constitute the training set. Alternatively, historical SPC numerical sequence data generated by multiple wafer manufacturing equipment can be obtained, and then a preset rule template can be used to match the corresponding anomaly types to the anomalous data segments in the SPC numerical sequence data.

[0055] In one embodiment, the machine learning model includes an encoder and decoder based on a Transformer model, and a feedforward prediction network; the step of inputting the feature vector sequence into the trained machine learning model to obtain several anomaly detection results includes: using the encoder to encode the feature vector sequence based on an attention mechanism to obtain an encoded vector sequence; using the decoder to decode the encoded vector sequence to obtain a decoded vector sequence; and using the feedforward prediction network to process the decoded vector sequence and output the several anomaly detection results.

[0056] In one possible implementation, the anomaly detection result is in the form of a triple, which includes the data anomaly type and the start and end position indices of the corresponding sequence segment in the target numerical sequence. For example, the triple can be in the form of (data anomaly type, start position index, end position index).

[0057] Based on this, step 104, which involves generating target knowledge points based on the plurality of anomaly detection results, includes: for any anomaly detection result, determining the corresponding anomaly occurrence location based on the start and end position indices in its corresponding triplet, combined with the relative position of the target numerical sequence in the SPC data; and generating target knowledge points based on the data anomaly type and the anomaly occurrence location.

[0058] A target numerical sequence can contain multiple sequence segments with anomalies. Each sequence segment has a corresponding data anomaly type, and each anomalous sequence segment corresponds to a start position index and an end position index. The start position index represents the starting position of this anomalous sequence segment in the target numerical sequence, and the end position index represents the ending position of this anomalous sequence segment in the target numerical sequence. For example, a target numerical sequence (v1, v2, ..., v...) 100 After detection by the machine learning model, two anomaly detection results were found. The first anomaly detection result has a data anomaly type of one, a start position index of 20, and an end position index of 30. This anomaly detection result can be represented as a triple (data anomaly type one, 20, 30), representing the sequence segment (v). 20 ,v 21 ,…,v 30 The first anomaly is identified as data anomaly type 1; the second anomaly detection result is data anomaly type 2, with a start position index of 60 and an end position index of 80. Therefore, this anomaly detection result can be represented as a triplet (data anomaly type 2, 60, 80), representing the sequence segment (v). 60 ,v 61 ,…,v 80 There is an anomaly, and the anomaly is data anomaly type two.

[0059] At this point, the location of the anomaly within the SPC data can be determined by combining the start and end indexes of the anomaly with the relative position of the target numerical sequence within the SPC data. Then, target knowledge points can be generated based on the data anomaly type and the location of the anomaly.

[0060] Regarding the setting of data anomaly types Figure 2 An SPC control chart is shown. (Example) Figure 2 As shown, the center line represents the mean μ of SPC data of the same type (outliers removed), and σ represents the standard deviation of the data. The range of 1 standard deviation above and below the mean is designated as Zone C, the range between 1 and 2 standard deviations is designated as Zone B, and the range between 2 and 3 standard deviations is designated as Zone A. The position of μ+3σ is the upper control limit (UCL), and the position of μ-3σ is the lower control limit (LCL). Those skilled in the art will understand that UCL means the upper limit of the specification for the characteristic value; that is, a product characteristic exceeding the UCL will result in a non-conformity in engineering. LCL means the lower limit of the specification for the characteristic value; that is, a product characteristic falling below the LCL will result in a non-conformity in engineering.

[0061] based on Figure 2This invention provides several types of data anomalies, including at least: N data points above the UCL line, N data points below the LCL line, N data points continuously above the center line, N data points continuously below the center line, N data points continuously rising, N data points continuously falling, N data points alternating between rising and falling, N out of M points being in or beyond the upper A zone of the center line, N out of M points being in or beyond the lower A zone of the center line, N out of M points being in or beyond the upper B zone of the center line, N out of M points being in or beyond the lower B zone of the center line, N out of M points being in or beyond the lower B zone of the center line, N out of M points being in or beyond the lower B zone of the center line, N consecutive points in the C zone, and N consecutive points not in the C zone.

[0062] In this context, M and N are both preset values ​​that can be set by engineers based on actual conditions or experience, or calculated by some data models based on historical data.

[0063] In some possible implementations, the method further includes: step 105, adding the target knowledge point to a temporary knowledge base and performing deduplication processing on the knowledge points in the temporary knowledge base; step 106, using the deduplicated knowledge points in the temporary knowledge base to generate or update the knowledge graph.

[0064] Figure 3 A flowchart illustrating a more specific embodiment is shown. For example... Figure 3 As shown, a complete dictionary sequence of SPC database information is obtained from the relational database RDB. This dictionary sequence includes the type of SPC numerical sequence data and the specific SPC numerical sequence data. A type of SPC data is selected and extracted from the dictionary sequence to obtain the SPC numerical sequence data.

[0065] The SPC numerical sequence data is divided into several segments of the same size. These segments are then input into a feature extraction model to obtain a feature vector sequence. This feature extraction model can be a CNN or an RNN. Next, the feature vector sequence is positionally embedded to obtain a feature vector sequence V containing positional information.

[0066] The feature vector sequence V containing location information is input into the encoder of the Transformer model for encoding, resulting in the encoded feature vector sequence E. The encoded feature vector sequence E and the query vector sequence Q are then input into the decoder of the Transformer model to obtain the decoded feature vector sequence D. The query vector sequence Q can be understood as the output of the decoder from the previous round. The decoded feature vector sequence D is then passed through a feedforward neural network, followed by linear and softmax layers to obtain the probability corresponding to each anomaly type. The anomaly type with the highest probability is selected, thus determining the anomaly type.

[0067] Based on the location information, the start and end positions of the anomaly are determined in the SPC numerical sequence data; the anomaly type, the start position, and the end position are combined to obtain a triple (data anomaly type, start position index, end position index).

[0068] To generate a knowledge graph from triples, we first define two types of entities in the knowledge graph: event entities (Entity) and object entities (Event), as well as two types of relationship edges: the relationship edge between object entities and event entities (Entity-Event, En-Ev), and the relationship edge between object entities (Entity-Entity, En-En).

[0069] The process of constructing a knowledge graph involves retrieving an existing knowledge graph from the Knowledge Graph Database (GDB), or creating a new knowledge graph if none exists. For any triple, the process first retrieves its corresponding production process information from the original dictionary sequence, such as the production step (`step`), the production equipment (`tool_entity`), the production time (`time`), and the wafer ID (`wafer_id`). Then, based on this information, the corresponding SPC event node (`SPCEvent`) is searched for in the graph (if it doesn't exist, a new one is created). Next, based on the `tool_entity`, the corresponding device entity node (`ToolEntity`) is searched for in the graph (if it doesn't exist, a new one is created). An En-Ev relationship edge is constructed between the SPCEvent node and the ToolEntity node. Finally, other production information, such as `step`, `time`, and `wafer_id`, is filled into the attribute information of the En-Ev relationship edge. This completes the process of adding a triple to the knowledge graph.

[0070] Figure 4This is a schematic block diagram of an apparatus for generating knowledge points based on SPC numerical sequence data, as disclosed in an embodiment of the present invention. The apparatus 400 includes: a data acquisition unit 401 configured to acquire a target numerical sequence from Statistical Process Control (SPC) data produced in semiconductor manufacturing; a feature extraction unit 402 configured to extract features from the target numerical sequence to obtain a feature vector sequence; an anomaly detection unit 403 configured to input the feature vector sequence into a trained machine learning model to obtain several anomaly detection results, each anomaly detection result including at least the data anomaly type of a sequence segment in the target numerical sequence and the corresponding wafer number; and a knowledge point generation unit 404 configured to generate target knowledge points based on the several anomaly detection results, the target knowledge points being used to generate or update a knowledge graph in the semiconductor field.

[0071] In some possible implementations, the device 400 further includes: a deduplication unit 405 configured to add the target knowledge points to a temporary knowledge base and perform deduplication processing on the knowledge points in the temporary knowledge base; and a knowledge graph generation unit 406 configured to generate or update the knowledge graph using the deduplicated knowledge points in the temporary knowledge base.

[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0073] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0074] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating knowledge points based on SPC numerical sequence data, comprising: Obtain the target numerical sequence from the statistical process control (SPC) data produced in semiconductor manufacturing; Feature extraction is performed on the target numerical sequence to obtain a feature vector sequence; The feature vector sequence is input into the trained machine learning model to obtain several anomaly detection results. Each anomaly detection result includes at least the data anomaly type of a sequence segment in the target numerical sequence and the corresponding wafer number. The anomaly detection result is in the form of a triplet, which includes the data anomaly type and the start and end position indices of the corresponding sequence segment in the target numerical sequence. For any anomaly detection result, the corresponding anomaly location is determined based on the start and end position indices in its corresponding triplet, combined with the relative position of the target numerical sequence in the SPC data. Based on the data anomaly type and the location of the anomaly, target knowledge points are generated, which are used to generate or update the knowledge graph in the semiconductor field.

2. The method according to claim 1, characterized in that, The statistical process control (SPC) data includes a dictionary sequence consisting of key-value pairs and sorted by key. Obtain the target numerical sequence from the statistical process control (SPC) data of equipment output in semiconductor manufacturing, including: Extract subsequences from the dictionary sequence, and extract the values ​​from the key-value pairs contained in the subsequences to form the target numerical sequence.

3. The method of claim 1, wherein, Feature extraction is performed on the target numerical sequence to obtain a feature vector sequence, including: The first vector sequence is obtained by processing each value in the target numerical sequence using a feature extraction model. Based on the position of each value in the target value sequence, a second vector sequence is obtained by embedding the values. The feature vector sequence is determined based on the first vector sequence and the second vector sequence.

4. The method of claim 3, wherein, The feature extraction model is a convolutional neural network or a recurrent neural network.

5. The method of claim 1, wherein, The machine learning model includes a Transformer-based encoder and decoder, as well as a feedforward prediction network; The feature vector sequence is input into the trained machine learning model to obtain several anomaly detection results, including: The encoder is used to perform attention-based encoding on the feature vector sequence to obtain an encoded vector sequence. The decoder is used to decode the encoded vector sequence to obtain a decoded vector sequence. The feedforward prediction network is used to process the decoded vector sequence and output the several anomaly detection results.

6. The method of claim 1, wherein, Also includes: The target knowledge points are added to a temporary knowledge base, and the knowledge points in the temporary knowledge base are deduplicated. The knowledge graph is generated or updated using the knowledge points in the temporary knowledge base after deduplication.

7. The method of claim 1, wherein, The machine learning model is trained on a training set, which includes multiple training samples, including sample numerical sequence data and labeled anomaly types.

8. The method of claim 1, wherein, The data anomaly types include at least the following: several data points are above the UCL line, several data points are below the LCL line, several data points are continuously above the center line, several data points are continuously below the center line, several data points are continuously rising, several data points are continuously falling, and several data points are alternating between rising and falling.

Images