Semiconductor product logistics whole-course quality safety tracing method

Abstract

The invention discloses a whole-course quality safety tracing method for semiconductor product logistics, relates to the technical field of quality tracing analysis, and solves the technical problems that an encryption strategy does not distinguish field-level sensitivity and cannot give consideration to safety and transmission efficiency at the same time. Field-level differential encryption is carried out on structured / unstructured data, the bandwidth and the calculation overhead are remarkably reduced while the security of core intellectual property is guaranteed, a triple generation mechanism with source evidence and 0-10 quantization confidence is provided, multi-source cross validation and expert confirmation state automatic scoring are combined, and the security of the core intellectual property is improved. The reliability and auditing compliance of the initial quality safety knowledge graph are ensured, the key node weight is automatically learned from the knowledge graph based on the graph neural network, the multi-factor coupling risk is effectively identified in combination with the yield + / -3 sigma dynamic threshold value of nearly three months, and misjudgment of a traditional fixed threshold value in the yield fluctuation period is avoided.

Description

Technical Field

[0001] This invention relates to the field of quality traceability analysis technology, specifically a method for tracing the quality and safety of semiconductor product logistics throughout the entire process. Background Technology

[0002] In the semiconductor manufacturing industry, product quality is highly dependent on the stability of the production environment and the precise control of process parameters. Currently, companies generally deploy sensor networks to monitor environmental parameters such as temperature, humidity, cleanliness, and air pressure in real time, and record process data and quality events through systems such as MES and QMS.

[0003] However, the existing quality and safety traceability system has significant shortcomings: on the one hand, the collected data is mostly stored in the form of isolated structured logs or unstructured files, lacking a unified semantic association;

[0004] On the other hand, data transmission generally adopts full encryption or static hierarchical encryption without combining a dynamic adjustment strategy based on content sensitivity, resulting in insufficient protection of highly sensitive data or excessive encryption of low-sensitivity data, which affects system performance.

[0005] Furthermore, while some systems have incorporated knowledge graphs for quality issue analysis, their construction relies on manual rules or static FMEA documents. The sources of triples are untraceable, confidence levels are subjective, and the graphs are disconnected from real-time risk assessment. Existing risk warnings are mostly based on fixed thresholds or simple statistical models, failing to identify potential failure modes under multi-factor coupling and hindering feedforward intervention. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for tracing the quality and safety of semiconductor products throughout the entire logistics process, which solves the problem that encryption strategies do not differentiate between field-level sensitivity and cannot balance security and transmission efficiency.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for tracing the quality and safety of semiconductor product logistics throughout the entire process, the method specifically including the following steps:

[0008] Step 1: Assign a globally unique identifier to the semiconductor product, collect production environment parameters in real time and bind them to the identifier, preprocess the collected data to obtain preprocessed data;

[0009] Step 2: Identify the field types in structured and unstructured data, classify them into high-sensitivity fields and low-sensitivity fields. For high-sensitivity fields, use strong encryption algorithms to encrypt them, and for low-sensitivity fields, use lightweight encryption algorithms to encrypt them. For image files in unstructured data, extract key feature information, convert it into structured description fields, and then encrypt it. The original file is encrypted using digital envelope technology.

[0010] Step 3: Construct an initial quality and safety knowledge graph based on the preprocessed data after encrypted transmission. Convert the data into triples with accompanying source evidence and confidence scores, and import them into a graph database to generate the graph.

[0011] Step 4: Use graph neural networks to learn features from the knowledge graph and build a risk assessment model based on the attention mechanism to identify key quality-affecting nodes. Set a dynamic risk threshold based on historical yield data, calculate the real-time risk value through the model, and trigger a graded early warning mechanism when the risk value exceeds the dynamic threshold to generate early warning information containing the location of the risk source and the solution.

[0012] As a further aspect of the present invention, the preprocessing includes missing value imputation, outlier identification, and standardization;

[0013] Missing value imputation specifically involves using linear interpolation to imput data that is missing for no more than 5 minutes consecutively. If the missing time exceeds 5 minutes, it is marked as an abnormal data segment.

[0014] As a further aspect of the present invention, the highly sensitive fields include wafer batch number, key process parameters, defect coordinates, customer order information and intellectual property-related data, and the low-sensitivity fields include workshop temperature and humidity records, general equipment status identifiers and publicly disclosed process node names.

[0015] As a further aspect of the present invention, the unstructured, highly sensitive data employs a content-aware encryption strategy:

[0016] First, the defect type, size, and location coordinates are extracted through image recognition, converted into structured fields, and then strongly encrypted. The original file is encrypted using digital envelope technology, that is, the file is encrypted with a random symmetric key, and then the symmetric key is encrypted with the recipient's public key.

[0017] For unstructured data exceeding 100MB, fragmented transmission is used, with each fragment having a unique sequence number, checksum, and total number of fragments. Resumption of transmission after interruption and reassembly after integrity verification are supported.

[0018] As a further aspect of the present invention, the confidence score is determined comprehensively based on the amount of evidence, data consistency, and expert confirmation status:

[0019] Single unconfirmed evidence is scored 4–6 points, dual-source cross-validation is scored 7–8 points, and key data reviewed by experts is scored 9–10 points.

[0020] As a further aspect of the present invention, the initial quality and safety knowledge graph creates node labels and relationship types in the graph database according to the ontology model, and establishes indexes for equipment numbers and material codes, outputting statistical information including the number of nodes, the number of relationships, and the core business processes covered.

[0021] As a further aspect of the present invention, the graph neural network automatically identifies key nodes that significantly affect product quality through attention weights, including the lithography machine temperature control module, wafer cleaning time, and photoresist viscosity parameters.

[0022] As a further aspect of the present invention, the graded early warning mechanism is as follows: a risk value greater than the threshold of 120% triggers a level one early warning, suspends the production line and pushes the information to the engineer's mobile device; a risk value between 100% and 120% triggers a level two early warning, notifies the process supervisor via email and initiates manual review; a risk value between 80% and 100% triggers a level three early warning, marks the risk value in yellow in the monitoring system and records the potential risk point.

[0023] As a further aspect of the present invention, the globally unique identifier is generated according to the product batch, wafer number, and chip location hierarchy, and is encoded using a combination of letters and numbers, and cannot be repeated throughout its entire life cycle.

[0024] This invention provides a method for tracing the quality and safety of semiconductor products throughout the entire logistics process. Compared with existing technologies, it has the following advantages:

[0025] This invention assigns a globally unique identifier to each semiconductor product and implements field-level differentiated encryption for structured / unstructured data. While ensuring the security of core intellectual property rights, it significantly reduces bandwidth and computational overhead. It proposes a triple generation mechanism with source evidence and 0-10 quantitative confidence levels, combined with multi-source cross-validation and automatic scoring of expert confirmation status, to ensure the reliability and audit compliance of the initial quality and safety knowledge graph.

[0026] This invention uses graph neural networks to automatically learn the weights of key nodes from knowledge graphs and combines them with a dynamic threshold of ±3σ yield over the past 3 months to effectively identify multi-factor coupling risks. It avoids misjudgments caused by traditional fixed thresholds during yield fluctuations, and innovatively designs a three-level early warning response mechanism. It also automatically generates early warning information that includes risk source location, impact path, and solutions for similar historical cases, thus shortening the response time for quality anomalies. Attached Figure Description

[0027] Figure 1 This is a flowchart of the quality and safety traceability method of the present invention. Detailed Implementation

[0028] 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, and 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.

[0029] Please see Figure 1This application provides a method for tracing the quality and safety of semiconductor products throughout the entire logistics process, which specifically includes the following steps:

[0030] Step 1: Assign a globally unique identifier to each semiconductor product according to hierarchical information such as product batch, wafer number, and chip location. This identifier uses a combination of numbers and letters to ensure non-repeatability throughout the entire production lifecycle, facilitating subsequent data tracking and correlation analysis. Environmental parameters such as temperature, relative humidity, airborne particle concentration, and atmospheric pressure are collected in real time during the semiconductor production process using devices deployed in the production workshop, with a sampling frequency of once per minute. The collected data is then bound to the product's unique identifier in real time. Simultaneously, the collected data undergoes preprocessing to obtain preprocessed data, and the specific preprocessing method is as follows:

[0031] First, missing value detection is performed on the collected environmental data. For data that is missing for no more than 5 minutes, linear interpolation is used to fill it in. If the missing time exceeds 5 minutes, it is marked as an abnormal data segment and the missing duration is recorded.

[0032] Then, outlier identification is performed. By setting normal fluctuation thresholds for each environmental parameter, the data is filtered using the 3σ principle. Data exceeding the threshold range is marked as outliers. The cause of the anomaly is analyzed in conjunction with production logs. If it is caused by sensor failure, the data for that time period is removed. If it is caused by temporary environmental fluctuations, the data is retained and an anomaly is added.

[0033] Finally, the data were standardized to convert the values ​​of each environmental parameter into values ​​within the range of [0, 1], thus eliminating the influence of dimensional differences on subsequent analysis.

[0034] The preprocessed data is classified to obtain structured and unstructured data.

[0035] Step 2: Obtain the categorized structured and unstructured data, identify the field types in both based on predefined rules, classify the identified field types as high-sensitivity fields and low-sensitivity fields, and encrypt the transmission of both structured and unstructured data separately.

[0036] For highly sensitive fields such as wafer batch numbers, key process parameters, yield data, defect coordinates, core equipment parameters, customer order information, and intellectual property-related data, strong encryption must be applied using national commercial cryptography standards (such as the SM4 block cipher algorithm with a key length of 128 bits) or internationally recognized advanced encryption standards (AES-256). For low-sensitivity fields, such as production workshop environmental temperature and humidity records, general equipment operating status indicators, basic attributes of non-critical materials, and publicly available process flow node names, lightweight encryption algorithms can be used to balance security and system performance. The SM7 national cryptographic algorithm or a simplified version of AES is recommended. The encrypted structured data should be transmitted using standard protocols.

[0037] In the data transmission stage, all encrypted structured data must be transmitted through standard transmission protocols, with priority given to secure communication protocols based on the TCP / IP protocol suite, such as TLS 1.3 (Transport Layer Security). This protocol has forward confidentiality features, which can prevent the risk of batch data decryption caused by the leakage of historical session keys. At the same time, it is necessary to combine the specific communication specifications of the semiconductor industry to implement data integrity verification and two-way authentication at the protocol layer. Among them, the data integrity verification uses the HMAC-SHA256 algorithm to ensure that the data is not eavesdropped, tampered with or forged during cross-device transmission. Before transmission, the encrypted data must be standardized and encapsulated, and the field identifiers and encryption algorithm version information must be clearly defined so that the receiving end can accurately decrypt and parse it, ensuring the security and controllability of quality monitoring data throughout the entire link of collection, transmission and storage.

[0038] For highly sensitive fields in unstructured data, such as wafer surface defect images, process flow diagrams, equipment maintenance manuals, and R&D experiment records, a content-based encryption strategy is adopted. First, key feature information, such as defect type, size, and location coordinates, is extracted from the image using image recognition technology. After the feature information is converted into structured description fields, it is processed using the same strong encryption algorithm as structured data. The original image file is encrypted using digital envelope technology. A random symmetric key is generated to encrypt the image file, and then the symmetric key is encrypted with the recipient's public key to form a digital envelope that is transmitted together with the encrypted image.

[0039] For low-sensitivity fields in unstructured data, such as videos of general equipment operation, photos of the workshop environment, and 3D model files of non-core materials, an integrated compression and encryption solution can be adopted. The AES-256 encryption function built into the ZIP or 7Z format can be used to complete the encryption process synchronously during data compression, thereby reducing storage space occupation and transmission bandwidth consumption.

[0040] In the process of unstructured data transmission, a fragmentation transmission mechanism needs to be established. For large files exceeding 100MB, they are divided into fixed-size data packets. Each data packet is given a unique sequence number, checksum, and total number of fragments. The receiving end verifies the integrity of the data packets through the checksum. After receiving all fragments, they are reassembled. At the same time, a breakpoint resume transmission technology is adopted. When the transmission is interrupted, it can continue the transmission from the breakpoint to avoid the waste of resources caused by repeated transmission.

[0041] Step 3: Obtain the preprocessed data after encryption transmission, and perform product quality monitoring and analysis based on the obtained preprocessed data. First, construct an initial quality and safety knowledge graph based on the obtained preprocessed data. The specific construction method is as follows:

[0042] The preprocessed data is converted into a standard triple structure. Each triple must clearly contain a subject, predicate, and object. To ensure data reliability, each triple is accompanied by a source evidence identifier and a confidence score. The confidence score ranges from 0 to 10 and is calculated based on the number of evidence, data consistency verification results, and expert confirmation status. For example, a single piece of unconfirmed evidence is scored 4-6 points, a cross-validation item with two pieces of evidence is scored 7-8 points, and key data reviewed by experts is scored 9-10 points. Source evidence identifiers include 8D report #D2024-087, equipment log #LOG-20241201, quality inspection record #QC-20241128, etc.

[0043] After constructing the triples, based on the pre-defined ontology model, which covers core entity types such as quality risk, production equipment, operating procedures, and testing indicators, node labels and relationship types are defined in the graph database. Indexes are then created for key attributes such as equipment numbers and material codes to improve query efficiency. Finally, basic statistical information for the initial quality and safety knowledge graph is output, including:

[0044] Total number of nodes, such as 52 equipment nodes, 28 process nodes, and 35 defect nodes;

[0045] The number of relationships, such as 78 causal relationships, 43 subordinate relationships, and 61 executive relationships;

[0046] The core business processes covered by the map include raw material inspection process, assembly process control process, finished product quality traceability process, and abnormal event handling process.

[0047] Step 4: Based on the obtained initial quality and safety knowledge graph, firstly, use graph neural networks to learn the features of entity relationships in the graph, extract multi-dimensional feature vectors such as equipment operating status, environmental parameter fluctuations, and material property changes, and construct a risk assessment model;

[0048] This model employs a graph attention network based on an attention mechanism to automatically identify key nodes with high weights affecting product quality, such as the lithography machine temperature control module, wafer cleaning process duration, and photoresist viscosity parameters. Combining historical quality incident data, it sets dynamic thresholds based on the average yield of the past three months ± 3σ. When the risk value calculated by the model from real-time monitoring data exceeds the threshold, the system automatically triggers a three-level early warning mechanism and generates warning information.

[0049] If the risk value exceeds the threshold of 120%, a Level 1 warning is triggered, production on the corresponding production line is immediately suspended, and the risk is pushed to the quality engineer's mobile device. If the threshold is between 100% and 120%, a Level 2 warning is triggered, an email is sent to the process supervisor, and the manual review process is initiated. If the threshold is between 80% and 100%, a Level 3 warning is triggered, a yellow warning is marked in the monitoring system, and potential risk points are recorded.

[0050] The early warning information includes risk source identification, correlation impact analysis, and solutions for similar historical cases.

[0051] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0052] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for tracing the quality and safety of semiconductor products throughout the entire logistics process, characterized in that, The method specifically includes the following steps: Step 1: Assign a globally unique identifier to the semiconductor product, collect production environment parameters in real time and bind them to the identifier, preprocess the collected data to obtain preprocessed data; Step 2: Identify the field types in structured and unstructured data, classify them into high-sensitivity fields and low-sensitivity fields. For high-sensitivity fields, use strong encryption algorithms to encrypt them, and for low-sensitivity fields, use lightweight encryption algorithms to encrypt them. For image files in unstructured data, extract key feature information, convert it into structured description fields, and then encrypt it. The original file is encrypted using digital envelope technology. Step 3: Construct an initial quality and safety knowledge graph based on the preprocessed data after encrypted transmission. Convert the data into triples with accompanying source evidence and confidence scores, and import them into a graph database to generate the graph. Step 4: Use graph neural networks to learn features from the knowledge graph and build a risk assessment model based on the attention mechanism to identify key quality-affecting nodes. Set a dynamic risk threshold based on historical yield data, calculate the real-time risk value through the model, and trigger a graded early warning mechanism when the risk value exceeds the dynamic threshold to generate early warning information containing the location of the risk source and the solution.

2. The semiconductor product logistics end-to-end quality and safety traceability method according to claim 1, characterized in that, The preprocessing includes missing value imputation, outlier identification, and standardization; Missing value imputation specifically involves using linear interpolation to imput data that is missing for no more than 5 minutes consecutively. If the missing time exceeds 5 minutes, it is marked as an abnormal data segment.

3. The semiconductor product logistics end-to-end quality and safety traceability method according to claim 1, characterized in that, The highly sensitive fields include wafer batch number, key process parameters, defect coordinates, customer order information, and intellectual property-related data. The low-sensitivity fields include workshop temperature and humidity records, general equipment status indicators, and publicly disclosed process node names.

4. The semiconductor product logistics end-to-end quality and safety traceability method according to claim 1, characterized in that, Unstructured, highly sensitive data employs a content-aware encryption strategy: First, the defect type, size, and location coordinates are extracted through image recognition, converted into structured fields, and then strongly encrypted. The original file is encrypted using digital envelope technology, that is, the file is encrypted with a random symmetric key, and then the symmetric key is encrypted with the recipient's public key. For unstructured data exceeding 100MB, fragmented transmission is used, with each fragment having a unique sequence number, checksum, and total number of fragments. Resumption of transmission after interruption and reassembly after integrity verification are supported.

5. The semiconductor product logistics end-to-end quality and safety traceability method according to claim 1, characterized in that, The confidence score is determined based on a comprehensive assessment of the amount of evidence, data consistency, and expert confirmation. Single unconfirmed evidence is scored 4–6 points, dual-source cross-validation is scored 7–8 points, and key data reviewed by experts is scored 9–10 points.

6. The semiconductor product logistics end-to-end quality and safety traceability method according to claim 1, characterized in that, The initial quality and safety knowledge graph creates node labels and relationship types in the graph database according to the ontology model, and establishes indexes for equipment numbers and material codes, outputting statistical information including the number of nodes, the number of relationships, and the core business processes covered.

7. The semiconductor product logistics end-to-end quality and safety traceability method according to claim 1, characterized in that, The graph neural network automatically identifies key nodes that significantly affect product quality through attention weights, including the lithography machine temperature control module, wafer cleaning time, and photoresist viscosity parameters.

8. The method for full-process quality and safety traceability of semiconductor product logistics according to claim 1, characterized in that, The tiered early warning mechanism is as follows: a risk value greater than 120% triggers a Level 1 early warning, which suspends the production line and sends a notification to the engineer's mobile device; a risk value between 100% and 120% triggers a Level 2 early warning, which notifies the process supervisor via email and initiates manual review; and a risk value between 80% and 100% triggers a Level 3 early warning, which marks a yellow warning in the monitoring system and records potential risk points.

9. A method for tracing the quality and safety of semiconductor products throughout the logistics process according to claim 1, characterized in that, The globally unique identifier is generated hierarchically according to product batch, wafer number, and chip location, and is encoded using a combination of letters and numbers, and cannot be repeated throughout its entire life cycle.

Images