CN-121998526-A - Intelligent prediction method and prediction system for physical distribution state of AMHS (automated mechanical transmission) in semiconductor wafer manufacturing

CN121998526ACN 121998526 ACN121998526 ACN 121998526ACN-121998526-A

Abstract

The invention provides an AMHS logistics state intelligent prediction method and a prediction system for semiconductor wafer manufacturing, wherein the method comprises the steps of extracting multi-scale time sequence features and modeling layering space relations through an encoder, capturing short-term, medium-term and long-term dynamic features in a time dimension by utilizing a multi-scale attention mechanism, respectively modeling local neighborhood and global dependency relations in the space dimension by adopting a graph attention network GAT and an improved space Transformer in parallel, realizing cross-scale space-time feature fusion by connecting a gating fusion mechanism with residual errors, decoding the fused complex space-time features by utilizing a decoder comprising double-layer convolution decoding and inverse normalization processing, and predicting a logistics state matrix sequence of a plurality of time steps in the future. By deeply coupling the multi-scale space-time features, the problems of space-time feature separation, incomplete multi-scale information capture and difficult modeling of a complex topological structure in the traditional method are effectively solved, and the accuracy and stability of AMHS logistics state prediction are remarkably improved.

Inventors

WU LIHUI
ZHAO WENKAI
HUANG JINGLUN
CHEN ZHE
DONG WANJIAO
YANG FAN

Assignees

上海应用技术大学

Dates

Publication Date: 20260508
Application Date: 20251229

Claims (7)

1. An intelligent prediction method for the physical distribution state of an AMHS for manufacturing a semiconductor wafer is characterized by comprising the following steps: step S100, generating a high-dimensional space-time characteristic representation: Collecting historical logistics data and node network topology relation data of an AMHS (automated mechanical transmission) system of a semiconductor manufacturing factory, preprocessing the historical logistics data and the node network topology relation data, and carrying out space-time feature fusion on the preprocessed data to generate high-dimensional space-time feature representation; Step 200, performing multi-scale space-time feature fusion on the high-dimensional space-time feature representation to obtain fusion space-time features of fusion space features and time features; and step S300, decoding and predicting the fused space-time characteristics to acquire a logistics state sequence of a plurality of time steps in the future.
2. The intelligent prediction method for the physical distribution state of the AMHS for semiconductor wafer fabrication according to claim 1, wherein the multi-scale spatio-temporal feature fusion in step S200 comprises the following steps: Respectively extracting short-term, medium-term and long-term time features in a time dimension through a multi-scale time attention mechanism, and fusing the short-term, medium-term and long-term time features into multi-scale time features; In the space dimension, a graph attention network and a space Transformer are adopted in parallel, the graph attention network is combined with a local multi-head attention mechanism to obtain local space features, the space Transformer is combined with a global multi-head attention mechanism to obtain global space features, and then the local space features and the global space features are fused to form a unified space representation; and fusing the multi-scale time features and the unified space characterization to obtain the fused space-time features.
3. The method for intelligently predicting the physical distribution of an AMHS for semiconductor wafer fabrication according to claim 2, wherein the step of obtaining the multi-scale time features comprises: s201, embedding the time characteristics into corresponding nodes; S202, dividing time sequence data into three time scales of short term, medium term and long term; S203, respectively applying an attention mechanism to each time scale, and calculating self-attention weights in different time scales; s204, fusing the output characteristics of different time scales.
4. The intelligent prediction method for the physical distribution state of the AMHS for semiconductor wafer fabrication according to claim 2, wherein the step of obtaining the unified spatial characterization comprises: S200A, extracting local space features and global space features based on high-dimensional space-time features: S200B, fusing the local spatial features and the global spatial features to form a unified spatial characterization.
5. The intelligent prediction method for the physical distribution state of the AMHS for manufacturing semiconductor wafers according to claim 4, wherein the step S200A extracts the global spatial feature and introduces the spatial position embedding.
6. The intelligent prediction method for the physical distribution state of the semiconductor wafer manufacturing AMHS according to claim 4, wherein the fusion is performed in S200B by a gate-controlled fusion mechanism.
7. An intelligent prediction system for the physical distribution state of a semiconductor wafer manufacturing AMHS, for implementing the intelligent prediction method for the physical distribution state of the semiconductor wafer manufacturing AMHS according to any one of claims 1 to 6, comprising: the system comprises a GUI module, a data preprocessing module, a space-time coupling module and a prediction module; The GUI module is used for receiving historical logistics data and node network topology relation data of the AMHS of the semiconductor manufacturing factory in real time and taking the historical logistics data and the node network topology relation data as original data; the data preprocessing module is used for cleaning, normalizing and constructing time characteristics of the acquired data; The space-time coupling module is used for capturing time dynamic characteristics and space relation characteristics of AMHS logistics state data through multi-scale time sequence characteristic extraction and layered space relation modeling, and carrying out space-time coupling characteristic fusion; The prediction module is used for predicting a logistics state matrix sequence of a plurality of time steps in the future through double-layer convolution decoding and inverse normalization processing, and transmitting the generated prediction data to the GUI module.

Description

Intelligent prediction method and prediction system for physical distribution state of AMHS (automated mechanical transmission) in semiconductor wafer manufacturing Technical Field The invention belongs to the technical field of semiconductor manufacturing, and particularly relates to an AMHS logistics state intelligent prediction method and a prediction system for semiconductor wafer manufacturing. Background In the field of semiconductor manufacturing, automated materials handling systems (Automated MATERIAL HANDLING SYSTEM, AMHS) are a core infrastructure supporting efficient operation of wafer manufacturing, which is responsible for accurately and timely transferring wafer lots among different process equipment, directly affecting the overall throughput, yield and manufacturing cost of a production line. Along with the continuous miniaturization of semiconductor process nodes, the requirements of wafer manufacturing on timeliness and accuracy of material transmission are increasingly improved, and the accurate prediction of the AMHS logistics state becomes a key precondition for realizing intelligent scheduling and optimization of the production process. By predicting the logistics state change trend of the AMHS in advance, the problems of material congestion, equipment idle and the like can be effectively avoided, the resource allocation is optimized, and the overall production efficiency is improved. However, AMHS flow state prediction faces complex technical challenges. On one hand, the physical distribution state data of the AMHS has obvious space-time coupling characteristics, namely, the physical distribution state change of a certain node is not only related to the historical data of the node, but also influenced by the physical distribution state change of surrounding nodes, and meanwhile, the influence relation can dynamically evolve along with time, and on the other hand, the physical distribution state data contains multi-scale time characteristics, has a short-term fluctuation rule and a periodic variation mode. In addition, the physical layout of AMHS forms a complex network topology, and the actual reachability and transmission cost between nodes often depend on the physical connection and layout of transportation rails, rather than simple geometric distances, which makes it difficult to effectively model its spatial relationship by conventional methods based on regular grids or euclidean distance assumptions, further increasing the difficulty of prediction. These characteristics make it difficult for conventional time series prediction methods to achieve the desired prediction accuracy. The existing AMHS logistics state prediction method mainly comprises a traditional statistical method, a machine learning method and a deep learning technology. The traditional statistical method is difficult to capture complex nonlinear relation and space-time coupling characteristics, the machine learning method such as a support vector machine can process nonlinear problems to a certain extent, but has the limitations of calculation efficiency and generalization capability when processing large-scale high-dimensional data, the traditional deep learning technology such as a cyclic neural network (RNN), a long-short-term memory network (LSTM) and the like has advanced to a certain extent in time sequence prediction, but has the defects when processing multi-scale space-time characteristics and complex network topological structures, and is difficult to fully mine complex space-time correlation information in AMHS logistics state data, so that the prediction precision is difficult to meet the high precision requirement of semiconductor manufacturing. Therefore, developing an AMHS logistics state intelligent prediction method and system for semiconductor wafer manufacturing has important significance for improving the intelligent level of semiconductor manufacturing. Disclosure of Invention The invention aims to provide an intelligent prediction method and a prediction system for the physical distribution state of an AMHS (automatic mechanical system) manufactured by a semiconductor wafer, which are used for solving the problems of space-time characteristic decoupling, insufficient multi-scale characteristic capturing, difficult spatial relation modeling and the like in the physical distribution state prediction of the semiconductor AMHS in the prior art and realizing the accurate prediction of the physical distribution state of the AMHS. The technical scheme adopted is as follows: An intelligent prediction method for the physical distribution state of an AMHS for manufacturing a semiconductor wafer comprises the following steps: step S100, generating a high-dimensional space-time characteristic representation: Collecting historical logistics data and node network topology relation data of an AMHS (automated mechanical transmission) system of a semiconductor manufacturing factory, preprocessing the historic