AI prediction system for optical properties

An AI prediction system addresses the issue of cumulative errors in optical integrated circuit manufacturing by using a feedforward approach to measure and predict optical properties, enhancing manufacturing accuracy and efficiency.

JP7590671B2Active Publication Date: 2024-11-27NIPPON TELEGRAPH & TELEPHONE CORP
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
JP2023501987
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2021-02-26
Publication Date
2024-11-27
Estimated Expiration
2041-02-26

AI Technical Summary

Technical Problem

Existing optical integrated circuit manufacturing processes struggle with inaccurate reflection of inspection results due to cumulative errors, leading to reduced accuracy and efficiency in manufacturing optical devices.

Method used

An artificial intelligence prediction system that utilizes a feedforward approach to measure and predict optical properties during the manufacturing process, adjusting manufacturing conditions in real-time based on immediate data from previous steps to improve accuracy and throughput.

Benefits of technology

Enhances the quality and efficiency of optical integrated circuit manufacturing by accurately reflecting inspection results and reducing manufacturing errors, thereby improving the precision and speed of the production process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007590671000003
    Figure 0007590671000003
  • Figure 0007590671000004
    Figure 0007590671000004
  • Figure 0007590671000005
    Figure 0007590671000005
Patent Text Reader

Abstract

Disclosed are a production system, a production method, a configuration for an optical circuit that improves the production method, and the like that make it possible for inspection results / data that can be acquired during a production process for an optical circuit to be more accurately and efficiently reflected in the process. This artificial intelligence prediction system for an optical integrated circuit comprises a central processing device that executes processing that receives input data, processing that performs learning and inference on the basis of the input data, and processing that generates prediction data for optical characteristics. The input data is process data that is acquired during a production process for the optical integrated circuit and includes at least in-wafer film thickness distribution and refractive index distribution data for a core layer. The input data may also include in-wafer film thickness distribution and refractive index distribution data for a lower cladding layer and in-wafer film thickness distribution and refractive index distribution data for an upper cladding layer.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to the fabrication of optical circuits, and more particularly to an artificial intelligence prediction system for optical properties. [Background technology]

[0002] Semiconductor lasers, photodiodes, optical wavelength multiplexers / demultiplexers, optical switches, etc. are key optical devices in optical fiber communications. Semiconductor lasers, as optical oscillators, generate light waves that serve as carriers for superimposing information signals. Photodiodes operate as elements that convert the intensity of optical signals into electrical signals. Optical wavelength multiplexers / demultiplexers, as typified by arrayed waveguide gratings, are used in wavelength division multiplexing communications as elements that multiplex / demultiplex different optical wavelengths (see, for example, Non-Patent Document 2). Furthermore, optical switches have an important function in ROADM (Reconfigurable Optical Add / Drop Multiplexing) systems as elements that route optical paths. These optical devices can be configured as optical integrated circuits. In optical fiber communications, not only optical fiber, which is a transmission medium, but also optical integrated circuits of these optical devices that process optical signals play an important role (see, for example, Non-Patent Document 1).

[0003] Optical integrated circuits are generally composed of optical waveguides formed on a substrate. An optical waveguide consists of a core through which an optical signal propagates and a cladding that surrounds it. Semiconductor lasers and photodiodes are made of semiconductor materials such as InP, while arrayed waveguide gratings and optical switches are mainly made of optical waveguide materials made of quartz glass.

[0004] FIG. 1 is a diagram showing a conventional method for manufacturing an optical waveguide. Here, a typical manufacturing process will be described using a silica-based planar lightwave circuit made of silica-based glass as an example. First, in 1: lower clad deposition process, a glass film that will become the lower clad 12 is deposited on a silicon substrate (wafer) 11. The lower clad 12 is made of SiO2 to which P2O5 and B2O3 have been added, which is deposited by flame hydrolysis deposition (FHD). The soot-like glass particles deposited by the FHD method are heated at a high temperature of 1000°C or higher to obtain a transparent lower clad 12. Next, in 2: core deposition process, a thin-film glass that will become the core 13, which has a higher refractive index than the lower clad 12, is deposited by the same FHD method. When depositing the core 13, a desired refractive index value can be obtained by adding GeO2 to SiO2. As in 1: lower clad deposition process, the transparent core 13 is formed by heating at a high temperature of 1000°C or higher.

[0005] In step 3: photoresist film formation, a photoresist film 14 is formed on the substrate by spin coating. Next, in step 4: circuit pattern exposure, the photoresist film is irradiated with UV light 16 through a photomask 15 to expose a circuit pattern according to the mask pattern. Then, in step 5: photoresist development, the circuit pattern of the photoresist film is developed to obtain a photoresist pattern 17.

[0006] Next, in 6: Etching step, the photoresist pattern 17 is transferred to the core by reactive ion etching (RIE) to obtain a core pattern 18. Furthermore, in 7: Resist Removal step, the photoresist remaining on the core is removed by ashing. Finally, in 8: Upper Clad Deposition Step, an upper clad 19 is deposited by the same method as the lower clad deposition in the above-mentioned 1: Lower Clad Deposition Step.

[0007] Various inspections are performed on the optical waveguide obtained by the above manufacturing process, such as the pattern size and optical characteristics of the optical circuit. Conventionally, when reflecting these inspection results in the manufacturing process, manufacturing conditions that reflect the inspection results were set for each process after all of the series of processes shown in Figure 1 above were completed. With this method of reflecting the inspection results, manufacturing errors accumulate in each process. Due to this accumulated error, the accuracy of the inspection results obtained in each process decreases the later the process shown in Figure 1, and there was a problem that the reflection of the inspection results became inaccurate. It is also possible to suppress the accumulation of manufacturing errors by resetting the manufacturing conditions for the current process based on the inspection results obtained at the time of the completion of one process. However, with this reflection method, there was a problem that the characteristics of the device during manufacturing could not be corrected. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] T. Miya, "Silica-based planar lightwave circuits: passive and thermally active devices," in IEEE Journal of Selected Topics in Quantum Electronics, vol. 6, no. 1, pp. 38-45, Jan.-Feb. 2000, doi: 10.1109 / 2944.826871 [Non-Patent Document 2] Inoue, Y.; Kaneko, A.; Hanawa, F.; Takahashi, H.; Hattori, K.; Sumida, S.: 'Athermal silica-based arrayed-waveguide grating multiplexer', Electronics Letters, 1997, 33, (23), p. 1945-1947, DOI: 10.1049 / el:19971317 [Non-Patent Document 3] Marcatili, EAJ: “Dielectric Rectangular Waveguide and Directional Coupler for Integrated Optics”, 1969 Bell Syst. Tech. J., 48, p.2071-2102 [Non-Patent Document 4] Fundamentals of Optical Waveguides 2nd edition (2010 / 8 / 4), P. 425) [Non-Patent Document 5] A. Himeno, K. Kato and T. Miya, "Silica-based planar lightwave circuits," in IEEE Journal of Selected Topics in Quantum Electronics, vol. 4, no. 6, pp. 913-924, Nov.-Dec. 1998, doi: 10.1109 / 2944.736076. [Non-Patent Document 6] Hida, Y.; Takato, N.; Jinguji, K.: 'Wavelength division multiplexer with wide passband and stopband for 1.3μm / 1.55μm using silica-based planar lightwave circuit', Electronics Letters, 1995, 31, (16), p. 1377-1379, DOI: 10.1049 / el: 19950927 Summary of the Invention [Problem to be solved by the invention]

[0009] There is a need for a prediction system that can effectively and efficiently reflect the inspection results of optical integrated circuits in the manufacturing process. [Means for solving the problem]

[0010] One embodiment of the present invention is an artificial intelligence prediction system for an optical integrated circuit, comprising a central processing unit that performs a process of receiving input data, a process of learning and inference based on the input data, and a process of generating predicted data for optical characteristics, wherein the input data is process data obtained during a manufacturing process of the optical integrated circuit, and includes at least film thickness distribution and refractive index distribution data within a wafer of a core layer. Effect of the Invention

[0011] The quality and throughput of the manufacturing process of optical integrated circuits can be improved. [Brief description of the drawings]

[0012] [Figure 1] 1A to 1C are diagrams illustrating a series of steps in a typical method for manufacturing an optical waveguide according to the prior art. [Diagram 2] 1A to 1C are diagrams illustrating an outline of a manufacturing method for the optical circuit according to the first embodiment. [Diagram 3] FIG. 2 is a diagram of a generalized feedforward system of the first embodiment. [Figure 4] FIG. 13 is a diagram illustrating an AI prediction system and related data according to the second embodiment. [Diagram 5] FIG. 11 is a diagram for explaining feature amounts acquired by the AI ​​prediction system of embodiment 2. [Figure 6] FIG. 11 is a diagram showing an example of image analysis of an optical waveguide pattern according to the second embodiment; [Figure 7] FIG. 1 is a diagram explaining improvement in prediction accuracy through learning of an AI prediction system. [Figure 8] FIG. 13 is a diagram illustrating the basic configuration of a marker according to a third embodiment. [Figure 9] FIG. 13 is a diagram for explaining another marker according to the third embodiment. [Figure 10] 13 is a flowchart of correction of a manufacturing process using a marker of embodiment 3. [Figure 11] FIG. 13 is a diagram illustrating an example of prediction using markers in the third embodiment. [Figure 12]FIG. 13 is a diagram showing a marker according to a second example of the third embodiment. [Figure 13] FIG. 13 is a diagram showing a marker according to a third example of the third embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] The inventors have reconsidered from a new perspective how to use information data obtained from inspection results in the manufacturing process of optical circuits using conventional techniques. In the series of manufacturing methods of the conventional techniques shown in Figure 1, after obtaining the inspection results in the final stage, information based on the inspection results was fed back to the preceding process. As a result, the inspection result values ​​such as optical characteristics obtained in the final stage contained accumulated errors in each process. In addition, in terms of process improvement through feedback, the process conditions were changed and adjusted based on information containing accumulated errors, including to processes that were not directly related. It could not be said that this was sufficiently effective in terms of the amount and accuracy of feedback, and the destination and timing of feedback.

[0014] However, if information on the processing results obtained in a certain process, for example, information on the resist pattern width obtained in a photolithography development process, can be known immediately after the development process, the information on the pattern width can be reflected in the subsequent etching process. Also, if the film thickness and refractive index of the core obtained in the core deposition process can be known immediately after the deposition process, the optical characteristics of the optical waveguide formed in the subsequent photolithography development process and etching process can be predicted. In this way, by acquiring the characteristic values ​​of the optical waveguide components obtained in the preceding pre-process during or immediately after the pre-process, they can be reflected in the processing conditions of the subsequent post-process and used to predict the optical characteristics obtained in the subsequent process.

[0015] In the series of inventions disclosed herein, process information and process data obtained in a plurality of manufacturing processes are "feed forward" to the next process during the process in which the information and data are obtained. Improvement of the manufacturing process by such feed forward is also advantageous in terms of throughput of the manufacturing process.

[0016] The following embodiments disclose a manufacturing system and manufacturing method that can more efficiently and accurately reflect the inspection results and data obtained in the manufacturing process of an optical circuit in the process, and an optical circuit configuration that improves the manufacturing method. In particular, this specification will focus on an artificial intelligence (AI) prediction system that predicts the optical characteristics of an optical circuit. First, the basic concept of the manufacturing method of the optical circuit disclosed herein will be explained with reference to the drawings.

[0017] [Embodiment 1] In the method for manufacturing an optical circuit according to the present embodiment, attention is focused on data on the components or characteristics of an optical device that is formed and acquired during one step of the manufacturing process. This data can also be called real data or current data. Measurements are performed on the components or characteristics of the optical device, and manufacturing conditions for subsequent steps are adjusted or corrected based on the data obtained from the measurements (hereinafter, this method is also called a "feedforward system"). The feedforward system suppresses variations in the optical characteristics of the optical device, making it possible to obtain desired optical characteristics for the final optical device.

[0018] 2 is a diagram showing a manufacturing method of an optical circuit including an optical waveguide according to the first embodiment of the present disclosure. The feedforward system performs "measurement" of components of an optical device formed in one process, and performs "optical characteristic estimation" based on the measurement results by optical characteristic estimation process 21. Then, based on the estimation results, "control" of the post-process is performed by process control process 22.

[0019] For example, the refractive index and thickness of the lower cladding film formed in the lower cladding deposition step 1 in Fig. 2, and the refractive index and thickness of the core layer deposited in the core deposition step 2 are "measured". Based on the measurement results, the final optical characteristics of the device fabricated with the standard (nominal) design values ​​are estimated. Then, based on this estimation, the etching strength or etching time is "controlled" in the subsequent etching step 6.

[0020] Specifically, based on the "measured" thickness and refractive index of the core layer and the refractive index of the cladding film, the ideal core width of the pattern to satisfy the performance required for the optical device is estimated (predicted). Then, in the etching process 6, etching is performed based on this predicted value. For example, if the predicted information indicates that the "waveguide width after core processing is too wide" with the standard (nominal) design value and the desired performance cannot be satisfied, a correction is made in the etching process to narrow the core width. The adjustment (correction) method at this time is to shorten / lengthen the etching time or weaken / strengthen the etching strength to make the core width wider / narrower. Furthermore, the core width and step in the waveguide pattern formed in the etching process 6 can be "measured". Based on the measurement results, the refractive index of the upper cladding film formed in the upper cladding deposition process 8 can be "controlled" to adjust the optical properties of the optical waveguide finally obtained.

[0021] As described above, the feedforward system of the present embodiment measures the shape, characteristics, etc. of components of a formed optical device during or after a front-end process among multiple processes for manufacturing an optical device, and adjusts or corrects the manufacturing conditions in the back-end process based on the measurement results.

[0022] FIG. 3 is a diagram showing a generalization of the feedforward system of Embodiment 1. The feedforward system includes a manufacturing procedure of an optical device consisting of M steps, and the optical device to be manufactured is carried out in the order of Step 1, Step 2 ···, Step i, ··· Step j, ··· Step M. Here, when i < j, Step j is a step temporally subsequent to Step i. The feedforward system includes a measurement data processing unit 31 and a control data processing unit 32. The measurement data processing unit 31 executes the optical property estimation process 21 described in FIG. 2, and the control data processing unit 32 executes the process control process 22 described in FIG. 2. The measurement data processing unit 31 and the control data processing unit 32 can be in the form of a computer configured to include a CPU, a RAM, a ROM, and the like.

[0023] In the feedforward system of FIG. 3, the solid line indicates the flow according to the steps of the object to be manufactured. The dashed line indicates the measurement data obtained by the "measurement" of each step, and the dash-dotted line indicates the control data for the "control" of each step, respectively. As shown in FIG. 3, in Step i of the feedforward system of the present embodiment, measurement data is acquired from the manufacturing apparatus or the measurement apparatus of that step and transferred to the measurement data processing unit 31. The measurement data processing unit 31 predicts the shape or properties of the components of the optical device formed in Step i based on the transferred measurement data. Also, based on the measurement data, the optical properties of the finally obtained optical device can be predicted in Step i.

[0024] The predicted value derived by the measurement data processing unit 31 is passed to the control data processing unit 32. The control data processing unit 32 determines the manufacturing conditions for process j, which is a subsequent process, based on the predicted value. When process j is performed, the control data processing unit 32 supplies control data for process j to be set in the manufacturing equipment according to the determined manufacturing conditions. The control data based on the previous process supplied when performing the subsequent process j may be only control data based on the previous process i, or may be multiple types of control data based on several of the previous processes. The form of the control data is naturally determined according to conditions such as the actually configured manufacturing equipment and the object to be manufactured.

[0025] [Embodiment 2] In this embodiment, an artificial intelligence (AI) prediction system for optical characteristics that can be used in the feedforward system of the above-mentioned embodiment 1 will be described. As described in embodiment 1, the feedforward system of Fig. 3 acquires measurement data from the manufacturing equipment or its measurement equipment in each process and transfers it to a measurement data processing unit 31. The measurement data processing unit 31 updates the manufacturing conditions and predicts the optical characteristics for updating, and feeds forward the data to the manufacturing process. In this embodiment, an AI prediction system that predicts the optical characteristics of an optical circuit including an optical waveguide is disclosed.

[0026] AI systems are already being used in a wide range of fields, from daily life to industry, by utilizing machine learning, deep learning and big data, and are expected to be increasingly used in manufacturing lines that include optical integrated circuits. In optical integrated circuits that are composed of optical waveguides, the input and output of optical signals via optical fibers is required for evaluating the optical characteristics. Measurements via optical fibers at the wafer level are costly and time-consuming. If the optical characteristics of optical devices can be predicted from process data during the manufacturing process, for example, it would be possible to select good products during the process. The selection of good products includes selecting defective products within a wafer and selecting wafers by lot, which can reduce unnecessary inspection and testing costs for defective products.

[0027] In the feedforward system in the manufacturing method of the above-mentioned embodiment 1, it is necessary to predict what the optical characteristics of the optical device will be if the process proceeds under the same process conditions, based on the information data obtained in each process. At least a part of the optical characteristics estimation process 21 in Fig. 2 corresponds to the process in the AI ​​prediction system of this embodiment, and at least a part of the measurement data processing unit 31 in Fig. 3 corresponds to the AI ​​prediction system of this embodiment.

[0028] 4 is a diagram illustrating an AI prediction system 1000 for optical properties and related data according to the present disclosure. In the AI ​​prediction system 1000 according to the present disclosure, process data 1001 acquired from a manufacturing process 1004 is input as learning data. For training to improve the prediction accuracy of the AI ​​system 1000, a data set 1003 combining past process data and actual measured values ​​of corresponding device optical properties is also input as teacher data. The AI ​​prediction system 1000 outputs predicted value data and predicted value 1002 of optical properties as a result of inference.

[0029] The AI ​​prediction system 1000 in Fig. 4 includes at least a central processing unit (CPU), and the CPU performs at least a process of receiving input data, a process of learning and inference based on the input data, and a process of generating predicted data of optical properties. The input process data 1001 can be acquired from the manufacturing process 1004 and input sequentially, or can be temporarily stored in a memory device not shown in Fig. 4 and input to the CPU at an appropriate time. In addition, teacher data, which will be described later, can also be stored in this memory device. Furthermore, the learning, inference, or generation of predicted data of optical properties may be performed by a graphics processing unit (GPU).

[0030] The multiple manufacturing processes 1004 in Fig. 4 correspond to process 1 to process M shown in Fig. 3, and "current process data" is input from each process to the AI ​​prediction system 1000. Fig. 4 shows M processes, and current process data is input from each process to the AI ​​prediction system 1000, but it is not necessary for process data to be supplied from all processes. Here, "current" means data acquired while each process is in operation or before the final process is completed, and it should be noted that the process data input does not strictly require real-time performance.

[0031] The teacher data 1003 is a data set that combines past process data acquired from a plurality of processes 1004 and actual measurement values ​​of device optical characteristics corresponding to the past process data. In the AI ​​prediction system 1000, the teacher data 1003 is used in a "learning phase" to improve the accuracy of inference, and is used to optimize the weighting values ​​of the synapses of the inference model in the AI ​​prediction system 1000. The teacher data set 1003 is past process data, and is supplied to the AI ​​prediction system 1000 separately from the current process data 1001. The teacher data can be accumulated in a memory provided in the AI ​​prediction system 1000. Data stored in an external memory outside the AI ​​prediction system 1000 can also be transmitted to the AI ​​prediction system 1000 as necessary.

[0032] For example, a multi-layer neural network, ridge regression, or Lasso regression can be used as a prediction algorithm in the AI ​​prediction system of the present disclosure. There is no limitation on the prediction algorithm as long as it is available in a general AI prediction system. Meanwhile, in the AI ​​prediction system of the present disclosure, the inventors have found an optimal feature for the manufacturing process of an optical integrated circuit including an optical waveguide as input data. In general, in an AI prediction system, in order to improve the accuracy of an inference model by learning, it is important to select an appropriate feature as input data. It is clear that if an incorrect variable (feature) is selected as input data in machine learning or deep learning, the prediction accuracy will deteriorate. Furthermore, if an unnecessary feature is selected as input data, the feature acts as noise in machine learning or deep learning, and the learning time will be extended. Furthermore, the selection of an inappropriate feature can lead to overlearning, which can reduce the prediction accuracy for the new state of the manufacturing process.

[0033] [Important parameters that affect the prediction accuracy of AI prediction systems] From the viewpoint of the manufacturing process of an optical integrated circuit, highly accurate prediction of optical characteristics is realized by identifying parameters that essentially affect the optical characteristics of an optical device and the optical characteristics to be predicted, determining appropriate feature values, and having the AI ​​prediction system learn the parameters. It is also important to select process data suitable for the manufacturing process of an optical integrated circuit, taking into consideration the realistic conditions in the actual manufacturing process and the throughput of the entire process. The inventors have carefully considered the functions of the components of an optical integrated circuit and the essential features of its manufacturing process, and have examined what feature values ​​improve the accuracy and efficiency of the manufacturing process of an optical integrated circuit. Below, in the AI ​​prediction system of the present disclosure, (1) the parameters of an optical device that essentially affect the optical characteristics to be predicted, and (2) the feature values ​​corresponding to these parameters and suitable for predicting optical characteristics with high accuracy will be described in detail.

[0034] As mentioned at the beginning, optical integrated circuits that integrate key optical devices contain optical waveguides as essential components. Optical devices can also be called optical waveguide devices. Therefore, it is effective to predict the structure and optical characteristics of optical waveguides with high accuracy to suppress manufacturing errors. Many important and large-scale optical integrated circuits also contain Mach-Zehnder interferometers (MZIs) and arrayed waveguide gratings (AWGs). The functions of MZIs and AWGs are based on the interference phenomenon that utilizes the optical phase difference caused by the path of one or more optical waveguides that the light passes through. The phase of an optical signal propagating through an optical waveguide is directly affected by the propagation characteristics of the optical waveguide, and these propagation characteristics are expressed as the equivalent refractive index n e It can be said that it is determined by

[0035] An optical signal with wavelength λ has length L and equivalent refractive index n e When an optical signal propagates through an optical waveguide, the phase lead φ of the optical signal is expressed by the following equation.

number

[0036] When the cross-sectional shape of the optical waveguide is rectangular, the equivalent refractive index of the optical waveguide is n e is determined by the width and height of the optical waveguide core, the refractive index of the core, and the refractive index of the cladding. For example, Non-Patent Documents 3 and 4 state that for a rectangular core, the core refractive index n core , cladding refractive index n clad Using the dispersion equations for the x and y components of the electric field, the propagation constant β and the equivalent refractive index n e A method for determining is disclosed.

[0037] As described above, in optical integrated circuits that utilize the optical interference phenomenon, such as AWG and MZI, it is found that highly accurate evaluation of the width and height of the optical waveguide core, the refractive index of the core, and the refractive index of the cladding is also important for predicting optical characteristics. In the AI ​​prediction system for optical characteristics of optical circuits disclosed herein, the physical configuration (thickness, width) of the optical waveguide core and the equivalent refractive index of the core are determined as the parameters that have the most significant effect on the prediction of optical characteristics, and the feature quantities corresponding to these parameters are further determined.

[0038] [Features of AI prediction system] FIG. 5 is a diagram for explaining the feature amount acquired in the AI ​​prediction system of the present disclosure. FIG. 5 provides an overview of the feature amount acquired from each process in a feedforward system implementing the optical circuit manufacturing method of the embodiment 1 shown in FIG. 2. In FIG. 5, only an arrow labeled "measurement" from 2: core deposition process is shown for the optical property estimation process 21 including the AI ​​prediction system 1000. The feature amount acquired by the AI ​​prediction system 1000 is different in each process, Feature amount 1 to Feature amount 6. As described later, Feature amount 6 related to particle / defect analysis can be acquired in all processes as shown by the thick arrow 1014 and input to the AI ​​prediction system 1000.

[0039] (1) Thickness distribution and refractive index distribution of core and cladding As described above, the width and height of the core, the refractive index of the core, and the refractive index of the cladding are the most important parameters in predicting optical characteristics. Naturally, at the stage of depositing each layer of the optical integrated circuit, the film thickness distribution and the refractive index distribution become the feature quantities of the AI ​​prediction system. Specifically, the lower cladding refractive index distribution obtained in the lower cladding deposition process, indicated by the arrow 1010, 1, becomes the first important feature quantity (feature quantity 1). It is preferable to obtain the in-plane distribution of the lower cladding refractive index of the wafer.

[0040] Next, the core film thickness distribution and core refractive index distribution obtained in 2: Core Deposition Process indicated by arrow 1011 become feature quantities (feature quantity 2). As already mentioned, the shape and refractive index of the core have a strong influence on the optical characteristics. The core height corresponds to the core film thickness and refractive index in 2: Core Deposition Process. It is preferable to obtain the in-plane distribution of the wafer for the core film thickness and refractive index as well.

[0041] Furthermore, the upper cladding refractive index distribution obtained in the upper cladding deposition step 8 indicated by the arrow 1013 becomes a feature (feature 3).

[0042] In the above-mentioned feature quantities 1 to 3, the film thickness distribution and refractive index distribution data in the wafer of the core layer are the most important feature quantities. This is because the structure of the core layer directly affects the propagation characteristics of the optical waveguide. Since the light mainly passes through the core, the influence of the film thickness distribution and refractive index distribution of the cladding is relatively small compared to the core. Therefore, in the AI ​​prediction system disclosed herein, the film thickness distribution and refractive index distribution data in the wafer of the core layer are important for improving the accuracy of the inference model. The refractive index of the cladding can be substituted with the refractive index of either the lower cladding or the upper cladding, or with the average value or the design value. In addition, the above-mentioned core layer distribution includes distribution information of the partial or entire waveguide width in the chip, or partial or entire waveguide image information in the chip, as described later.

[0043] The thickness distribution and refractive index distribution of each of the above-mentioned layers are preferably measured in a non-contact manner on the multilayer film at the location where the optical waveguide is formed. That is, when measuring the core, the lower cladding film is separated and measured in a non-contact manner. This non-contact measurement can be performed by a non-contact evaluation of the thickness and refractive index of the multilayer film by light reflection spectrum analysis. According to this measurement method, the lower oxide film is measured before the deposition of the core film, and the lower oxide film is fixed and analyzed when the core film is measured, so that the characteristics of the core film can be evaluated with high accuracy. Similarly, when measuring the upper cladding film, the core film is separated and measured in a non-contact manner. In addition, it is preferable to obtain the temperature at which the thickness and refractive index of each of the above-mentioned films are measured as a feature value. This is because the thickness and refractive index are also significantly affected by the measurement temperature.

[0044] (2) Waveguide pattern measurement results / images Among the physical configurations of the core, the core width needs to be observed in the completed state of the core pattern after etching. Therefore, the waveguide measurement results and image analysis results obtained after the etching process is completed (feature amount 4) indicated by arrow 1012 6. Specifically, the information corresponding to the core width can be learned by obtaining the results of measuring the waveguide structure at a representative point in the chip using a microscope or the like, or the entire or partial waveguide image of a specific optical waveguide, and using them as features. As a preprocessing step for training the AI ​​prediction system 1000, the image data is converted into information that detects only the sidewalls of the core, so that information on the core width and its distribution can be extracted and noise can be removed.

[0045] Figure 6 shows an example of image analysis of an optical waveguide pattern. The left side shows a microscope photograph of the core part after etching is completed, and the right side shows the sidewall of the optical waveguide extracted using image analysis technology. As a preprocessing step for training AI on the image data of the microscope photograph of the core part in Figure 6, it is possible to convert it into information that detects only the sidewall of the core, thereby removing information other than information related to the core width (and its distribution), i.e., information that becomes noise. In the extracted sidewall view, the waveguide width is about 80 pixels wide, and fluctuations in the core width can be detected with a resolution of at least about 1.25%.

[0046] (3) Element structure collection product Marker Image In addition to directly acquiring data on the core width of the specific optical waveguide described above, the chip can be provided with an element structure integrated marker in which the waveguide elements used on the chip of the optical integrated circuit are integrated within a certain area. In addition to individually measuring each of the important structural elements distributed on the chip, or instead of individually measuring each of the important structural elements, the element structure integrated marker can be imaged with a microscope or the like to acquire the feature quantity to be input to the AI ​​prediction system 1000.

[0047] Referring again to FIG. 5, the marker analysis result obtained after the etching process is completed (feature amount 5) is indicated by arrow 1012 6. It is practically difficult to capture an image of the entire optical waveguide with high resolution in order to obtain processing information of the optical waveguide. Alternatively, feature amounts can be obtained using element structure integrated markers that reflect and represent important optical waveguide elements, which also helps to improve the throughput of the manufacturing process. Structural elements included in the element structure integrated marker include bends, tapers, intersections, and straight lines. Details of the element structure integrated marker will be described later as embodiment 3.

[0048] (4) Images of foreign objects and defects If foreign matter or defects are found on a wafer during the manufacturing process of an optical waveguide device, the shape of the optical waveguide may be distorted or disconnected or cause reflection. Such foreign matter or defects can cause defects and failures in optical integrated circuits. Failure modes caused by foreign matter or defects in optical integrated circuits occur when an optical waveguide through which an optical signal passes is formed at the location of the foreign matter or defect. If an optical waveguide overlaps with a foreign matter or defect, the foreign matter or defect location will cause an increase in propagation loss, reflection, and interruption of the optical signal.

[0049] On the other hand, foreign matter and defects in places where optical waveguides are not formed do not cause defects. When the optical waveguide is away from foreign matter or defects, the optical signal is confined in the optical waveguide, and foreign matter and defects do not affect the optical signal. Therefore, in order to consider foreign matter and defects that affect the quality of optical integrated circuits, the position and shape of the foreign matter or defect on the wafer are used as features to input microscope images, etc. into machine learning. Referring again to FIG. 5, the foreign matter and defect analysis results that can be obtained at any process indicated by arrow 1014 become features (feature 6). Since foreign matter and defects can occur in all processes, this feature can be obtained at any process.

[0050] There are two possible formats for data to be input as features to the AI ​​prediction system 1000 of the present disclosure. One format is the image data of the optical waveguide section itself, or data obtained by converting each pixel of the image into grayscale or the like. The other format is a calculated value using a function that evaluates the overlap of the optical waveguide with foreign bodies or defects. The imaging and evaluation of foreign bodies or defects can be weighted according to their position on the optical waveguide, and only those parts that are important in terms of the function of the optical integrated circuit can be evaluated. For example, weights can be given to the phase adjustment section of the MZI or the phase adjustment section of the DPOH (Dual Polarization Optical Hybrid).

[0051] Therefore, the AI ​​prediction system disclosed herein is an artificial intelligence prediction system for optical integrated circuits, and includes a central processing unit that performs processes of receiving input data, learning and inference based on the input data, and generating predicted data for optical properties, and the input data is process data obtained during the manufacturing process of the optical integrated circuit, and can be implemented as at least film thickness distribution and refractive index distribution data within a wafer of a core layer.

[0052] Next, we will show the improvement in prediction accuracy when the AI ​​prediction system disclosed herein is applied to the central wavelength of an AWG as an example of optical properties. EXAMPLES

[0053] In an AWG, the phase difference caused by the difference in optical path length of each arrayed waveguide produces an optical multiplexing / demultiplexing effect. By inputting a wavelength-multiplexed optical signal containing multiple optical signals of different wavelengths, it is possible to separate and output the optical signals of each wavelength to different ports. The central wavelength of each output port in an AWG is determined by the phase difference caused by the difference in optical path length of the arrayed waveguide. The difference in optical path length in the arrayed waveguide of an AWG is determined by the refractive index of the cores and claddings of the optical waveguides that make up the arrayed waveguide, and the shape of the cores.

[0054] Using the feature quantities explained in Fig. 5, the machine learning of the AI ​​prediction system was processed as a regression problem to investigate the variation in the central wavelength of one transmission band of the AWG. In this example, the predicted characteristic in the AI ​​prediction system is the "central wavelength of the AWG."

[0055] Referring to Non-Patent Document 5, the theoretical formula for the central wavelength λ0 in an AWG is expressed by the following formula:

number

[0056] where n e is the equivalent refractive index of the core, ΔL is the physical length difference of the waveguide between adjacent arms, and m is the diffraction order. The features used for inference were the core width, core refractive index, core thickness, and chip coordinates in the wafer of the arrayed waveguide.

[0057] Fig. 7 is a diagram for explaining the improvement of prediction accuracy by learning the AI ​​prediction system of the present disclosure. Fig. 7(a) shows the variation of the central wavelength at a given central wavelength in a state where the AI ​​prediction system has not performed learning for prediction. Here, the variation is the difference between the design value (theoretical value) and the actual measured value of the central wavelength. The horizontal axis shows the central wavelength difference value (nm), and the vertical axis shows the frequency, and is shown as a histogram of the variation of the actual measured value of the AWG central wavelength (1299 nm) for 600 chips.

[0058] On the other hand, FIG. 7(b) shows the correlation between the predicted value and the actual value after AI learning was performed using a data set consisting of the above-mentioned feature amount and the actual measured value of the AWG central wavelength as training data in the AI ​​prediction system of the present disclosure. The horizontal axis shows the actual measured value of the AWG central wavelength, and the vertical axis shows the predicted value of the AWG central wavelength by the AI ​​system in nm. Each point on the graph represents one sample of a different AWG chip. The number of data sets of AWG central wavelengths used for learning was about 1200 chips, which was different from the approximately 600 chip sets shown in FIG. 7(a).

[0059] Compared with (a) of FIG. 7, it can be seen that the variation in the difference between the predicted value (design value) and the actual value before and after learning has converged to a smaller range. According to (b) of FIG. 7, a clear correlation can be confirmed between the predicted value and the actual value of the central wavelength of the AWG. That is, in the AI ​​prediction system of optical characteristics disclosed herein, it has been confirmed that the spread of the difference between the predicted value and the actual value of the central wavelength of the AWG can be reduced to about half (approximately 0.1 nm in standard deviation) by learning the optical characteristic value of the central wavelength of the AWG using teacher data, thereby improving the prediction accuracy. It can be confirmed that the prediction accuracy of the central wavelength is significantly improved by simply using the core width, core refractive index, core thickness, and chip coordinates in the wafer as features for the arrayed waveguide of the AWG.

[0060] By changing the teaching data, it is also possible to predict the characteristics of the transmission spectrum. Specifically, it is possible to predict the 3 dB bandwidth of the transmission spectrum and the bandwidth from the peak transmittance to a specified amount of attenuation.

[0061] In the above-mentioned embodiment, the effect of the AI ​​prediction system of the present disclosure was shown by taking the central wavelength of the AWG as an example of the optical characteristic to be predicted in the optical integrated circuit. The optical characteristic to be predicted is not limited to this, and can be, for example, the optical characteristic to predict the central wavelength of the MZI. In the MZI, the output port changes due to the phase difference caused by the optical path length difference of the two arm waveguides. It can also be used as a switch by heating the arm. Here, if the optical path length difference of the two arm waveguides deviates from the design value, the extinction wavelength changes. The optical path length difference in the MZI is determined by the refractive index of the core and cladding that form the optical waveguide, and the shape of the core (see Non-Patent Document 5).

[0062] Non-Patent Document 5 describes the structure and operation principle of a two-input, two-output MZI switch. In an MZI switch, the optical path length difference between the two arm waveguides changes the interference state in the directional coupler, and the output intensity changes. If the branching ratio of the directional coupler is 50%, when the optical path length difference between the two arms is 0±2nπ in phase conversion, all the light input to input port 1 is output from output port 2. On the other hand, when the optical path length difference is π±2nπ in phase conversion, all the light input to input port 1 is output from output port 1. The theoretical formula for the MZI transmission spectrum is disclosed in Non-Patent Document 6.

[0063] As in the first embodiment, the core width, core refractive index, core thickness, and chip coordinates in the wafer of the arrayed waveguide can be selected as feature quantities, and machine learning can be processed as a regression problem.

[0064] As described above in detail, the AI ​​prediction system for optical properties disclosed herein enables accurate prediction of optical properties. By using the predicted value in the feedforward system described in the first embodiment, it is possible to realize efficient and high throughput manufacturing processes for optical integrated circuits.

[0065] [Embodiment 3] In this embodiment, a collection of element structures that can be used as features in the above-mentioned AI prediction system is productA specific example of the marker will be described below. First, a basic configuration for acquiring processing information according to an embodiment of the present invention will be described below.

[0066] [Basic configuration] Fig. 8 is a diagram for explaining the basic configuration of this embodiment, and is a diagram for explaining a waveguide pattern formed on a substrate and a marker formed together with this pattern. Fig. 8 shows an AWG waveguide pattern 100 as an example of a waveguide, and a marker 200 formed together with this pattern.

[0067] As shown in FIG. 8, the AWG waveguide pattern 100 has a bent portion 101, a tapered portion 102, a straight portion 103, and the like, as components that determine the shape of the waveguide. The components are not limited to these, and can be determined according to how the processing information obtained from the marker, which will be described later, is used, such as the shape of the target waveguide pattern. For example, components that determine the shape of the waveguide include an intersection portion and a directional coupler portion in addition to the bent portion, tapered portion, and straight portion described above. In the example shown in FIG. 8, the marker 200 is formed in the almost center of the AWG waveguide pattern 100 in the same process as the optical waveguide described in FIG. 1. The marker 200 is approximately square and has a size of about 100 μm×100 μm, as shown on the right side of FIG. 8. In contrast, the waveguide pattern 100 has a size of about several tens of mm×several tens of mm, as shown on the left side of FIG. 8 (see Non-Patent Document 2).

[0068] The marker 200 is formed by integrating elements having basically the same shape as a part of the constituent elements of the waveguide. In the example shown in Fig. 8, for example, the tapered portion 102 of the waveguide is shown as a constituent element of the waveguide, and this is formed as a processed information element 102M in the marker 200. Similarly, the straight portion 103 of the waveguide is formed as a straight processed information element 103M in the marker 200.

[0069] In this way, by forming and imaging the marker 200 in which the components of the waveguide are integrated during the manufacturing process of the waveguide, it is possible to represent the processing information of the manufacturing process. The processed shapes of the waveguide components, such as bends, tapers, intersections, and straight lines, affect the optical propagation characteristics of the waveguide. Furthermore, due to manufacturing variations during manufacturing, even if the shape is the same, the processing results may have unevenness within the wafer surface. For this reason, by aggregating information on the processing information elements (hereinafter also simply referred to as "processing information") in the marker 200 formed in the same process, corrections in subsequent processes and predictions of optical characteristics can be made based on this processing information.

[0070] The processing information element formed in the marker 200 is basically the same as the parameter value of the waveguide component in the waveguide pattern 100 to be manufactured. For example, if the component is a "straight line", the straight line in the marker 200 is formed with the same width as the "width" of the component as a parameter. Similarly, if the component is a "bend", the parameters are "width" and "bend radius", if it is a "taper", the parameters are "taper rate" and "taper start point & end point width", if it is a "crossing", the parameters are "crossing angle", if it is a "directional coupler", the parameters are "waveguide width" and "spacing" of the waveguide. In the marker 200, the processing information element is formed with these parameters having the same values ​​as the waveguide pattern. It is needless to say that the parameters for each component of the waveguide are not limited to the above example. It is needless to say that they can be appropriately determined according to the processing information to be obtained and the purpose of use thereof.

[0071] In this embodiment, the marker 200 described above is formed, and processing information is obtained from it. This makes it possible to obtain processing information by imaging only the marker. As a result, it becomes possible to obtain processing information with high throughput. If the marker of this embodiment is not used, for example, when imaging components in the entire waveguide with high resolution, there are inconveniences such as an increase in the number of images taken and an increase in the memory capacity of the image data. As a result, in order to determine the processing state, it is necessary to observe and measure the shape after processing over the entire wafer surface, which makes the process complicated and reduces the throughput.

[0072] It is desirable that the size and shape of the marker 200 are as small as possible so as not to affect the waveguide pattern. On the other hand, the entire marker must fit within the imaging field of view, and the shape of the imaging element is generally rectangular. From these points of view, it is desirable that the marker 200 also fits within the rectangular range accordingly. For example, for a chip size of 10 mm x 10 mm, if the size of the marker 200 is 1 mm x 1 mm, the imaging size required for inspection is 1 / 100, and if the size is 100 μm x 100 μm, the imaging size required for inspection is 1 / 10,000. This indicates that the throughput when measured with the same microscope magnification is 100 times and 10,000 times, respectively. In order to improve the throughput, it is desirable that the length of one side of the marker 200 is 1 / 10 of the length of one side of the chip or the area of ​​the marker 200 is 1 / 10 of the chip. In addition, even if the chip size is 500 μm × 500 μm, if the size of the marker 200 is 100 μm × 100 μm, the imaging size required for inspection is 1 / 25. This means that the throughput when measured with the same microscope magnification is 25 times higher.

[0073] It should be noted that it is not necessary for the processing information elements in the marker 200 to have the same parameter values ​​for all components of the waveguide. Examples of such elements will be described below.

[0074] For example, when the component is a "bend", the radius of the formed waveguide pattern is large, such as on the order of mm in a silica-based waveguide, and the range required for imaging becomes large. In such a case, a processing information element with a smaller value of the parameter "bend radius" is formed in the marker. When acquiring the processing information element from the marker 200, processing information according to the ratio of the smaller "bend radius" is acquired. Specifically, according to the above-mentioned "bend radius" that is smaller than the actual bend radius in the waveguide pattern, a plurality of waveguides with bend radii are prepared in advance, and the formation results (waveguide width, etc.) are compared with the actual waveguide pattern. Then, in the comparison result, the formation result (processing information) of the actual bend radius in the waveguide pattern can be extrapolated and estimated according to the ratio of the above-mentioned bend radii. Note that, as a characteristic of the "bend" part, even if the bend radius in the marker 200 is different from the actual bend radius in the waveguide pattern, even if the shape (processing information) of the bend part is estimated based on the different bend radii of the marker 200, the estimated result of the "bend" part is close to the shape of the bend part actually formed. As a result, even if the radius of the actually formed waveguide pattern becomes large, it is possible to fit the processing information within the size of the marker 200.

[0075] Fig. 9 is a diagram for explaining the configuration of another marker of this embodiment. As another example, only a part of the bend of the waveguide component may be formed as a processing information element in the marker 200. Fig. 9(a) is a diagram showing the configuration. As shown in Fig. 9(a), only a part of the bend 100 of the waveguide component is formed as a processing information element 100M. Ma The processing information is formed within the marker 200. In this way, by making the bent portion corresponding to a part of the angle of the bend the processing information element 100M, it is possible to fit the processing information within the size of the marker 200. That is, although it is not possible to know the difference (angle dependency) according to the positional relationship with the bent portion in the actual waveguide pattern from the processing information element 100M that is only a part, it is possible to estimate that the same formation result (processing information) will be obtained even at other angles from the formation result of the processing information element 100M at the representative point (angle) of the processing information element 100M.

[0076] As another example, when the waveguide component is a circular arc as a kind of bend, it may be divided equally and formed as a processing information element in the marker 200. FIG. 9B is a diagram showing the configuration. As shown in FIG. 9B, the circular arc 100 of the waveguide component is divided into four equal parts 100A, 100B, 100C, and 100D. Then, the processing information elements 100MA, 100MB, 100MC, and 100MD corresponding to them are formed in the marker 200. At this time, the directions or angles of the processing information elements 100MA, 100MB, 100MC, and 100MD are made different. Specifically, they are arranged at equal intervals in the circumferential direction. This makes it possible to make the formation range more compact. In this way, by dividing the bend equally and forming the processing information elements 100M, it becomes possible to fit the processing information within the size of the marker 200. In other words, by interpolating the four discrete pieces of information (values) of the formation results (such as core width) for each representative point (angle) of the four processing information elements 100MA, 100MB, 100MC, and 100MD, the formation results (processing information) for the actual waveguide pattern can be estimated.

[0077] As shown in the example of FIG. 9(b), the processing information element 100M may be formed in the marker 200 in a direction or angle different from the direction or angle at which the waveguide component 100 is formed. In this case, the effect of so-called angle dependency is taken into consideration. That is, from information on the formation results (such as core width) of the processing information element at different angles in the marker 200, the respective angle dependencies can be calculated, and the formation results (processing information) at other angles can be estimated based on the calculation results. In particular, as long as the processing information elements are mutually orthogonal, they can be expressed for any angle. Specifically, the degree of influence in the manufacturing process may differ depending on the direction or angle of the waveguide component on the substrate. For this reason, the following configuration can be adopted.

[0078] First, when the arrangement direction of the same components in the waveguide pattern is constant, it is preferable that the processing information elements formed in the marker are formed in the same direction. Second, it is preferable that the same components in the waveguide pattern are arranged at different angles, preferably at orthogonal angles, in the marker. Third, when the components are "arcs," it is preferable that the arcs are divided at equal intervals and the divided arcs are arranged at different directions or angles in the marker, as explained in FIG. 9(b).

[0079] In addition, there are cases where it is necessary to measure the distribution of a specific pattern in the waveguide pattern to be formed. For such cases, it is desirable to have multiple identical patterns on the substrate. This makes it possible to calculate the change in the formation result (such as core width) corresponding to the coordinates on the wafer based on the difference in the formation result between the markers in each of the multiple patterns.

[0080] The processing information is not limited to the shape information of the waveguide components described above. For example, the processing information may be the following information.

[0081] The average value of the width of the waveguide that is actually formed compared to the desired waveguide width (design value) of the waveguide pattern, and the waviness and roughness of the waveguide surface and side surfaces. The actual waveguide width formed for each design value of the waveguide width (dependence of the difference between the design value and the actual value on the design waveguide width) The width of the waveguide that is formed varies depending on the bending radius, or the width of the waveguide that is formed at the bending radius and the waveguide angle used varies. -Waveguide width when the waveguide is dense or close together (including average, waviness, roughness, etc.) -Brightness distribution and brightness of the captured image of the waveguide pattern. Based on this, information on the roughness of the processed surface and the processing height can be obtained.

[0082] The processing information elements in the mark 200 described above can be formed and acquired during the manufacturing process, and used for correction in a later process. In addition, the optical characteristics of the finally manufactured optical waveguide can be estimated and predicted based on the acquired processing information. Note that the estimation and prediction of characteristics based on the acquired processing information described above is not limited to the finally manufactured optical waveguide, and it is also possible to estimate and predict the characteristics of a resist pattern formed during the manufacturing process, as will be apparent from the following description. Next, their basic configurations will be described.

[0083] [Correction based on processing information] In one embodiment of the present invention, a marker is formed during one step of the manufacturing process of a waveguide, processing information is obtained from the marker, and the processing information is used to correct the processing in the subsequent step.

[0084] For example, in a photoresist developing process, if processing information that "the resist pattern is thicker than a specified value" is obtained from the processing information element of the mark 200 formed in the same process, a correction of "lengthening / strengthening the etching time / strength" is made in the subsequent etching process, and the width of the waveguide formed in the etching process is set to a desired value (design value). Conversely, if processing information that "the resist pattern is thinner than a specified value" is obtained, a correction of "shortening / weakening the etching time / strength" is made.

[0085] As another example, after the etching process of the waveguide manufacturing, the core width is measured from the processing information element of the mark 200 formed in that process. In addition, the width of the waveguide pattern through which the optical signal actually propagates can be estimated from the measured value. In the etching process, etching can be performed as an additional process based on the measured value of the core width. For example, in the case of information that "the waveguide width after etching is wider than the ideal value," a correction is performed to narrow the core width formed by additional etching. When the width is narrower than the desired value, it is possible to deal with the problem by, for example, designing the waveguide to be wider than the desired waveguide width, or by dividing the etching into two steps and acquiring the intermediate state. The ideal value of the waveguide width here may be a design value, or it may be an ideal waveguide width calculated from the characteristic prediction described later.

[0086] When etching conditions are determined based on processing information, it is desirable to determine the etching time and the like taking into consideration the wraparound of the etching gas.

[0087] Fig. 10 is a flow chart for explaining an example of correction using markers in the manufacturing process of an optical waveguide according to an embodiment of the present invention. Steps S1 to S8 in Fig. 10 indicate the respective steps of manufacturing the optical waveguide described above in Fig. 1. Also, the dicing step shown in step 9 is a step of cutting the wafer of optical waveguides manufactured up to that point into chips.

[0088] In the above manufacturing process, in this example, a waveguide pattern is formed in the photoresist development process of step S5, and a marker 200 is also formed (see FIG. 1). As described in FIG. 8, the marker 200 has processing information elements corresponding to the components of the waveguide pattern formed therein. Then, in the next step S201, the marker 200 is imaged, processing information is obtained from the imaged processing information elements, and the etching conditions in the subsequent etching process (S6) are calculated based on this processing information. Thereafter, the etching process is performed in step S6. As a result, as described above, for example, when the formed "resist pattern is thicker than a predetermined value", the etching conditions in the etching process are calculated (corrected) to "increase the etching time / intensity" as the etching conditions. As a result, the width of the waveguide formed in the etching process can be set to a desired value (design value).

[0089] [Optical property prediction based on processing information] In one embodiment of the present invention, a marker is formed during one step of the manufacturing process of the waveguide, processing information is obtained from the marker, and the optical characteristics of the final optical waveguide are estimated and predicted using the processing information.

[0090] FIG. 11 is a diagram for explaining an example of prediction using a marker in the manufacturing process of an optical waveguide according to an embodiment of the present invention. This example utilizes the fact that the optical characteristics of an optical waveguide are determined according to the shape (width and thickness) and refractive index of the core through which light propagates. As shown in FIG. 11, the film thickness and refractive index of the core are obtained after the film formation process of the core film. In addition, after the etching process, the width of the core pattern is obtained as processing information from the formed marker 200. Then, the optical characteristics of the optical waveguide finally manufactured are predicted based on the obtained width of the core pattern, thickness of the core film, and refractive index. Note that, as shown in FIG. 11, the acquisition (imaging) of processing information after the etching process is performed by imaging, measuring, and acquiring the width and refractive index of the core pattern in the area corresponding to the marker 200 of the core film 18 formed in this process (this area is also referred to as the "marker" in this specification).

[0091] In addition, correction may be made in a later process based on the processing information obtained as described above in FIG. 11. For example, processing information on the thickness and refractive index of the core in the film forming process of the core film and the core pattern width in the etching process is obtained. Then, etching conditions in an additional process of the etching process are determined. As a result, for example, in the case where the optical characteristics estimated from the thickness and refractive index of the core and the core width measured after the etching process, assuming that the core width is narrower, are closer to the design value. In other words, as described above, in the case of processing information that "the core width of the waveguide pattern is wider than the ideal value," a correction can be made in the etching process to narrow the core width formed in consideration of the wraparound of the etching gas.

[0092] As described above, the same correction is possible when capturing an image of the marker 200 in the resist pattern formation stage. For example, when the optical characteristics estimated from the film thickness and refractive index of the core and the resist pattern width measured after the resist pattern formation process are closer to the design value when the resist pattern width is assumed to be narrower. In other words, when the processing information is that the "resist pattern width is wider than the ideal value," a correction can be made in the etching process to narrow the core width formed by taking into account the wraparound of the etching gas.

[0093] As described above, the marker 200 is formed in the same process as the process for manufacturing the optical waveguide shown in FIG. 1. The timing for acquiring the processing information from the processing information element of the marker 200 varies depending on the purpose of use. As shown in the above example, the processing information may be acquired after the etching process and the photoresist development process. Also, the processing information may be acquired after the upper cladding deposition process, in which case processing information close to the final waveguide pattern can be acquired.

[0094] Moreover, instead of acquiring the processing information in some of the steps, the processing information may be acquired from the marker 200 in all of the steps shown in FIG.

[0095] More preferably, the process is performed after the core deposition process and before the upper clad deposition process, because in an optical waveguide structure, most of the light passes through the core, and the shape of the core has the greatest effect on the characteristics, and also because after the upper clad is deposited, it is necessary to image the structure inside the upper clad, and in this case, noise may be introduced into the acquired data.

[0096] Below is the list of element structures explained so far. product A more specific example of the marker will now be described.

[0097] [First Example] As a first embodiment of the present invention, the waveguide components and corresponding processing information elements of the marker when manufacturing an AWG waveguide will be described with reference to Fig. 8, which has been described for the basic configuration. The bend 101, taper 102, and straight line 103, which are the waveguide components constituting the AWG waveguide, correspond to processing information elements 101M, 102M, and 103M of the marker 200, respectively.

[0098] The parameters of the waveguide components are, for the bend 101, the waveguide width, the bend radius, and the spacing; for the taper 102, the taper ratio, and the taper start / end waveguide width; and for the straight line 103, the waveguide width. With regard to these parameters, in the marker 200, the processing information elements 102M and 103M are formed with the same values ​​(taper ratio and taper start / end waveguide width, waveguide width) as the corresponding taper 102 and straight line 103, respectively. In addition, the processing information elements 102M and 103M are formed with at least two orthogonal orientations or angles.

[0099] On the other hand, in the marker 200, the processing information element 101M may have a parameter value different from that of the corresponding bend 101. In detail, as described above in FIG. 9(b), the bend radius is set to a value smaller than that of the actual waveguide component. The waveguide width and interval have the same parameter values ​​as the corresponding bend 101. In addition, the processing information element 101M is formed in four places with different orientations or angles. This allows the processing information element 101M corresponding to the bend 101 to be formed within the size of the marker 200.

[0100] In addition, when the taper rate of the taper 102 is small and the difference between the taper start / end waveguide widths is large, the taper becomes longer and the size of the processing information element 102M becomes larger. In this case, the parameter values ​​are not made the same, but, for example, the difference between the taper start / end waveguide widths is made smaller. As described above, it is not necessary for all parameter values ​​of the waveguide components and the processing information elements to be the same.

[0101] [Second Example] 12 is a diagram showing a marker according to the second embodiment. It shows the waveguide components when manufacturing the waveguide constituting the MZI switch and the corresponding processing information elements of the marker. The bend 101, the straight line 103, the intersection 104, and the directional coupler 105, which are the waveguide components constituting the MZI switch waveguide, correspond to the processing information elements 101M, 103M, 104M, and 105M of the marker 200, respectively.

[0102] The parameters of the waveguide components are: bend 101 is the waveguide width and bend radius, and straight line 103 is the waveguide width. Also, intersection 104 is the intersection angle, and directional coupler 105 is the straight line of the coupling section and the interval between them. With regard to these parameters, in marker 200, these processed information elements 103M, 104M, and 105M are formed with the same values ​​as the corresponding straight line, intersection, and directional coupler 105, respectively. In addition, processed information elements 103M and 105M are formed in at least two different directions or angles that are orthogonal to each other, and processed information element 104M is formed in two different directions or angles.

[0103] On the other hand, in the marker 200, the processing information element 101M has a parameter value different from that of the corresponding bend 101. In detail, as described in FIG. 9B, the bending radius is set to a value smaller than that of the actual waveguide component. In addition, the processing information element 101M is formed in four places with different orientations or angles. This allows the processing information element 101M corresponding to the bend to be formed within the size of the marker 200.

[0104] [Third Example] Fig. 13 is a diagram showing a marker according to the third embodiment, showing the waveguide components when manufacturing the waveguide constituting the MZI switch and the corresponding processing information elements of the marker. It is almost the same as the marker according to the third embodiment described in Fig. 12, and the following describes the differences.

[0105] In an MZI switch, for example, as in the PILOSS configuration, the MZI waveguides and intersections 104 are often arranged along the same direction. Therefore, it is desirable to make the processing information elements of the marker 200 the same as the direction in which they are actually arranged. This is because the characteristics of the MZI switch, such as the ON / OFF extinction ratio and its wavelength dependency, and the leakage light to other ports, can be determined by the configuration and intersections of the MZI.

Claims

1. 1. An artificial intelligence prediction system for optical integrated circuits, comprising: a central processing unit that receives input data, performs learning and inference based on the input data, and generates predicted data of optical properties; the input data is process data acquired in a manufacturing process of the optical integrated circuit, and includes film thickness distribution and refractive index distribution data in a wafer of a core layer, and refractive index distribution data in a wafer of a lower cladding layer, or refractive index distribution data in a wafer of an upper cladding layer; The process data includes: Image data of the fabricated optical waveguide core, or further comprising image data of integrated markers of optical waveguide components within the optical integrated circuit; the image data of the core is transformation data that extracts a sidewall of the core; The image data of the integrated marker includes transformation data that extracts a sidewall of a core of the component. An artificial intelligence prediction system.

2. The accumulation marker is The artificial intelligence prediction system of claim 1, further comprising at least one of a straight line, a bend, a taper, a cross, or a directional coupler having the same configuration parameters as the fabricated optical waveguide.

3. 2. The artificial intelligence prediction system according to claim 1, wherein a data set including the process data and the corresponding measured data of the optical characteristics is further input to the central processing unit as training data.

4. 4. The artificial intelligence prediction system according to claim 1, wherein the optical characteristics include a characteristic value determined by interference of an optical signal due to an optical path length difference between two or more optical waveguides.

5. The characteristic values ​​are The central wavelength of the transmission band of the arrayed waveguide grating, Spectral parameters of the transmission band of the arrayed waveguide grating, The central wavelength of the transmission band of the Mach-Zehnder interferometer, or Spectral parameters of the transmission band of a Mach-Zehnder interferometer.

5. The artificial intelligence prediction system according to claim 4, further comprising:

Citation Information

Patent Citations

  • Blast furnace operating method

    JP1978095814A

  • Manufacturing system

    JP1988249328A

  • Semiconductor manufacturing system

    JP1998163080A

  • Method of manufacturing optical waveguide, optical waveguide device using the method, and waveguide type optical multiplexer / Demultiplexer

    JP2003232948A

  • Optical waveguide structure and its manufacturing method

    JP2004206016A