Design method of test structure, storage medium and terminal
Through machine learning models, the historical test structure data is trained and detected to generate the test structure to be designed, which solves the problem of low efficiency in the test structure design in the existing technology, realizes a fast and efficient design process, and improves the three-dimensional visualization tool Design accuracy and efficiency.
Patent Information
- Application Number
- CN202411047359.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-19
AI Technical Summary
In the prior art, the testing structure design process is inefficient, requires a lot of manpower and time, and is prone to violate design rules, making it difficult to quickly generate a test structure that meets the needs.
The machine learning model is used to train and detect the feature data and measured parameter values of the historical test structure, generate the test structure to be designed, and three-dimensional visualization is performed through simulation tools to improve design efficiency.
Quickly generate the tested structure to be designed through machine learning models, reduce human participation, improve design efficiency, and ensure design accuracy and compliance with rules.
Smart Images

Figure CN118981950A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of semiconductor manufacturing technology, and in particular to a design method, storage medium and terminal of a test structure. Background Art
[0002] In the field of semiconductor manufacturing, with the development of technology, the size of semiconductor devices is getting smaller and smaller, while the complexity is getting higher and higher. In order to monitor the manufacturing process of semiconductor devices and ensure the reliability of semiconductor devices, it is common practice to form a test structure (test key) in the semiconductor device for testing and simulating some key parameters of the semiconductor device to ensure the quality of the semiconductor device before leaving the factory.
[0003] The test structure is usually manufactured in the same semiconductor process as the semiconductor device in the wafer, and the test structure and the semiconductor device have a mutual correspondence. Each layer of interconnection lines in the semiconductor device corresponds to the test line located in the same layer in the test structure, and each plug in the semiconductor device corresponds to the test plug located in the same layer in the test structure. Since the test structure and the semiconductor device are prepared in the same process and have a mutual correspondence, the performance of the semiconductor device in the wafer can be obtained by detecting the performance of the test structure. The performance of the semiconductor device in the wafer is reflected by the test structure, avoiding damage to the semiconductor device in the wafer.
[0004] However, the test structure in the prior art still has many problems in the design process. Summary of the invention
[0005] The technical problem solved by the present invention is to provide a design method, a storage medium and a terminal for a test structure, so as to improve the design generation efficiency of the test structure.
[0006] To solve the above problems, the present invention provides a test structure design method, including: providing several groups of test structure groups with different test parameter types, each group of the test structure groups having several historical test structures, and each of the historical test structures having corresponding measurement parameter values; providing a machine learning model; inputting the characteristic data associated with the historical test structures and the corresponding measurement parameter values into the machine learning model for model training; after the model training, inputting the characteristic data associated with the historical test structures into the trained machine learning model for model detection, and outputting predicted parameter values based on the machine learning model; comparing the predicted parameter values with the corresponding measurement parameter values until the detection accuracy reaches a preset accuracy, thereby completing the model detection of the machine learning model; providing design requirement data of the test structure to be designed; inputting the design requirement data into the machine learning model that has completed the model detection, and generating the test structure to be designed based on the machine learning model.
[0007] Optionally, the plurality of historical test structures in each group of the test structure groups are divided into a plurality of training test structures and a plurality of detection test structures.
[0008] Optionally, the characteristic data associated with the training test structure in each group of the test structure groups and the corresponding measurement parameter values are input into the machine learning model for model training.
[0009] Optionally, the feature data associated with the detection test structure in each group of the test structure groups is input into the trained machine learning model for model detection.
[0010] Optionally, the method of comparing the predicted parameter value with the corresponding measured parameter value and obtaining the detection accuracy includes: obtaining the degree of match between the predicted parameter value and the corresponding measured parameter value; when the degree of match reaches a preset degree of match, judging that the predicted parameter value has passed the match; and taking the ratio of the number of preset parameter values that have passed the match to the total number of preset parameter values as the detection accuracy.
[0011] Optionally, the method of detecting until the detection accuracy reaches a preset accuracy includes: when the detection accuracy of the machine learning model does not reach the preset accuracy, increasing the number of model training times of the machine learning model until the detection accuracy of the machine learning model reaches the preset accuracy.
[0012] Optionally, the method for obtaining the degree of match between the predicted parameter value and the corresponding measured parameter value includes: root mean square error or R-square indicator.
[0013] Optionally, the method for outputting predicted parameter values based on the machine learning model includes: associating the machine learning model with a simulation tool; and outputting the predicted parameter values by the simulation tool.
[0014] Optionally, the method for generating the test structure to be designed based on the machine learning model includes: the machine learning model transmits data for generating the test structure to be designed to the simulation tool; and the simulation tool generates a three-dimensional visualization of the test structure to be designed.
[0015] Optionally, different test parameter types include: one or more of current, voltage, resistance, capacitance and inductance.
[0016] Optionally, the characteristic data associated with the historical test structure includes: the number of process layers, design rules, characteristic dimensions of each component, and positional relationships between each component.
[0017] Optionally, the design requirement data of the test structure to be designed includes: characteristic dimensions of each component.
[0018] Optionally, the machine learning model includes: a neural network model or a random forest model.
[0019] Correspondingly, the technical solution of the present invention also provides a storage medium on which computer instructions are stored, and when the computer instructions are executed, the steps of the method described in any one of the above technical solutions are executed.
[0020] Correspondingly, the technical solution of the present invention also provides a terminal, including a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and when the processor executes the computer instructions, the steps of the method described in any one of the above technical solutions are executed.
[0021] Compared with the prior art, the technical solution of the present invention has the following advantages:
[0022] In the test structure design method of the technical solution of the present invention, the machine learning model is trained and tested through the characteristic data associated with the historical test structure and the corresponding measurement parameter values. After the machine learning model passes the test, the test structure to be designed can be quickly generated based on the machine learning model, thereby effectively reducing manpower and improving the generation efficiency of the test structure to be designed.
[0023] Furthermore, the method for generating a test structure to be designed based on a machine learning model includes: the machine learning model transmits data of the test structure to be designed to a simulation tool; the simulation tool generates a three-dimensional visualized test structure to be designed. The simulation tool converts the data of the test structure to be designed into a visualized three-dimensional structure, so as to facilitate the inspection of the test structure to be designed. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a flow chart of a design method of a test structure according to an embodiment of the present invention. DETAILED DESCRIPTION
[0025] As described in the background art, there are still many problems in the design process of the test structure in the prior art, which will be described in detail below.
[0026] At present, designing a test structure with actual testability usually requires the following conditions: mature EDA tools (Electronic Design Automation); engineers or teams who are proficient in EDA tools and familiar with the chip manufacturing process and the relationship between different process layers in the chip; using EDA tools to design and combine the test structure layer by layer, while constantly aligning it with the design rules, and finally being able to connect the test structure to the test pad.
[0027] However, the design method of the above test structure takes a lot of time, especially in the early stage of tape-out when the demand for test structures is particularly large; in addition, repeated checks are required. Although they need to be constantly aligned with the design rules, the structures of some test structures are relatively complex and involve a large number of process layers. It is difficult to avoid areas that violate the design rules when they are manually designed and combined.
[0028] On this basis, the present invention provides a test structure design method, storage medium and terminal, which train and detect the machine learning model through the characteristic data associated with the historical test structure and the corresponding measurement parameter values. After the machine learning model passes the detection, the test structure to be designed can be quickly generated based on the machine learning model, thereby effectively reducing manpower and improving the generation efficiency of the test structure to be designed.
[0029] Figure 1 : is a flow chart of a design method of a test structure according to an embodiment of the present invention, comprising:
[0030] Step S101, providing a plurality of test structure groups with different test parameter types, each test structure group having a plurality of historical test structures, each historical test structure having a corresponding measurement parameter value;
[0031] Step S102, providing a machine learning model;
[0032] Step S103, inputting the characteristic data associated with the historical test structure and the corresponding measurement parameter values into the machine learning model for model training;
[0033] Step S104, after model training, inputting feature data associated with the historical test structure into the trained machine learning model for model detection, and outputting predicted parameter values based on the machine learning model;
[0034] Step S105, comparing the predicted parameter value with the corresponding measured parameter value until the detection accuracy reaches a preset accuracy, thereby completing the model detection of the machine learning model;
[0035] Step S106, providing design requirement data of the test structure to be designed;
[0036] Step S107, input the design requirement data into the machine learning model that completes the model detection, and generate the test structure to be designed based on the machine learning model.
[0037] The following will describe in detail the various steps of the test structure design method.
[0038] In step S101 , a plurality of test structure groups with different test parameter types are provided, each test structure group has a plurality of historical test structures, and each historical test structure has a corresponding measurement parameter value.
[0039] In this embodiment, the historical test structure is the test structure that has been manufactured on the wafer. Based on the electrical property measurement of the completed test structure, the measurement parameter values corresponding to each historical test structure are obtained. These data will provide support for the subsequent training and detection of the machine learning model.
[0040] In this embodiment, different test parameter types include one or more of current, voltage, resistance, capacitance and inductance. The test structures corresponding to different test parameter types are different. For example, the test structure corresponding to the resistance parameter is usually a strip or snake-shaped polysilicon and an active area with ion implantation; the capacitance parameter is usually a transistor structure, and it is necessary to detect the capacitance generated between the opposite regions between the gate and the source and drain doping layers in the transistor structure.
[0041] In this embodiment, since a large amount of data is required to support both the training and detection phases of the machine learning model, the number of historical test structures in each test structure group cannot be only one. In addition, the several historical test structures in each test structure group need to be divided into several training test structures and several detection test structures. Among them, the data of several training test structures will be used in the training phase of the machine learning model, and the data of several detection test structures will be used in the detection phase of the machine learning model.
[0042] It should be noted that the ratio of the number of training test structures to the number of detection test structures in each test structure group is not fixed, and is usually 1:1. However, in the case of more diverse test structures, the ratio of training test structures needs to be increased, and the highest ratio should depend on the type of test structure.
[0043] In step S102, a machine learning model is provided.
[0044] In this embodiment, the machine learning model includes: a neural network model or a random forest model.
[0045] In step S103, the characteristic data associated with the historical test structure and the corresponding measurement parameter values are input into the machine learning model for model training.
[0046] In this embodiment, the characteristic data associated with the training test structures in each test structure group and the corresponding measurement parameter values are input into the machine learning model for model training.
[0047] In this embodiment, during the model training stage, the characteristic data associated with the historical test structure is used as a condition, and the measurement parameter value corresponding to the historical test structure is used as a result, and both are simultaneously input into the machine learning model, and the machine learning model learns and establishes a mapping relationship by itself.
[0048] In this embodiment, the characteristic data associated with the historical test structure includes: the number of process layers, design rules, characteristic dimensions of each component, and positional relationships between each component.
[0049] Taking the transistor structure as an example, the number of process layers required includes at least: active area layer, gate layer, contact hole layer and metal wiring layer. Moreover, depending on the transistor type and manufacturing process, the ion implantation layer is also different. Taking the NMOS transistor structure as an example, it also requires P-type well area implantation, N-type ion heavy doping, N-type ion light doping, P-type ion heavy doping, and ion implantation for threshold voltage adjustment (optional according to the process). The characteristic dimensions of each component include: gate line width and length, active area widening along the channel direction, contact hole size, metal lead line width, through hole size, etc. The design rules include: special regulations for the dimensions of each component, such as the gate line width cannot be less than the process setting value, the matching of the active area anti-breakdown well process, etc. The positional relationship between the various components includes: the spacing between the contact hole and the gate, the relative position of the gate and the active area, the alignment relationship between the metal lead and the contact hole / through hole, etc.
[0050] It should be noted that in the training stage of the machine learning model, the characteristic dimensions of each component are key dimensions, such as the width of the gate and the length of the source and drain doping layer (i.e. the channel length and width of the transistor structure). These dimensions are the most direct factors affecting the electrical properties of the device, and the key dimensions remain unchanged no matter how they are designed. Other data, such as the portion of the gate that exceeds the active area specified in the design rules, the distance between the contact hole and the active area and the gate, etc., are uncertain and can be optimized by the machine learning model itself.
[0051] In step S104, after model training, feature data associated with the historical test structure is input into the trained machine learning model for model detection, and predicted parameter values are output based on the machine learning model.
[0052] In this embodiment, the feature data associated with the detection test structure in each test structure group is input into the trained machine learning model for model detection.
[0053] In this embodiment, the method for outputting predicted parameter values based on a machine learning model includes: associating the machine learning model with a simulation tool; and outputting the predicted parameter values by the simulation tool.
[0054] In this embodiment, the simulation process may use a TCAD (Technology Computer Aided Design) simulation tool.
[0055] In step S105, the predicted parameter value is compared with the corresponding measured parameter value until the detection accuracy reaches a preset accuracy, thereby completing the model detection of the machine learning model.
[0056] In this embodiment, the method of comparing the predicted parameter value with the corresponding measured parameter value and obtaining the detection accuracy includes: obtaining the matching degree between the predicted parameter value and the corresponding measured parameter value; when the matching degree reaches the preset matching degree, it is judged that the predicted parameter value has passed the matching; and the ratio of the number of preset parameter values that have passed the matching to the total number of preset parameter values is used as the detection accuracy.
[0057] In this embodiment, the method for obtaining the matching degree between the predicted parameter value and the corresponding measured parameter value includes: Root Mean Squared Error (RMSE) or R-squared (R2). The Root Mean Squared Error or R-squared is used to measure the difference between the predicted parameter value and the measured parameter value. When the difference between the two is within a preset range, it is considered that the two are matched.
[0058] In this embodiment, the method of detecting until the detection accuracy reaches the preset accuracy includes: when the detection accuracy of the machine learning model does not reach the preset accuracy, increasing the number of model training of the machine learning model until the detection accuracy of the machine learning model reaches the preset accuracy.
[0059] In step S106, design requirement data of a test structure to be designed is provided.
[0060] In this embodiment, the design requirement data of the test structure to be designed includes: the characteristic dimensions of each component. For example, taking the test structure to be designed as a transistor structure, after the machine learning model is usually detected and put into practical application, only the key dimensions such as the width of the gate and the length of the source and drain doping layer in the transistor structure need to be input.
[0061] In step S107, the design requirement data is input into the machine learning model that has completed the model detection, and a test structure to be designed is generated based on the machine learning model.
[0062] The machine learning model is trained and tested through the characteristic data associated with the historical test structure and the corresponding measurement parameter values. After the machine learning model passes the test, the test structure to be designed can be quickly generated based on the machine learning model, thereby effectively reducing manpower and improving the generation efficiency of the test structure to be designed.
[0063] In this embodiment, the method for generating a test structure to be designed based on a machine learning model includes: the machine learning model transmits data of the test structure to be designed to a simulation tool; the simulation tool generates a three-dimensional visualized test structure to be designed. The data of the test structure to be designed is converted into a visualized three-dimensional structure by the simulation tool, so as to facilitate the inspection of the test structure to be designed.
[0064] Correspondingly, an embodiment of the present invention further provides a storage medium on which computer instructions are stored. When the computer instructions are executed, the steps of any one of the above-mentioned embodiment methods are executed.
[0065] Correspondingly, an embodiment of the present invention further provides a terminal, including a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and when the processor executes the computer instructions, the steps of any one of the above-mentioned embodiment methods are executed.
[0066] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.
Claims
1. A method for designing a test structure, characterized in that: include: Providing a plurality of test structure groups with different test parameter types, each of the test structure groups having a plurality of historical test structures, each of the historical test structures having a corresponding measurement parameter value; Provide machine learning models; Inputting the characteristic data associated with the historical test structure and the corresponding measurement parameter values into the machine learning model for model training; After model training, the feature data associated with the historical test structure is input into the trained machine learning model for model detection, and a predicted parameter value is output based on the machine learning model; Comparing the predicted parameter value with the corresponding measured parameter value until the detection accuracy reaches a preset accuracy, thereby completing the model detection of the machine learning model; Provide design requirement data of the test structure to be designed; The design requirement data is input into the machine learning model that completes model detection, and the test structure to be designed is generated based on the machine learning model.
2. The test structure design method according to claim 1, characterized in that: The plurality of historical test structures in each of the test structure groups are divided into a plurality of training test structures and a plurality of detection test structures.
3. The method for designing a test structure according to claim 2, wherein: The characteristic data associated with the training test structure in each group of the test structure groups and the corresponding measurement parameter values are input into the machine learning model for model training.
4. The method for designing a test structure according to claim 2, wherein: The feature data associated with the detection test structure in each group of the test structure groups is input into the trained machine learning model for model detection.
5. The method for designing a test structure according to claim 1, wherein: The method of comparing the predicted parameter value with the corresponding measured parameter value and obtaining the detection accuracy includes: obtaining the matching degree between the predicted parameter value and the corresponding measured parameter value; when the matching degree reaches a preset matching degree, judging that the predicted parameter value has passed the matching; and taking the ratio of the number of the preset parameter values that have passed the matching to the total number of the preset parameter values as the detection accuracy.
6. The method for designing a test structure according to claim 5, characterized in that: The method for detecting a detection accuracy rate reaching a preset accuracy rate includes: when the detection accuracy rate of the machine learning model does not reach the preset accuracy rate, increasing the number of model training times of the machine learning model until the detection accuracy rate of the machine learning model reaches the preset accuracy rate.
7. The method for designing a test structure according to claim 5, characterized in that: Methods for obtaining the degree of matching between the predicted parameter value and the corresponding measured parameter value include: root mean square error or R-square index.
8. The method for designing a test structure according to claim 1, wherein: The method for outputting predicted parameter values based on the machine learning model includes: associating the machine learning model with a simulation tool; The predicted parameter values are output by the simulation tool.
9. The method for designing a test structure according to claim 8, characterized in that: The method for generating the test structure to be designed based on the machine learning model includes: the machine learning model transmits data for generating the test structure to be designed to the simulation tool; and the simulation tool generates a three-dimensional visualized test structure to be designed.
10. The method for designing a test structure according to claim 1, wherein: The different test parameter types include: one or more of current, voltage, resistance, capacitance and inductance.
11. The method for designing a test structure according to claim 1, wherein: The characteristic data associated with the historical test structure includes: the number of process layers, design rules, characteristic dimensions of each component, and positional relationships between each component.
12. The method for designing a test structure according to claim 1, wherein: The design requirement data of the test structure to be designed includes: characteristic dimensions of each component.
13. The method for designing a test structure according to claim 1, wherein: The machine learning model includes: a neural network model or a random forest model.
14. A storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed, the steps of the method according to any one of claims 1 to 13 are executed.
15. A terminal comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, characterized in that: When the processor runs the computer instructions, the steps of the method according to any one of claims 1 to 13 are performed.
