Method and system for acquiring semiconductor process recipe, and semiconductor process apparatus

A multilayer deep neural network model optimizes process parameters for deep silicon etching, addressing inefficiencies in manual optimization and enhancing automation in semiconductor manufacturing.

JP2024540986AActive Publication Date: 2024-11-06BEIJING NAURA MICROELECTRONICS EQUIP CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2024523932
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2021-11-04
Filing Date
2022-10-28
Publication Date
2024-11-06
Estimated Expiration
2042-10-28

AI Technical Summary

Technical Problem

The existing methods for deep silicon etching in semiconductor manufacturing, such as the Bosch process, result in scalloped structures and increased sidewall roughness, and the process parameters are difficult to optimize manually, leading to inefficiencies and a lack of automation in achieving desired etching results.

Method used

A method using a trained multilayer deep neural network model to automatically acquire a process recipe by inputting process parameters, optimizing them through a gradient algorithm and self-consistent iteration to meet predetermined requirements, thereby generating optimized process parameters for deep silicon etching.

Benefits of technology

The method enables automated and efficient generation of process recipes that meet precise etching requirements, reducing sidewall roughness and improving the quality of deep silicon structures by optimizing process parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024540986000001_ABST
    Figure 2024540986000001_ABST
Patent Text Reader

Abstract

The present disclosure provides a semiconductor processing apparatus, a tray and a cassette, and relates to the field of semiconductor equipment. The semiconductor processing apparatus includes a process chamber, a transport chamber and a load chamber, the load chamber is used to accommodate a cassette, the cassette is used to place a tray, a transport device for transporting the tray between the process chamber and the load chamber is provided in the transport chamber, an image recognition device is provided at the boundary between the load chamber and the transport chamber to recognize a predetermined characteristic image on the tray in the process of the transport device moving the tray and determine characteristic information of the tray, and the semiconductor processing apparatus further includes a controller for acquiring the characteristic information, searching for transport parameters corresponding to the characteristic information from a predetermined tray information library, and controlling the transport device to transport the tray according to the transport parameters. The tray is applied to the semiconductor processing apparatus. The cassette is used to place the tray. The present application can solve problems such as low transport accuracy and laborious work.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to the technical field of semiconductor manufacturing, and more particularly to a method, system and semiconductor processing equipment for obtaining a semiconductor process recipe. [Background technology]

[0002] In the field of microelectronics manufacturing, plasma etching is an important process in device fabrication. For example, etching a trench gate in a transistor discrete device can suppress the short channel effect of the device and solve problems such as mobility reduction caused by electron scattering in the channel. Therefore, the transistor discrete device has undergone a transformation from the traditional planar gate to the trench gate, and is applied to silicon-based and silicon carbide-based metal-oxide-semiconductor field-effect transistor (MOSFET) devices. For example, etching a superjunction in a transistor discrete device can effectively reduce the conduction resistance and improve the switching speed of the device. Because the driving current when adopting a lateral structure superjunction is small, the mainstream products on the market adopt a vertical structure superjunction. The process technology has developed from a multi-epitaxial process to a deep trench process, and therefore it is necessary to adopt silicon deep etching. For example, in a gallium nitride high electron mobility transistor (HEMT) device, one of the methods to realize a normally-off device is to use a recessed groove in the gate. By etching a recess, the surface energy level of the gallium nitride gate can be changed by the recess, enabling the device to be normally off. Discrete devices that originally required trench etching are also developing to have high aspect ratios and high verticality. For example, in traditional silicon capacitor devices, the deeper the trench, the more effectively the device space can be utilized and the greater the capacitance value can be. The higher the verticality, the better the actual device can match the theoretical model and meet the design requirements. In advanced packaging, for example, etching through silicon via (TSV) structures can improve device performance, reduce power consumption, and miniaturize devices.

[0003] Therefore, with the development of fields such as semiconductor discrete devices and advanced packaging, the requirements for silicon deep etching are becoming higher and higher. Silicon deep microstructures have a large aspect ratio and high verticality, so they are difficult to form by traditional wet etching and need to be formed by dry etching methods. To obtain deep and vertical silicon microstructures, the "Bosch" process, a time-separated dry etching process, is mainly used. The process provides sidewall passivation protection and fluorine-based plasma chemical reaction by inducing fluorocarbon polymer with plasma to etch silicon downward, and the sidewall passivation protection (deposition step) and fluorine-based plasma chemical reaction (etch step) are alternated. The "Bosch" process alternates deposition and etching, which inevitably leads to the formation of a scallop structure and large sidewall roughness.

[0004] Generally, the scallops at the top of the silicon deep digging structure have the largest dimensions and the roughest sidewalls, and obvious scallops and lateral wrinkles are visible. Therefore, it is necessary to find a way to reduce the roughness of the top of the silicon deep digging structure. In addition to considering the sidewall roughness, always pay attention to the etching morphology (sidewall angle). The reason is that after the formation of the silicon deep digging structure, the plasma is difficult to penetrate into the silicon deep digging structure, and the products after etching are also difficult to extract from the silicon deep digging structure, and it is often easy to form a silicon deep digging structure with a large upper part and a small lower part. At this time, it is necessary to improve the mean free path of the plasma by adjusting the process parameters and help the plasma move to the wafer. In addition, the etching depth, etching rate, mask selectivity, etc. are also indicators to evaluate the quality of the silicon deep digging etching effect. In order to obtain a good silicon deep digging etching effect, it is necessary to optimize the process parameters, that is, adjust the process recipe. At present, mainly process personnel manually try various recipes one by one to determine the recipe that can meet the process requirements, and this process is very complicated and inefficient, and automation is often not possible. Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present invention is to provide a method, system, and semiconductor process apparatus for automatically acquiring a corresponding process recipe in accordance with predetermined process requirements, thereby improving the efficiency of acquiring the process recipe. [Means for solving the problem]

[0006] In order to achieve the above object, the present invention provides inputting a set of process parameters required for a process into the constructed deep neural network model randomly, and obtaining a process result evaluation index corresponding to the set of process parameters, wherein the set of process parameters includes a plurality of process parameters, and the deep neural network model is a trained multi-layer deep neural network, and the deep neural network model includes an input layer, at least one intermediate layer and an output layer, the input layer is used to input the set of process parameters, and the output layer is used to output the process result evaluation index corresponding to the set of process parameters; determining whether a difference between each of the process result evaluation indexes and a corresponding process requirement satisfies the requirement; if YES, using the set of process parameters as a process recipe for performing an actual process; if NO, optimizing, layer by layer, eigenvalues ​​in each of the intermediate layers and numerical values ​​of each of the process parameters in the input layer based on a gradient algorithm and a self-consistent iterative method along a direction from the output layer to the input layer according to the process requirements, until a set of process parameters that enables all process result evaluation indexes output from the deep neural network model in the input layer to satisfy the process requirements is obtained, and using the set of process parameters that satisfy the process requirements as a process recipe for performing an actual process.

[0007] Optionally, the method for constructing a deep neural network model comprises: determining the number of layers and the structure of the deep neural network model according to the number of process parameters included in the process parameter group and the number of the process result evaluation indexes corresponding to the process parameter group, the number of the intermediate layers is the difference between the number of the process parameters and the number of the process result evaluation indexes, the deep neural network of each layer in the deep neural network model includes a plurality of neurons, one neuron is sequentially decreased between adjacent deep neural networks along the direction from the input layer to the output layer, each neuron is connected to all neurons in the deep neural network of the previous layer, each process parameter is an input feature of one neuron in the input layer, each neuron in the intermediate layer is used to calculate the output features of all neurons in the deep neural network of the previous layer, the output feature of each neuron in the output layer is the process result evaluation index, and the number of the process parameters included in the process parameter group is greater than the number of the process result evaluation indexes corresponding to the process parameter group; and providing an activation function to be executed in the deep neural network model according to the number of layers of the deep neural network model.

[0008] Optionally, the method for constructing a deep neural network model comprises: training the deep neural network model using a deep learning method with a process database as training data, the process database including historical process parameter data and historical process result data for a particular process; The method further includes determining weight terms and bias terms of a deep neural network of each layer in the deep neural network model according to a training result of the activation function.

[0009] Optionally, the step of optimizing eigenvalues ​​in each of the hidden layers and values ​​of each process parameter in the input layer layer-by-layer based on a gradient algorithm and a self-consistent iteration method includes: A step S1 of finding one process result evaluation index from the output layer that has the largest difference from the process requirement; Step S2 of finding one process parameter from an intermediate layer adjacent to the output layer that has the largest value of the weight term associated with the process result evaluation index with the largest degree of difference; Step S3: fix the weight term value of the one process parameter having the largest weight term value and the bias term value, and optimize and step-adjust the eigenvalue of the one process parameter having the largest weight term value using a gradient algorithm until the difference between the process result evaluation index having the largest difference in the output layer and the process requirement satisfies a set requirement; At this time, if there are other process result evaluation indexes whose difference with the process requirement does not satisfy the requirement, the steps S1 to S3 are repeated until all the process result evaluation indexes in the output layer satisfy the process requirement, and at the same time, a step S4 of determining an eigenvalue of each process parameter in the intermediate layer is performed; and step S5 of continuing to optimize, layer by layer, the eigenvalues ​​of each process parameter in other intermediate layers and the numerical values ​​of the process parameters in the input layer using a gradient algorithm and a self-consistent iterative method along a direction from the output layer to the input layer based on the eigenvalues ​​of each process parameter in the intermediate layer obtained in step S4, until a set of process parameters that enables all process result evaluation indexes output from the deep neural network model in the input layer to satisfy the process requirements is obtained.

[0010] Optionally, after step S3, determine whether the difference between the process result evaluation index correspondingly adjusted in step S3 and the process requirement satisfies a set requirement, and if so, execute step S4; if not, fix one process parameter with the largest value of the weight term obtained by adjustment and optimization in step S3; Step S6 of finding one process parameter from a hidden layer adjacent to the output layer that has the second largest value of the weight term associated with the corresponding adjusted process result index in step S3; and performing step S7 of fixing the value of the weight term of the one process parameter having the second largest value of the weight term, and optimizing and step-adjusting the eigenvalue of the one process parameter having the second largest value of the weight term using a gradient algorithm until a difference between the process result evaluation index correspondingly adjusted in step S3 in the output layer and the process requirement satisfies a set requirement.

[0011] Optionally, after step S5, After optimizing and adjusting all the process parameters related to the process result index, if there is still a process result index in the output layer whose difference with the process requirement does not satisfy the set requirement, The method further includes a step S8 of forming a new set of process parameters by selecting the process parameters that enable the smallest sum of squares of the degree of difference between each process result evaluation index and the process requirement, and setting the new set of process parameters as a process recipe for performing an actual process.

[0012] Optionally, the formula for the gradient algorithm is:

[0013] x i =x i -θ×(∂y / ∂x i ) In the formula, x i is the adjusted process parameter in the intermediate layer, y is the adjusted process result evaluation index, and θ is the adjustment step.

[0014] Optionally, the dissimilarity is calculated according to the formula: As the set requirement, the degree of difference between the process result evaluation index and the corresponding set process requirement is 0 to 10%.

[0015] Δ=|(yy 実際 ) / y 実際 | In the formula, Δ is the difference between the process result index and the set process requirement, y is the process result index output from the deep neural network model, and y 実際 is the given process requirement.

[0016] In a second aspect, the present invention provides a method for producing a composition comprising the steps of: A deep neural network model is a trained multi-layer deep neural network, and is capable of outputting accurate process result evaluation indexes corresponding to a set of process parameters based on input process parameters; There is further provided a system for automatically acquiring a semiconductor process recipe, comprising: a calculation module for performing the method for acquiring a semiconductor process recipe according to the first aspect.

[0017] In a third aspect, the present invention further provides a semiconductor processing apparatus including the system for automatically acquiring a semiconductor process recipe according to the second aspect.

[0018] The beneficial effects of the present invention are as follows:

[0019] When process requirements are given, based on the constructed deep neural network model, the present invention optimizes and adjusts the eigenvalues ​​in each intermediate layer and the process parameter group of the input layer by layer using a gradient algorithm and a self-consistent iteration method until a process parameter group that enables all process result evaluation indexes output from the deep neural network model to satisfy the process requirements is derived in the backward direction from the output layer to the input layer according to the set process requirements as a process recipe of an actual process. Compared with the conventional method of repeatedly manually debugging the process recipe parameters, the present invention can automatically derive process recipe parameters that satisfy the process requirements by the deep neural network model according to the process requirements, thereby improving the efficiency of obtaining the process recipe and improving the automation degree of the etching apparatus.

[0020] The system of the present invention has other characteristics and advantages which will become apparent from or will be described in detail in the drawings and specific embodiments incorporated herein below, which together serve to interpret certain principles of the present invention. [Brief description of the drawings]

[0021] The above and other objects, features and advantages of the present invention will become more apparent by describing in more detail the exemplary embodiments of the present invention with reference to the drawings, in which like reference numerals generally refer to like elements.

[0022] [Figure 1]FIG. 1 is a schematic diagram of a deep neural network model applied to a photolithography apparatus in prior art 1. [Diagram 2] FIG. 13 is a schematic diagram of a neural network model for predicting the etching process result in Prior Art 3. [Figure 3A] 4 is a flow chart of steps of a method for obtaining a semiconductor process recipe in accordance with an embodiment of the present invention. [Figure 3B] 4 is another flow chart of steps of a method for obtaining a semiconductor process recipe in accordance with an embodiment of the present invention. [Figure 3C] 5 is yet another flowchart of steps of a method for obtaining a semiconductor process recipe in accordance with an embodiment of the present invention. [Figure 3D] 1 is a flowchart of a specific embodiment of a method for obtaining a semiconductor process recipe according to an embodiment of the present invention. [Figure 4] FIG. 2 is a schematic diagram of a deep neural network model in a method for acquiring a semiconductor process recipe according to an embodiment of the present invention. [Diagram 5] FIG. 2 is a schematic diagram of an etching process with a critical dimension of 3 microns in a method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention. [Figure 6] FIG. 2 is a schematic diagram illustrating the influence of chamber pressure on a trench etching process result evaluation index in a method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention. [Figure 7] FIG. 2 is a schematic diagram showing the influence of the final power of the lower electrode in the etching step on the trench etching process result evaluation index in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 8] FIG. 2 is a schematic diagram illustrating the influence of the final time of an etching step on a trench etching process result evaluation index in a method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 9] FIG. 2 is a schematic diagram illustrating the influence of the final deposition step time on the trench etching process result evaluation index in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 10] FIG. 2 is a schematic diagram of the influence of SF6 intake in the etching step on the trench etching process result evaluation index in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 11] FIG. 13 is a schematic diagram of the effect of C4F8 intake in the deposition step on the trench etching process result evaluation index in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 12] 1 is a SEM view of a trench etch process requirement in a method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; FIG. [Figure 13] FIG. 2 is a schematic diagram illustrating the influence of chamber pressure on hole etching process result evaluation indexes in a method for obtaining a semiconductor process recipe according to one embodiment of the present invention; [Figure 14] FIG. 2 is a schematic diagram of the influence of the final power of the lower electrode in the etching step on the hole etching process result evaluation index in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 15] FIG. 2 is a schematic diagram illustrating the influence of the final time of an etching step on an etching process result evaluation index in a method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention. [Figure 16] FIG. 2 is a schematic diagram illustrating the influence of the final deposition step time on the hole etching process result evaluation index in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 17] FIG. 2 is a schematic diagram of the influence of SF6 intake on the hole etching process result evaluation index in the etching step in the method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; [Figure 18] FIG. 2 is a schematic diagram of the effect of C4F8 intake in a deposition step on the hole etching process result evaluation index in a method for obtaining a semiconductor process recipe according to one embodiment of the present invention; [Figure 19]2 is a SEM diagram of a hole etching process requirement in a method for obtaining a semiconductor process recipe according to one specific embodiment of the present invention; FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Prior Art 1 discloses a deep neural network model applied to a photolithography apparatus, and the deep neural network model is shown in Figure 1. The method can automatically output a photolithography process result according to the photolithography process parameters input into the model based on the deep neural network model completing deep learning.

[0024] The deep neural network model is a conventional technology. Specifically, first, the activation function of the model needs to be defined and a linear transformation needs to be performed on the parameters of the input layer, y=f(Σ i w i x i +b), wherein x i is the input eigenvector (feature), w i are the weights for each eigenvalue, and b is the bias term. This is the case for the simplest neural network with only one layer. When there are multi-layer neural networks, the model activation function for the second layer is y=g(Σ j w j f(Σ i w i x i +b1)+b2), and the model activation function for the three layers is y=h(Σ k w k g(Σ j w j f(Σ i w i x i +b1)+b2)+b3), and so on.

[0025] Next, the input-output values ​​(x i ,y) and the actual dataset (xi ,y r ) and the objective function J(w i ,b)=(Σ i w i x i +by r ) 2 Define n real data sets (x i,n ,y r,n ), it is necessary to perform arithmetic averaging on the objective function, J(w i ,b)=(1 / n)Σ n (Σ i w i x i,n +by r,n ) 2 It is.

[0026] In neural network models, data training is an important part, mainly related to parameter optimization. At present, there are three mainstream optimizers, namely stochastic gradient descent, stochastic gradient descent with momentum, and Adam optimization. Take stochastic gradient descent as an example, its algorithm is as follows: i =w i -η×[∂J(w i ,b) / ∂w i ], b=b-η×[∂J(w i ,b) / ∂b].

[0027] That is, perform partial derivative calculations on the relevant parameters, and move according to the product of the step η and the partial derivative each time, until the objective function reaches a minimum value, i.e., the deviation between the data predicted by the model and the actual value is the smallest. For a multi-layer neural network, it can be run backwards for each layer to minimize the objective function of each layer, i.e., the deviation between the data predicted by the model and the actual value is the smallest.

[0028] As a result of research, the inventors found that Prior Art 1 has the following problems.

[0029] The neural network model only provides forward learning and prediction, i.e., it can only derive the process result from the process parameter recipe in the forward direction. During the forward derivation, the variables in the gradient optimization algorithm are weighted by w i , which cannot realize the backward derivation to obtain the process parameters according to the process requirements, and does not involve machine automation.

[0030] Photolithography equipment has high accuracy requirements and the algorithms of its optimizers are complex (while other semiconductor processing equipment such as etching equipment has a certain tolerance for accuracy requirements relative to computing speed).

[0031] Prior art 2 discloses a self-consistent iterative method, that is, when solving a transcendental equation, an analytical solution cannot be obtained, and only a numerical solution can be obtained. When obtaining the numerical solution, a successive iterative method is used, and the method finally converges to an optimal solution to form a self-consistency.

[0032] When an operator in an equation is a function of unknowns, the only way is to use a self-consistent iterative method, which involves first determining the operator for one set of assumed unknowns, then substituting it into the original equation to obtain a numerical solution for another set of the unknowns, until the assumed unknowns are completely consistent with the final calculated unknowns, i.e., self-consistent; this method is used in the field of quantized computation to solve the Hartree-Fock equations.

[0033] The method has not currently been applied to neural network models, nor has it been applied in the field of semiconductor process equipment.

[0034] Prior Art 3 discloses a neural network model for predicting an etching process result. As shown in FIG. 2, the inventor's research has revealed that the method uses a single-layer neural network model, which is not a deep neural network but an ordinary neural network, and can only predict the etching process result and cannot derive process parameters from the etching process result in reverse.

[0035] The present invention can use software algorithms to automatically provide a process recipe for a related process (such as dry etching, etc.) according to actual process requirements. In a specific application, if a user needs to obtain a silicon deep recess structure with a specific etching depth, sidewall angle, sidewall roughness, and mask selectivity, and expects to complete the etching at a specific etching rate, the corresponding process parameter recipe, such as chamber pressure, upper and lower electrode power, intake air flow rate, and etching time, can be automatically provided by the means of the present invention.

[0036] The present invention will be described in more detail below with reference to the drawings. The drawings show preferred embodiments of the present invention, but it should be understood that the present invention is not limited to the embodiments described herein and can be realized in various forms. Rather, these embodiments are provided to make the present invention clearer and more complete, and to fully convey the scope of the present invention to those skilled in the art.

[0037] 3A to 3D are flow charts of some steps of a method for obtaining a semiconductor process recipe according to an embodiment of the present invention. Referring to FIG. 3A to FIG. 3D in conjunction with FIG. 3A to FIG. 3D, the method for obtaining a semiconductor process recipe includes the following steps S101 to S104.

[0038] Step S101: A group of process parameters (including multiple process parameters) required for a process is randomly input into the constructed deep neural network model, and a process result evaluation index corresponding to the group of process parameters is obtained.

[0039] The deep neural network model is a trained multi-layer deep neural network, and the deep neural network model includes an input layer, at least one intermediate layer, and an output layer, where the input layer is used to input a set of process parameters, and the output layer is used to output process result evaluation indexes corresponding to the set of process parameters, that is, each of the input layer, the at least one intermediate layer, and the output layer is a one-layer deep neural network.

[0040] Specifically, before the above step S101, first, a deep neural network model is constructed, and as shown in FIG. 3B, the method for constructing the deep neural network model includes the following steps S201 to S202.

[0041] Step S201: The number of layers and the structure of a deep neural network model are determined according to the number of process parameters included in the process parameter group and the number of process result evaluation indexes corresponding to the process parameter group.

[0042] The deep neural network model includes an input layer, at least one intermediate layer, and an output layer, the number of intermediate layers is the difference between the number of process parameters included in the process parameter group and the number of process result evaluation indexes corresponding to the process parameter group, the deep neural network of each layer in the deep neural network model includes multiple neurons, one neuron is sequentially decreased between adjacent deep neural networks along the direction from the input layer to the output layer, each neuron is connected to all neurons in the deep neural network of the previous layer, each process parameter is the input feature of one neuron in the input layer, each neuron in the intermediate layer is used to calculate the output features of all neurons in the deep neural network of the previous layer, the output feature of each neuron in the output layer is the corresponding process result evaluation index, and the number of process parameters included in the process parameter group is greater than the number of process result evaluation indexes corresponding to the process parameter group.

[0043] Step S202: Provide an activation function to be executed in the deep neural network model according to the number of layers of the deep neural network model.

[0044] The deep neural network model of this embodiment is shown in Figure 4. Optionally, the deep neural network model building method further includes the following steps S203 to S204.

[0045] Step S203: Using the process database as training data, train a deep neural network model using a deep learning method.

[0046] The process database contains historical process parameter data and historical process result data for a particular process.

[0047] Step S204: Determine the weight terms and bias terms of the deep neural network of each layer in the deep neural network model according to the training result of the activation function, and finally complete the construction of the deep neural network model.

[0048] In the specific implementation process, for semiconductor processing machines such as etching equipment, the number of process result evaluation indexes is generally less than the number of process parameters. Therefore, the deep neural network model of the present invention can be constructed based on the neural network model construction method disclosed in Prior Art 1.

[0049] The number of layers of the neural network model is determined by the difference between the number of input variables (the number of process parameters) and the number of output variables (the number of process result evaluation indexes). One intermediate feature variable is gradually added between the deep neural networks of each layer. As shown in FIG. 4, in the case of an etching device, the input process parameter group has six process parameters, i.e., six variables, including chamber pressure, upper electrode power, lower electrode power, SF6 flow rate, O2 flow rate and etching time, and the finally obtained process result corresponding to the process parameter group has three evaluation indexes, including etching depth, sidewall angle and etching rate. Then, a deep neural network model with two layers of intermediate feature variables (three-layer deep neural network) is constructed.

[0050] The process database of process parameters and process results obtained during normal operation is used for deep learning training of the model, and the optimal model function y = h(Σ k w k g(Σ j w j f(Σ i w i x i +b1)+b2)+b3) can be obtained, i.e., a model relationship between the process parameters and the process results is obtained.

[0051] In a real problem, when the process result / process requirement is known, it is necessary to derive a suitable process recipe backwards, and as can be seen from the deep neural network model of the present invention, the number of indicators of the process result is less than the number of process parameters in the process recipe, so the problem is equivalent to solving an indeterminate equation, and theoretically there are infinite solutions; therefore, in order to quickly arrive at the optimal solution, the present invention uses a self-consistent iterative method to find the optimal solution.

[0052] Step S102: It is determined whether the degree of difference between each process result evaluation index and the corresponding process requirement satisfies the set requirement.

[0053] If step S103: YES, the input process parameter group is used as a process recipe for performing an actual process.

[0054] Specifically, in steps S101 to S103, first, one process parameter group (multiple process parameters x i ) is randomly given as the input feature of each neuron in the input layer of the deep neural network model, and the process result evaluation index y output from the output layer of the deep neural network model is expressed as the process requirement y 実際 If the set of process parameters satisfies the set requirements, the process is actually carried out using the set of process parameters.

[0055] The dissimilarity is calculated according to the following formula:

[0056] Δ=|(yy 実際 ) / y 実際 | In the formula, Δ is the difference between the process result index and the set process requirement, y is the process result index output from the deep neural network model, and y 実際 are the set process requirements, As a selectable set requirement, the degree of difference between the process result evaluation index and the corresponding set process requirement is 0 to 10%, and preferably 5%.

[0057] Step S104: If NO, according to the set process requirements, along the direction from the output layer to the input layer, the eigenvalues ​​in each intermediate layer and the numerical values ​​of each process parameter in the input layer are optimized layer by layer based on a gradient algorithm and a self-consistent iterative method, until a set of process parameters is obtained that enables all process result evaluation indexes output from the deep neural network model in the input layer to satisfy the corresponding set process requirements, and the set of process parameters that satisfy the process requirements is used as a process recipe for performing an actual process.

[0058] Specifically, as shown in FIG. 3C, the step of optimizing the eigenvalues ​​in each of the hidden layers and the process parameter values ​​in the input layer for each layer based on the gradient algorithm and the self-consistent iterative method includes the following steps S301 to S303.

[0059] Step S301: Find one process result evaluation index from the output layer that has the largest discrepancy with the correspondingly set process requirement.

[0060] Step S302: Find one process parameter from the hidden layer adjacent to the output layer that has the largest value of the weight term associated with the process result evaluation index with the largest degree of difference.

[0061] Step S303: Fix the weight term value and bias term value of the one process parameter with the largest weight term value, and use a gradient algorithm to optimize and step-adjust the eigenvalue of the one process parameter with the largest weight term value until the difference between the process result evaluation index with the largest difference in the output layer and the corresponding process requirement satisfies the set requirement.

[0062] After step S303, the following steps S304 to S307 are further included.

[0063] In step S303, it is determined whether the difference between the correspondingly adjusted process result evaluation index and the set process requirement satisfies the set requirement, and if so, step S304 is executed, and if not, step S306 is executed.

[0064] Step S304: At this time, if there are other process result evaluation indexes whose difference from the corresponding process requirements does not satisfy the set requirements, steps S301 to S303 are repeated until all process result evaluation indexes in the output layer satisfy the corresponding process requirements, and at the same time, the eigenvalues ​​of each process parameter in the intermediate layer are determined.

[0065] Step S305: Based on the eigenvalues ​​of each process parameter in the intermediate layer obtained in step S304, along the direction from the output layer to the input layer, continue to optimize the eigenvalues ​​of each process parameter in other intermediate layers and the numerical values ​​of the process parameters in the input layer by layer using the gradient algorithm and the self-consistent iterative method until a set of process parameters is obtained that enables all process result evaluation indexes output from the deep neural network model in the input layer to satisfy the corresponding set process requirements.

[0066] Step S306: Fix one process parameter corresponding to the weight term with the largest numerical value obtained by the adjustment and optimization in step S303, and find one process parameter with the second largest numerical value of the weight term related to the process result evaluation index correspondingly adjusted in step S303 from the hidden layer adjacent to the output layer.

[0067] Step S307: fix the value of the weight term of the one process parameter with the second largest weight term value, and use a gradient algorithm to optimize and step-adjust the eigenvalue of the one process parameter with the second largest weight term value until the difference between the process result evaluation index in the output layer correspondingly adjusted in step S303 and the corresponding set process requirement satisfies the set requirement.

[0068] In some other optional embodiments, after completing step S303, it is possible to directly execute step S304 without determining whether the discrepancy between the corresponding adjusted process result evaluation index in step S303 and the set process requirement satisfies the set requirement, and omit the above steps S306 and S307.

[0069] Preferably, after step S305, the method may further include step S308.

[0070] Step S308: after optimizing and adjusting all the process parameters related to the process result evaluation indexes, if there are still process result evaluation indexes in the output layer whose difference with the corresponding set process requirement does not meet the set requirement,

[0071] A new process parameter group is formed by selecting the process parameters that enable the sum of squares of the degree of difference between each process result evaluation index and the process requirement set corresponding to it to be the smallest, and the new process parameter group is set as a process recipe for performing an actual process.

[0072] In this embodiment, the calculation formula of the gradient algorithm is as follows:

[0073] x i =x i -θ×(∂y / ∂x i ) In the formula, x i is the adjusted process parameter in the intermediate layer, y is the adjusted process result evaluation index, and θ is the adjustment step.

[0074] In the specific implementation process of the above steps 301 to 308, first, the process result evaluation index y output from the output layer of the deep neural network model is converted into the process requirement y 実際 If the requirements are not met, the output process result indicator y and the process requirement y 実際 Difference Δ=|(yy 実際 ) / y 実際 According to |, find one process result evaluation index that has the largest difference from the set process requirements, and according to the magnitude of the weight term w of the process parameter group corresponding to the process result evaluation index, use a gradient algorithm x i =x i -θ×(∂y / ∂x i ) (θ is the step) i(a) is adjusted and repeated until the result is self-consistent, i.e., the process result obtained by the deep neural network model according to the given process parameters is approximately consistent with the process requirement (the output process result evaluation index only needs to reach 95% of the set process requirement, and the value can be 90%-100%).

[0075] If the process requirement cannot be met even after the process parameter with the largest weight term value is completely traversed according to the gradient algorithm, the process parameter with the largest weight term value obtained by the adjustment and optimization in step 303 is fixed, and the eigenvalue of the process parameter with the second largest weight term value is optimized according to the gradient algorithm, and so on. If the process requirement cannot be met even after all the process parameters are completely traversed, the process result evaluation index y and the process requirement y are calculated. 実際 In other words, a set of process parameters that enables the sum of squares of the differences between each process result evaluation index and the process requirement set corresponding thereto to be minimized is selected to form a new set of process parameters, and the new set of process parameters is used as a process recipe for performing the actual process.

[0076] Hereinafter, the process of finding the optimum solution by the self-consistent iterative method will be further described using an etching apparatus as an example.

[0077] First, a set of process parameters in the process recipe is randomly used and input into the deep neural network model to obtain process results in the output layer; then, find one process result evaluation index that is the most different from the set process requirements from the process results; for example, assume that the difference between the sidewall angle is the largest; then, search for one maximum weight term related to the sidewall angle in the deep neural network of the previous layer; then, fix the value of the weight term, and step-adjust the value of this weight term according to the gradient algorithm until the sidewall angle meets the requirement (for example, meets 95% of the set value); at this time, other process result evaluation indexes output from the model may also change, and if the other process result evaluation indexes also exceed the set process requirements, optimize the process result evaluation index in the same manner as above until all process result evaluation indexes meet the set process requirements; and at the same time, obtain the eigenvalues ​​of the process parameters in a set of intermediate layers adjacent to the value of the process result evaluation index; and similarly optimize the eigenvalues ​​of the process parameters in the intermediate layer in the same manner until the data of the above process parameter group in the process recipe, which is the leftmost input layer in FIG. 4, is obtained.

[0078] An embodiment of the present invention comprises: A deep neural network model is a trained multi-layered deep neural network, and is capable of outputting accurate process result evaluation indexes corresponding to a set of input process parameters based on the set of input process parameters; There is further provided a system for automatically acquiring a semiconductor process recipe, the system including: a calculation module for executing the method for acquiring a semiconductor process recipe according to the above embodiment.

[0079] The present invention further provides a semiconductor process apparatus including the semiconductor process recipe acquisition system according to the above embodiment.

[0080] The present invention can automatically generate a corresponding process recipe by software according to process requirements, improving the automation degree of the etching equipment, that is, the process engineers originally input process parameters according to their experience and repeatedly debug, and then the process requirements are directly input, and the corresponding process recipe is generated by software.

[0081] The present invention will be further explained and described below with reference to specific examples.

[0082] Specific Examples The present invention is applied to an etching apparatus, and a specific process is to realize an etching process with a critical dimension of 3 microns. The etching process with a critical dimension of 3 microns is shown in FIG.

[0083] Since the etching pattern needs to use the Bosch process, a plurality of process parameters x i include the chamber pressure of the deposition step, the top electrode center power of the deposition step, the top electrode edge power of the deposition step, the center C4F8 flow rate of the deposition step, the edge C4F8 flow rate of the deposition step, the deposition step start time, the deposition step start time, the chamber pressure of the etching step, the top electrode center power of the etching step, the top electrode edge power of the etching step, the bottom electrode start power of the etching step, the bottom electrode final power of the etching step, the center C4F8 flow rate of the etching step, the C4F8 flow rate of the etching step edge, the etching step start time, and the etching step final time, and the process evaluation index y includes the etching depth, the top opening dimension, the bottom opening dimension, the selectivity, the top sidewall roughness (scallop dimension), the bottom sidewall roughness (scallop dimension), etc.

[0084] Taking trench etching as an example, the deep neural network model of the above embodiment is trained according to the process database shown in FIGS.

[0085] From the process database shown in FIGS. 6 to 11, the values ​​of the weighting terms of the model function having the following trends can be trained.

[0086] The chamber pressure shows a high correlation with the etching morphology, and the weighting term has a large value. The lower electrode final (bias) power also shows a high correlation with the morphology, and the weighting term has a large value.

[0087] In addition to constructing the process database by ourselves in this embodiment, the process trends in the process database are also described in various documents, so it is also possible to construct a corresponding process database by referring to the documents.

[0088] After obtaining the trained model, in a recursive iteration manner, according to the process result shown in FIG. 12, a process recipe that meets the process requirements is derived backwards. For example, when deriving the etching angle backwards, the bottom electrode final power of etching and the final single-step time of etching have a large relationship (weight). According to the means of the present invention, these two process parameters are simultaneously optimized by gradient algorithm. The finally obtained process recipe is shown in Table 1.

[0089] [Table 1]

[0090] *The bottom electrode power is ramped from the starting value to the final value. **Single step time ramps from start value to final value.

[0091] For the hole etching process, we similarly construct the process database shown in Figures 13 to 18 to train the above deep neural network model, and after obtaining the trained model, we use a recursive iteration method to backwardly derive the process recipe according to the process result shown in Figure 19, as shown in Table 2.

[0092] [Table 2]

[0093] As described above, by using the means of the present invention, a process recipe can be automatically generated according to the process result in an etching device, and the degree of automation of the machine can be improved. Note that the present invention can be applied not only to etching devices, but also to other processes in the semiconductor field such as PVD, CVD, furnace tubes, cleaning machines, etc.

[0094] Although the embodiments of the present invention have been described above, the above description is illustrative and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for obtaining a semiconductor process recipe, comprising the steps of: inputting a set of process parameters required for a process into the constructed deep neural network model randomly, and obtaining a process result evaluation index corresponding to the set of process parameters, the set of process parameters including a plurality of process parameters, the deep neural network model being a trained multi-layer deep neural network, the deep neural network model including an input layer, at least one intermediate layer and an output layer, the input layer being used for inputting the set of process parameters, and the output layer being used for outputting the process result evaluation index corresponding to the set of process parameters; determining whether a difference between each of the process result evaluation indexes and a corresponding process requirement satisfies the requirement; if yes, using the set of process parameters as a process recipe for performing an actual process; if the result is NO, then optimizing, according to the process requirements, eigenvalues ​​in each of the intermediate layers and numerical values ​​of each of the process parameters in the input layer based on a gradient algorithm and a self-consistent iterative method along a direction from the output layer to the input layer until a set of process parameters that enables all process result evaluation indexes output from the deep neural network model in the input layer to satisfy the process requirements is obtained, and using the set of process parameters that satisfy the process requirements as a process recipe for performing an actual process.

2. The method for constructing a deep neural network model includes: determining the number of layers and the structure of the deep neural network model according to the number of process parameters included in the process parameter group and the number of the process result evaluation indexes corresponding to the process parameter group, the number of the intermediate layers is the difference between the number of the process parameters and the number of the process result evaluation indexes, the deep neural network of each layer in the deep neural network model includes a plurality of neurons, one neuron is sequentially decreased between adjacent deep neural networks along the direction from the input layer to the output layer, each neuron is connected to all neurons in the deep neural network of the previous layer, each process parameter is an input feature of one neuron in the input layer, each neuron in the intermediate layer is used to calculate output features of all neurons in the deep neural network of the previous layer, the output feature of each neuron in the output layer is the process result evaluation index, and the number of the process parameters included in the process parameter group is greater than the number of the process result evaluation indexes corresponding to the process parameter group; 2. The method for obtaining a semiconductor process recipe according to claim 1, further comprising: providing an activation function to be executed in the deep neural network model according to the number of layers of the deep neural network model.

3. The method for constructing a deep neural network model includes: training the deep neural network model using a deep learning method with a process database as training data, the process database including historical process parameter data and historical process result data for a particular process; 3. The method of claim 2, further comprising: determining weight terms and bias terms of the deep neural network of each layer in the deep neural network model according to a training result of the activation function.

4. The step of optimizing the eigenvalues ​​in each of the hidden layers and the values ​​of each process parameter in the input layer layer by layer based on a gradient algorithm and a self-consistent iterative method includes: A step S1 of finding one process result evaluation index from the output layer that has the largest difference from the process requirement; A step S2 of finding one process parameter having the largest value of the weight term associated with the process result evaluation index having the largest degree of difference from an intermediate layer adjacent to the output layer; Step S3: fix the weight term value and the bias term value of the one process parameter having the largest weight term value, and optimize and step-adjust the eigenvalue of the one process parameter having the largest weight term value using a gradient algorithm until the difference between the process result evaluation index having the largest difference in the output layer and the process requirement satisfies a set requirement; At this time, if there are other process result evaluation indexes whose difference with the process requirement does not satisfy the requirement set in the other process result evaluation indexes, the steps S1 to S3 are repeated until all the process result evaluation indexes in the output layer satisfy the process requirement, and at the same time, a step S4 is performed to determine an eigenvalue of each process parameter in the intermediate layer; and a step S5 of continuing to optimize, layer by layer, the eigenvalues ​​of each process parameter in other intermediate layers and the numerical values ​​of the process parameters in the input layer using a gradient algorithm and a self-consistent iterative method along a direction from the output layer to the input layer based on the eigenvalues ​​of each process parameter in the intermediate layer acquired in step S4, until a set of process parameters that enables all process result evaluation indexes output from the deep neural network model in the input layer to satisfy the process requirements is acquired.

5. After the step S3, it is determined whether the difference between the process result evaluation index correspondingly adjusted in the step S3 and the process requirement satisfies the set requirement, and if it does, the step S4 is executed; if it does not, the one process parameter having the largest value of the weight term obtained by the adjustment and optimization in the step S3 is fixed; Step S6: finding one process parameter from a hidden layer adjacent to the output layer that has the second largest value of the weight term associated with the process result evaluation index correspondingly adjusted in step S3; and performing a step S7 of fixing a value of a weight term of one process parameter having a second largest value of the weight term, and optimizing and step-adjusting an eigenvalue of the one process parameter having a second largest value of the weight term using a gradient algorithm until a difference between the process result evaluation index correspondingly adjusted in the step S3 in the output layer and the process requirement satisfies a set requirement.

6. After step S5, After optimizing and adjusting all the process parameters related to the process result index, if there is still a process result index in the output layer whose difference with the process requirement does not satisfy the set requirement, 6. The method for obtaining a semiconductor process recipe according to claim 5, further comprising the step of: forming a new group of process parameters by selecting process parameters that enable a sum of squares of the degree of difference between each process result evaluation index and the process requirement to be smallest, and setting the new group of process parameters as a process recipe for performing an actual process.

7. 7. The method for obtaining a semiconductor process recipe according to claim 1, wherein the gradient algorithm is calculated using the following formula: x i =x i -θ×(∂y / ∂x i ) (In the formula, x i is the adjusted process parameter in the intermediate layer, y is the adjusted process result index, and θ is the adjustment step.

8. The dissimilarity is calculated according to the following formula: The method for obtaining a semiconductor process recipe according to any one of claims 1, 4 to 6, characterized in that, as the set requirement, a difference between the process result evaluation index and the process requirement set corresponding thereto is 0 to 10%. Δ=|(y-y 実際 ) / y 実際 | (In the formula, Δ is the difference between the process result evaluation index and the set process requirement, y is the process result evaluation index output from the deep neural network model, y 実際 is the given process requirement.)

9. A system for automatically acquiring a semiconductor process recipe, comprising: A deep neural network model is a trained multi-layer deep neural network, and is capable of outputting accurate process result evaluation indexes corresponding to a set of process parameters based on input process parameters; A system for automatically acquiring a semiconductor process recipe, comprising: a calculation module for executing the method for acquiring a semiconductor process recipe according to any one of claims 1 to 8.

10. A semiconductor process device comprising the system for automatically acquiring a semiconductor process recipe according to claim 9.

Citation Information

Patent Citations

  • Magnetron film plating instrument process parameter optimizing method based on genetic algorithm and BP neural network

    CN110262233A

  • Etching effect prediction method and input parameter determination method

    JP2019083306A

  • Method and process for performing machine learning on complex multivariate wafer processing equipment

    JP2019537240A

  • Production specification determination method, production method, and production specification determination device for metal material

    WO2020148917A1