Etching process evaluation method and system based on artificial intelligence

Abstract

The present invention provides an artificial intelligence-based etching process evaluation method and system, which relate to the field of artificial intelligence technology. In the present invention, first, a photolithography pattern semantic vector is determined; secondly, a vector space conversion unit included in an etching process evaluation network is used to perform a space conversion operation on the photolithography pattern semantic vector, and a corresponding photolithography pattern conversion vector is output; then, based on the photolithography pattern conversion vector, a plurality of local etching processes included in a target etching process are jointly mined and evaluated by an etching process evaluation unit included in an etching process evaluation network, and target process evaluation data corresponding to the target etching process is output. Based on the above content, the problem of relatively low efficiency and reliability of etching process evaluation existing in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an etching process evaluation method and system based on artificial intelligence. Background Art

[0002] Flexible Printed Circuit (FPC) is an electronic structural component based on a flexible substrate (such as polyimide or polyester film, etc.), which is widely used in various electronic products, such as smart phones, wearable devices, medical instruments, etc. Among them, in the manufacturing process of FPC, etching (etching is the process of removing the unprotected part of copper or other metal materials to form the required circuit pattern) and other processes are crucial steps, which affect the accuracy of the circuit, the protection of the material and the overall performance of the circuit. Therefore, it is necessary to conduct a corresponding evaluation of the etching process. However, in the prior art, the evaluation is generally based on manual evaluation, which makes it easy to have the problem of low evaluation efficiency. In addition, due to the limitation of the experience and knowledge of the evaluators, there is also the problem of low reliability. Summary of the invention

[0003] In view of this, an object of the present invention is to provide an etching process evaluation method and system based on artificial intelligence to improve the problem of relatively low efficiency and reliability of etching process evaluation in the prior art.

[0004] To achieve the above object, the present invention adopts the following technical solution:

[0005] An etching process evaluation method based on artificial intelligence, comprising:

[0006] Determine a photolithography pattern semantic vector, wherein the photolithography pattern semantic vector is formed based on a photolithography pattern semantic space or formed by a photolithography pattern mining unit in an etching process evaluation network performing a semantic mining operation on a target photolithography pattern, the photolithography pattern semantic space is a space where semantic vectors formed by the photolithography pattern mining unit performing a semantic mining operation on each training photolithography pattern after training is completed, and the target photolithography pattern is a pattern formed by an exposure process and is a pattern that needs to be etched by a target etching process to be evaluated;

[0007] Performing a space conversion operation on the photolithography pattern semantic vector through a vector space conversion unit included in the etching process evaluation network, and outputting a corresponding photolithography pattern conversion vector;

[0008] According to the photolithography pattern conversion vector, the etching process evaluation unit included in the etching process evaluation network performs joint mining and evaluation operations on the multiple local etching processes included in the target etching process, and outputs target process evaluation data corresponding to the target etching process.

[0009] In a preferred option of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, the etching process evaluation unit includes a joint mining subunit and a process evaluation subunit;

[0010] Among them, the step of performing joint mining and evaluation operations on multiple local etching processes included in the target etching process through the etching process evaluation unit included in the etching process evaluation network according to the photolithography pattern conversion vector, and outputting the target process evaluation data corresponding to the target etching process includes:

[0011] For each local etching process among the multiple local etching processes included in the target etching process, the following operations are performed on the local etching process:

[0012] By means of the joint mining subunit, the local etching process is subjected to M stages of joint mining operations according to the lithography pattern conversion vector, and an etching process joint vector of the Mth stage is output, wherein data processed by the joint mining operation of the n+1th stage includes the etching process joint vector of the nth stage, and n+1 is less than or equal to M;

[0013] Through the process evaluation subunit, the etching process joint vector is mapped and outputted to form local process evaluation data corresponding to the local etching process;

[0014] After obtaining the local process evaluation data corresponding to each of the local etching processes, the target process evaluation data corresponding to the target etching process is output based on the local process evaluation data corresponding to each of the local etching processes.

[0015] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, the step of performing M-stage joint mining operations on the local etching process according to the photolithography pattern conversion vector through the joint mining subunit and outputting the etching process joint vector of the M-th stage includes:

[0016] Determining an etching process semantic vector corresponding to the local etching process, wherein the etching process semantic vector is obtained by performing a word embedding operation on the local etching process;

[0017] Through the joint mining subunit, based on the lithography pattern conversion vector and the etching process semantic vectors corresponding to other local etching processes, the etching process semantic vector corresponding to the local etching process is subjected to M stages of joint mining operations, and the etching process joint vector of the Mth stage is output.

[0018] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, the step of performing M-stage joint mining operations on the etching process semantic vector corresponding to the local etching process based on the photolithography pattern conversion vector and the etching process semantic vectors corresponding to other local etching processes through the joint mining subunit, and outputting the etching process joint vector of the Mth stage includes:

[0019] In the joint mining operation of the first stage, the center vector of the etching process semantic vector corresponding to each other local etching process is determined, and based on the vector association parameter between the center vector and the etching process semantic vector corresponding to the local etching process, the etching process semantic vector corresponding to the local etching process is optimized to form the etching process optimization vector of the first stage, and the etching process optimization vector is optimized by using the vector association parameter between the photolithography pattern conversion vector and the etching process optimization vector to form the etching process joint vector of the first stage;

[0020] In the joint mining operation of the second stage and each subsequent stage, internal semantic optimization operation is performed on the internal optimization vector of the previous stage of each other local etching process, and the internal optimization vector of the current stage of each other local etching process is output, and internal semantic optimization operation is performed on the photolithography pattern optimization vector of the previous stage, and the photolithography pattern optimization vector of the current stage is output, wherein the internal optimization vector of the first stage is the corresponding etching process semantic vector, and the photolithography pattern optimization vector of the first stage is the photolithography pattern conversion vector;

[0021] Determine the center vector of the internal optimization vector of each other local etching process in the current stage, and optimize the etching process joint vector based on the vector association parameters between the center vector and the etching process joint vector of the previous stage to form the etching process optimization vector of the current stage, and optimize the etching process optimization vector using the vector association parameters between the lithography pattern optimization vector of the current stage and the etching process optimization vector to form the etching process joint vector of the current stage.

[0022] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, before the step of performing joint mining and evaluation operations on a plurality of local etching processes included in the target etching process according to the photolithography pattern conversion vector by the etching process evaluation unit included in the etching process evaluation network, and outputting the target process evaluation data corresponding to the plurality of local etching processes, the etching process evaluation method based on artificial intelligence also includes:

[0023] Determine a training first lithography pattern, a training first pattern semantic vector corresponding to the training first lithography pattern, and Q stages configured for the training first lithography pattern, wherein the training first pattern semantic vector is formed by other lithography pattern mining units performing semantic mining operations on the training first lithography pattern, the other lithography pattern mining units do not belong to the etching process evaluation network, and Q is less than or equal to M;

[0024] By means of the lithography pattern mining unit, a semantic mining operation is performed on the trained first lithography pattern, and a corresponding first lithography pattern semantic vector is output;

[0025] By means of the joint mining subunit, the first lithography pattern semantic vector is subjected to Q stages of interference application operations, and a corresponding lithography pattern interference vector is output;

[0026] Through the joint mining subunit, the lithography pattern interference vector is subjected to Q stages of interference removal operations according to the trained first pattern semantic vector, and a corresponding interference information removal vector is output, wherein the interference removal operation refers to performing a joint mining operation on the trained first pattern semantic vector and the lithography pattern interference vector to remove interference information;

[0027] Through the first output branch in the process evaluation subunit, the interference information removal vector is subjected to a pattern restoration operation, and a corresponding first restored photolithography pattern is output, wherein the process evaluation subunit further includes a second output branch, and the second output branch is used to perform a mapping output operation on the etching process joint vector to form corresponding local process evaluation data;

[0028] By means of the other photolithography pattern mining unit, the first restored photolithography pattern is subjected to a semantic mining operation, and a corresponding first restored pattern semantic vector is output;

[0029] Determining a corresponding first error index according to the trained first pattern semantic vector and the first restored pattern semantic vector;

[0030] The joint mining subunit is updated according to the first error indicator.

[0031] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, the step of determining the corresponding first error index based on the trained first pattern semantic vector and the first restored pattern semantic vector includes:

[0032] Analyzing corresponding error adjustment parameters according to the Q stages, wherein there is a negative correlation between the error adjustment parameters and the Q stages;

[0033] Determining a vector difference index between the trained first pattern semantic vector and the first restored pattern semantic vector;

[0034] Based on the error adjustment parameter, the vector difference indicator is adjusted and a corresponding first error indicator is output.

[0035] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, the step of updating the joint mining subunit according to the first error index includes:

[0036] The network parameters of the lithography pattern mining unit and the process evaluation sub-unit are maintained unchanged, and the network parameters of the joint mining sub-unit are updated according to the first error index so that the first error index converges.

[0037] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, before the step of performing a semantic mining operation on the trained first lithography pattern through the lithography pattern mining unit and outputting the corresponding first lithography pattern semantic vector, the etching process evaluation method based on artificial intelligence includes:

[0038] Determining a training second lithography pattern;

[0039] By means of the lithography pattern mining unit, the training second lithography pattern is subjected to a semantic mining operation, and a corresponding second lithography pattern semantic vector is output;

[0040] Performing a pattern restoration operation on the second photolithography pattern semantic vector through the first output branch in the process evaluation subunit, and outputting a corresponding second restored photolithography pattern;

[0041] Obtaining a second error indicator based on the trained second photolithography pattern and the second restored photolithography pattern;

[0042] Based on the second error indicator, network parameters of the lithography pattern mining unit and the process evaluation subunit are updated so that the second error indicator converges.

[0043] In a preferred embodiment of the present invention, in the above-mentioned etching process evaluation method based on artificial intelligence, after the step of updating the joint mining subunit according to the first error index, the etching process evaluation method based on artificial intelligence further includes:

[0044] Determine a training etching process, a training third photolithography pattern, and a third photolithography pattern semantic vector corresponding to the training third photolithography pattern, wherein the third photolithography pattern semantic vector is formed by performing a semantic mining operation on the training third photolithography pattern by the photolithography pattern mining unit;

[0045] Performing a space conversion operation on the third lithography pattern semantic vector by the vector space conversion unit, and outputting a corresponding third lithography pattern conversion vector;

[0046] For each local training etching process among the multiple local training etching processes included in the training etching process, the following operations are performed on the local training etching process:

[0047] By means of the joint mining subunit, the local training etching process is subjected to M stages of joint mining operations according to the third photolithography pattern conversion vector, and a joint vector of the training etching process of the Mth stage is output;

[0048] Through the process evaluation subunit, the training etching process joint vector is mapped and outputted to form local process evaluation data corresponding to the local training etching process;

[0049] After obtaining the local process evaluation data corresponding to each of the local training etching processes, outputting the training process evaluation data corresponding to the training etching process based on the local process evaluation data corresponding to each of the local training etching processes;

[0050] Based on the target error index between the training process evaluation data and the actual process evaluation data corresponding to the training etching process, the network parameters of the lithography pattern mining unit, the vector space conversion unit, the joint mining sub-unit and the process evaluation sub-unit are updated so that the target error index converges.

[0051] On the basis of the above, the present invention also provides an etching process evaluation system based on artificial intelligence, comprising:

[0052] Memory for storing computer programs;

[0053] A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-mentioned etching process evaluation method based on artificial intelligence.

[0054] The artificial intelligence-based etching process evaluation method and system provided by the present invention firstly determines the semantic vector of the photolithography pattern; secondly, the vector space conversion unit included in the etching process evaluation network performs a space conversion operation on the semantic vector of the photolithography pattern, and outputs the corresponding photolithography pattern conversion vector; then, according to the photolithography pattern conversion vector, the etching process evaluation unit included in the etching process evaluation network performs a joint mining and evaluation operation on the various local etching processes included in the target etching process, and outputs the target process evaluation data corresponding to the target etching process. Based on the above content, on the one hand, the powerful learning ability and data analysis ability of the neural network can be used to perform process evaluation, so that compared with the conventional technical solution based on manual evaluation, it can have higher efficiency and reliability. On the other hand, since the target etching process is jointly mined and evaluated in combination with the lithography pattern conversion vector, the basis of the evaluation operation not only has the semantic information of the target etching process, but also has the semantic information of the target lithography pattern or the training lithography pattern. There is a certain correlation between the etching process and the lithography pattern. Therefore, the basis of the evaluation operation has richer and more reliable semantic information, so that the reliability of the target process evaluation data obtained by evaluation can be higher. Therefore, based on the solution provided by the present invention, the problem of relatively low efficiency and reliability of etching process evaluation in the prior art can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings.

[0056] Figure 1 A structural block diagram of an etching process evaluation system based on artificial intelligence provided in an embodiment of the present invention.

[0057] Figure 2 A schematic flow chart of an etching process evaluation method based on artificial intelligence provided in an embodiment of the present invention.

[0058] Figure 3 A schematic diagram of a joint mining operation in M ​​stages provided in an embodiment of the present invention.

[0059] Figure 4 A schematic diagram of Q-stage interference removal operations provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0061] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] like Figure 1 As shown, an embodiment of the present invention provides an etching process evaluation system based on artificial intelligence. Wherein, the etching process evaluation system based on artificial intelligence may include a memory and a processor. In detail, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The processor is used to execute an executable computer program stored in the memory to implement the etching process evaluation method based on artificial intelligence provided by an embodiment of the present invention.

[0063] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc.

[0064] Furthermore, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0065] Understandably, Figure 1 The structure shown is for illustration only. The etching process evaluation system based on artificial intelligence may also include Figure 1 More or fewer components as shown, or with Figure 1 The different configurations shown, for example, may also include a communication unit for information exchange with other devices. In addition, the etching process evaluation system based on artificial intelligence may be a server or a server cluster.

[0066] Combination Figure 2 The embodiment of the present invention further provides an artificial intelligence-based etching process evaluation method that can be applied to the above-mentioned artificial intelligence-based etching process evaluation system. Among them, the method steps defined in the process related to the artificial intelligence-based etching process evaluation method can be implemented by the artificial intelligence-based etching process evaluation system (hereinafter referred to as the etching process evaluation system). Figure 2 The specific process shown is explained in detail.

[0067] Step S110, determining a lithography pattern semantic vector.

[0068] In an embodiment of the present invention, the etching process evaluation system can determine the semantic vector of the photolithography pattern. The semantic vector of the photolithography pattern is formed based on the semantic space of the photolithography pattern or formed by the photolithography pattern mining unit in the etching process evaluation network performing semantic mining operations on the target photolithography pattern. The photolithography pattern semantic space is the space where the semantic vector formed by the photolithography pattern mining unit performing semantic mining operations on each training photolithography pattern after the training is completed (for example, the photolithography pattern semantic vector can be the mean vector of each semantic vector, etc.), and the target photolithography pattern is a pattern formed by the exposure process and belongs to the pattern to be etched by the target etching process to be evaluated, that is, after the target photolithography pattern is formed, the corresponding etching can be performed by the target etching process to obtain the required circuit pattern. In addition, the target photolithography pattern can be obtained by image acquisition of the circuit board after the exposure process, and the target etching process can be text data designed by the corresponding personnel or text data generated by other equipment. For example, a target etching process can include: etching copper with ferric chloride (FeCl3) at a concentration of 35%; during the copper etching process, the temperature should be maintained at 45°C; and the etching time is controlled to be 2 to 3 minutes. In other embodiments, the target etching process may further include more data, such as etching solution flow rate and stirring method, line width and spacing, substrate thickness and etching depth, etc.

[0069] Step S120, performing a space conversion operation on the lithography pattern semantic vector through the vector space conversion unit included in the etching process evaluation network, and outputting a corresponding lithography pattern conversion vector.

[0070] In an embodiment of the present invention, after determining the semantic vector of the lithography pattern, the etching process evaluation system can perform a spatial conversion operation on the semantic vector of the lithography pattern through the vector space conversion unit included in the etching process evaluation network, and output the corresponding lithography pattern conversion vector. That is to say, since the semantic information corresponding to the lithography image and the etching process is generally not in the same semantic space, in order to facilitate reliable joint mining and evaluation operations in subsequent steps, the lithography pattern semantic vector can be subjected to a spatial conversion operation. For example, the vector space conversion unit can include a weight matrix, an offset matrix, and a nonlinear activation function, so that the weight matrix and the lithography pattern semantic vector can be multiplied, and the multiplication result and the offset matrix are added, and the addition result is processed based on the nonlinear activation function to obtain the corresponding lithography pattern conversion vector. In addition, the weight matrix, the offset matrix, and the nonlinear activation function can be multiple groups. After a group of weight matrices, offset matrices, and nonlinear activation functions are processed accordingly, the processing result of the previous group can be processed again by a later group of weight matrices, offset matrices, and nonlinear activation functions, thereby forming a lithography pattern conversion vector.

[0071] Step S130, based on the lithography pattern conversion vector, the etching process evaluation unit included in the etching process evaluation network jointly mines and evaluates the multiple local etching processes included in the target etching process, and outputs target process evaluation data corresponding to the target etching process.

[0072] In an embodiment of the present invention, after outputting the photolithography pattern conversion vector, the etching process evaluation system can perform joint mining and evaluation operations on the multiple local etching processes included in the target etching process through the etching process evaluation unit included in the etching process evaluation network according to the photolithography pattern conversion vector, and output the target process evaluation data corresponding to the target etching process. Exemplarily, the target process evaluation data can be a value in the continuous interval of 0-1, and the larger the value, the higher the process reliability represented by the corresponding process evaluation data. In addition, the target process evaluation data can also be classified as better, average, or poor.

[0073] Based on the above content, on the one hand, the powerful learning ability and data analysis ability of the neural network can be used to perform process evaluation, so that compared with the conventional technical solution based on manual evaluation, it can have higher efficiency and reliability. On the other hand, because the target etching process is also combined with the lithography pattern conversion vector to perform joint mining and evaluation operations, the basis of the evaluation operation not only has the semantic information of the target etching process, but also has the semantic information of the target lithography pattern or the training lithography pattern, and there is a certain correlation between the etching process and the lithography pattern. Therefore, the basis of the evaluation operation has relatively rich and reliable semantic information, so that the reliability of the target process evaluation data obtained by the evaluation can also be higher. Therefore, based on the solution provided by the present invention, the problem of relatively low efficiency and reliability of etching process evaluation in the prior art can be improved.

[0074] It should be noted that for the above step S130, the specific manner of jointly mining and evaluating the multiple local etching processes included in the target etching process is not limited and can be selected according to actual needs.

[0075] For example, in an alternative embodiment, the photolithography pattern conversion vector and the semantic vectors corresponding to the multiple local etching processes included in the target etching process can be spliced ​​to form a corresponding splicing vector, and then the splicing vector can be evaluated to obtain target process evaluation data corresponding to the target etching process.

[0076] For example, in another alternative embodiment, in order to improve the reliability of the joint mining and evaluation operations, the reliability of the obtained target process evaluation data can be higher, and the etching process evaluation unit includes a joint mining sub-unit and a process evaluation sub-unit, so that the above-mentioned step S130 can further include step S131 and step S132, and the specific contents are described as follows.

[0077] Step S131, for each local etching process among the multiple local etching processes included in the target etching process, the following operations are performed on the local etching process: through the joint mining sub-unit, the local etching process is subjected to M stages of joint mining operations according to the lithography pattern conversion vector, and the etching process joint vector of the Mth stage is output; through the process evaluation sub-unit, the etching process joint vector is mapped and output, so as to form local process evaluation data corresponding to the local etching process.

[0078] In an embodiment of the present invention, for each of the multiple local etching processes included in the target etching process (such as etching solution concentration, etching temperature, etching time, etc.), the following operations can be performed on the local etching process: first, the local etching process can be subjected to M stages of joint mining operations according to the photolithography pattern conversion vector through the joint mining subunit, and the etching process joint vector of the Mth stage can be output; secondly, the etching process joint vector can be mapped and output through the process evaluation subunit to form local process evaluation data corresponding to the local etching process. Among them, the data processed by the joint mining operation of the n+1th stage includes the etching process joint vector of the nth stage, that is, the data output by the joint mining operation of the previous stage can be used as the input data of the joint mining operation of the next stage. For example, the data output by the joint mining operation of the first stage can be used as the input data of the joint mining operation of the second stage, n+1 is less than or equal to M, and M is greater than or equal to 2. In addition, the process evaluation subunit may include a convolutional network, a pooling network, a fully connected network and a classification function (such as softmax, etc.). In this way, the convolutional network can process the etching process joint vector of the Mth stage (or the mean vector of the etching process joint vectors of each stage, etc.) to obtain the corresponding convolution vector, the pooling network can process the convolution vector to obtain the corresponding pooling vector, and then, the fully connected network can process the pooling vector to obtain the corresponding fully connected vector. Finally, the fully connected vector can be processed by the classification function to obtain the probabilities of each type, and then, the type with the highest probability can be used as local process evaluation data.

[0079] Step S132, after obtaining the local process evaluation data corresponding to each of the local etching processes, outputting the target process evaluation data corresponding to the target etching process based on the local process evaluation data corresponding to each of the local etching processes.

[0080] In an embodiment of the present invention, after obtaining the local process evaluation data corresponding to each of the local etching processes, the target process evaluation data corresponding to the target etching process can be output based on the local process evaluation data corresponding to each of the local etching processes. For example, the local process evaluation data corresponding to each of the local etching processes can be averaged to obtain the target process evaluation data corresponding to the target etching process.

[0081] In the above-mentioned step S131, the specific method of performing M-stage joint mining operations on the local etching process according to the lithography pattern conversion vector is not limited, and can also be configured accordingly according to actual needs. For example, in an alternative embodiment, in order to ensure the reliability of the joint mining operation so that the obtained etching process joint vector has better semantic representation capabilities, the above-mentioned step S131 can further include step S131 and step S132, and the specific contents of each step are described as follows.

[0082] Step S131, determining an etching process semantic vector corresponding to the local etching process.

[0083] In an embodiment of the present invention, the etching process semantic vector corresponding to the local etching process can be determined. In addition, the etching process semantic vectors corresponding to each other local etching process can also be determined separately. Among them, the etching process semantic vector is obtained by performing a word embedding operation on the local etching process. For example, the joint mining sub-unit or the etching process evaluation network also includes a word embedding model. In this way, the local etching process can be word-embedded by the word embedding model to obtain the word vectors of each word, and then the word vectors of each word can be spliced ​​or averaged to obtain the corresponding etching process semantic vector, or the result of splicing or average calculation can be self-attention processed to obtain the corresponding etching process semantic vector, or the word vectors of each word can be added to the position vector of each word (such as the embedding vector of the position information), and then splicing or average calculation can be performed. For example, for the local etching process "control the etching time to 2 to 3 minutes", each word vector can be:

[0084] Control: [0.23, 0.12, 0.45, 0.67, 0.89, 0.23, −0.45, 0.56, −0.12, 0.33, 0.77, −0.44, 0.11, 0.34, −0.67, 0.22, 0.88, −0.99, 0.15, 0.10, 0.55, −0.76, …, 0.43];

[0085] Etch: [0.31, 0.24, 0.37, 0.41, 0.76, 0.33, −0.28, 0.63, −0.22, 0.48, 0.59, −0.35, 0.17, 0.25, −0.33, 0.49, 0.76, −0.62, 0.08, 0.52, 0.13, −0.84, …, 0.61];

[0086] time: [0.12, 0.56, 0.42, 0.28, 0.63, 0.12, 0.19, 0.46, −0.15, 0.29, 0.53, −0.19, 0.35, 0.23, −0.67, 0.44, 0.91, −0.73, 0.18, 0.31, 0.27, −0.82, …, 0.12];

[0087] is: [0.04, 0.22, 0.09, 0.17, 0.33, 0.02, −0.15, 0.25, 0.02, 0.18, 0.27, −0.10, 0.12, 0.03, −0.22, 0.21, 0.36, −0.29, 0.16, 0.27, 0.15, −0.65, …, 0.08];

[0088] 2: [0.19, 0.27, 0.14, 0.36, 0.49, 0.23, −0.12, 0.28, 0.17, 0.48, 0.54, −0.33, 0.27, 0.17, −0.53, 0.39, 0.66, −0.44, 0.12, 0.29, 0.27, −0.68, …, 0.75];

[0089] ~: [0.38, 0.27, 0.17, 0.32, 0.49, 0.27, 0.12, 0.25, 0.19, 0.41, 0.54, −0.32, 0.27, 0.15, −0.53, 0.39, 0.76, 0.44, 0.12, 0.29, 0.23, −0.69, …, 0.34];

[0090] 3: [0.58, 0.26, 0.15, 0.32, 0.49, 0.13, −0.12, 0.24, 0.19, 0.41, 0.54, −0.31, 0.27, 0.11, 0.57, 0.36, 0.66, −0.44, 0.12, 0.23, 0.22, −0.68, …, 0.25];

[0091] minutes: [0.11, 0.43, 0.36, 0.29, 0.59, 0.25, 0.02, 0.35, 0.08, 0.44, 0.48, −0.24, 0.21, 0.15, 0.31, 0.48, 0.55, −0.62, 0.19, 0.38, 0.23, −0.72, …, 0.41].

[0092] Step S132, through the joint mining subunit, based on the lithography pattern conversion vector and the etching process semantic vectors corresponding to other local etching processes, the etching process semantic vector corresponding to the local etching process is subjected to M-stage joint mining operations, and the etching process joint vector of the Mth stage is output.

[0093] In an embodiment of the present invention, after obtaining the etching process semantic vector corresponding to the local etching process, the etching process semantic vector corresponding to the local etching process can be subjected to M stages of joint mining operations based on the photolithography pattern conversion vector and the etching process semantic vectors corresponding to other local etching processes through the joint mining subunit, and the etching process joint vector of the Mth stage is output. In other words, not only is it necessary to fuse the semantic information in the photolithography pattern conversion vector into the etching process semantic vector, but also, considering that different local etching processes may also have correlations, the etching process semantic vectors corresponding to other local etching processes may be fused into the etching process semantic vector corresponding to the local etching process, so that the obtained etching process joint vector can have richer and more reliable semantic information, and can more accurately semantically represent the corresponding local etching process.

[0094] In the above step S132, the specific method of performing M-stage joint mining operations on the etching process semantic vector corresponding to the local etching process based on the photolithography pattern conversion vector and the etching process semantic vector corresponding to other local etching processes is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to mine semantic information with depth, so that the semantic representation ability of the etching process joint vector can be better, and in order to make the semantic information in the photolithography pattern conversion vector and the etching process semantic vector corresponding to other local etching processes more reliably integrated into the etching process semantic vector corresponding to the local etching process, the above step S132 can further include the following content (combined with Figure 3 ):

[0095] First, in the joint mining operation of the first stage, the central vector of the etching process semantic vector corresponding to each other local etching process can be determined (for example, the mean of each etching process semantic vector can be calculated, and the obtained mean vector can be used as the central vector, or the etching process semantic vector with the greatest similarity to the mean vector can be used as the central vector), and based on the vector association parameters between the central vector and the etching process semantic vector corresponding to the local etching process (for example, the dot product of the central vector and the transposed vector of the etching process semantic vector can be performed to obtain the corresponding vector association parameters, wherein the vector association parameters can be a A matrix, for example, a row vector with a center vector of 1*a, a transposed vector of an etching process semantic vector of a*1, and a dot product can be performed to obtain an a*a matrix, i.e., a vector association parameter), to optimize the etching process semantic vector corresponding to the local etching process to form an etching process optimization vector of the first stage (for example, the etching process semantic vector can be weighted summed based on the vector association parameter to obtain the corresponding etching process optimization vector, for example, for each row in the vector association parameter, the parameters in the etching process semantic vector are weighted summed based on the parameters of the row to obtain a parameter in the etching process optimization vector, such as Therefore, the information associated with the central vector can be mined from the etching process semantic vector, that is, the etching process semantic vector can be optimized), and the etching process optimization vector can be optimized using the vector association parameters between the photolithography pattern conversion vector and the etching process optimization vector (the same as the aforementioned processing procedure) to form the first stage of the etching process joint vector; in addition, it should be noted that before determining the vector association parameters between the vectors and optimizing them, the corresponding vectors can also be mapped, and the mapping parameters can be determined by the joint mining subunit in the process of learning the training data. In this way, since the mapping can be achieved through different learning The process captures the complex relationship between inputs, especially in high-dimensional space, the model can learn the nonlinear relationship between inputs through linear transformation, so more reliable joint mining can be achieved; for example, the first parameter matrix and the photolithography pattern conversion vector can be multiplied (parameters at corresponding positions are multiplied) to obtain a first output matrix, and the second parameter matrix and the third parameter matrix can be multiplied with the etching process optimization vector respectively to obtain a second output matrix and a third output matrix, then, the dot product between the transposed matrices of the first output matrix and the second output matrix can be calculated, and then, the third output matrix can be weighted summed based on the dot product to obtain an etching process joint vector;

[0096] Secondly, in the joint mining operation of the second stage and each subsequent stage (such as the joint mining operation of the second stage, the joint mining operation of the third stage, the joint mining operation of the fourth stage, etc.), the internal semantic optimization operation is performed on the internal optimization vector of the previous stage of each other local etching process, and the internal optimization vector of the current stage of each other local etching process is output, and the internal semantic optimization operation is performed on the photolithography pattern optimization vector of the previous stage, and the photolithography pattern optimization vector of the current stage is output, wherein the internal optimization vector of the first stage is the corresponding etching process semantic vector, and the photolithography pattern optimization vector of the first stage is the photolithography pattern conversion vector; the internal semantic optimization operation may refer to multiplying the internal optimization vector of the previous stage with the three parameter matrices in the joint mining sub-unit (which may be different from the three parameter matrices in the aforementioned steps), and then performing dot product calculation and weighted sum calculation in the above manner to obtain the internal optimization vector of the current stage, and the photolithography pattern optimization vector may also be obtained based on the same manner;

[0097] Then, the central vector of the internal optimization vector of each other local etching process in the current stage can be determined, and based on the vector association parameters between the central vector and the etching process joint vector of the previous stage, the etching process joint vector is optimized to form the etching process optimization vector of the current stage, and the etching process optimization vector is optimized using the vector association parameters between the photolithography pattern optimization vector of the current stage and the etching process optimization vector to form the etching process joint vector of the current stage, as described above. In this way, the etching process joint vector of the second stage, the etching process joint vector of the third stage, ..., the etching process joint vector of the last stage, that is, the etching process joint vector of the Mth stage, can be obtained in sequence.

[0098] It should be noted that in order to ensure that the etching process evaluation network has the above-mentioned semantic information mining and evaluation output capabilities, that is, to ensure the reliable execution of the above-mentioned steps S110, S120 and S130, the artificial intelligence-based etching process evaluation method may also include the step of training to form the etching process evaluation network. In an alternative embodiment, the step may specifically include step S101, step S102, step S103, step S104, step S105, step S106, step S107 and step S108, and the specific content of each step is described as follows.

[0099] Step S101 : determining a training first lithography pattern, a training first pattern semantic vector corresponding to the training first lithography pattern, and Q stages configured for the training first lithography pattern.

[0100] In an embodiment of the present invention, a training first lithography pattern, a training first pattern semantic vector corresponding to the training first lithography pattern, and Q stages configured for the training first lithography pattern can be determined. The training first pattern semantic vector is formed by other lithography pattern mining units performing semantic mining operations on the training first lithography pattern, and the other lithography pattern mining units do not belong to the etching process evaluation network. That is, the etching process evaluation network can be assisted in training based on other lithography pattern mining units other than the etching process evaluation network, thereby improving the training efficiency. Specifically, the other lithography pattern mining units can be units used for semantic mining in other trained neural networks. For example, a network for evaluating pattern quality may be trained based on other needs, and the network also needs to perform semantic mining operations on the pattern, such as a convolutional neural network. In addition, Q is less than or equal to M. For example, Q can be 5 and M can be 10.

[0101] Step S102: performing semantic mining operation on the trained first lithography pattern through the lithography pattern mining unit, and outputting a corresponding first lithography pattern semantic vector.

[0102] In an embodiment of the present invention, after obtaining the first trained photolithography pattern, the photolithography pattern mining unit may perform semantic mining operations on the first trained photolithography pattern to output a corresponding first photolithography pattern semantic vector. For example, the photolithography pattern mining unit may include a convolution subunit, a pooling subunit, etc. Thus, the first trained photolithography pattern may be processed by the convolution subunit and the pooling subunit to obtain a first photolithography pattern semantic vector. For example, in a specific application scenario, the photolithography pattern mining unit is composed as follows:

[0103] Convolutional layer C1 (first convolutional layer):

[0104] Number of convolution kernels: 6 convolution kernels (filters); convolution kernel size: 5x5; stride: 1; padding: no padding (validpadding);

[0105] Pooling layer S2 (first pooling layer):

[0106] Type: Average Pooling; Pooling Window: 2x2; Stride: 2;

[0107] Convolutional layer C3 (second convolutional layer):

[0108] Number of convolution kernels: 16; convolution kernel size: 5x5; stride: 1; padding: no padding (validpadding);

[0109] Pooling layer S4 (second pooling layer):

[0110] Type: Average Pooling; Pooling Window: 2x2; Stride: 2.

[0111] Step S103 : performing Q stages of interference application operations on the first lithography pattern semantic vector through the joint mining subunit, and outputting a corresponding lithography pattern interference vector.

[0112] In an embodiment of the present invention, after obtaining the first lithography pattern semantic vector, the first lithography pattern semantic vector can be subjected to Q stages of interference application operations through the joint mining subunit to output a corresponding lithography pattern interference vector. Exemplarily, Q random vectors can be generated, and then the Q random vectors are sequentially applied to the first lithography image semantic vector to obtain a corresponding lithography pattern interference vector; exemplarily, the first random vector and the first lithography image semantic vector can be added to obtain a first lithography pattern interference vector, and the second random vector and the first lithography pattern interference vector can be added to obtain a second lithography image interference vector.

[0113] Step S104 : performing Q stages of interference removal operations on the lithography pattern interference vector according to the trained first pattern semantic vector through the joint mining subunit, and outputting a corresponding interference information removal vector.

[0114] In an embodiment of the present invention, after obtaining the lithography image interference vector, the joint mining subunit can perform Q stages of interference removal operations on the lithography pattern interference vector according to the trained first pattern semantic vector, and output a corresponding interference information removal vector. The interference removal operation refers to performing a joint mining operation on the trained first pattern semantic vector and the lithography pattern interference vector to remove interference information. For example, in combination Figure 4In the interference removal operation of the first stage, the last lithography pattern interference vector can be optimized based on the vector association parameters between the training first pattern semantic vector and the last lithography image interference vector to obtain the corresponding optimization vector, and the optimization vector can be optimized based on the vector association parameters between the training first pattern semantic vector and the optimization vector to obtain the interference information removal vector of the first stage. In the interference removal operation of the second stage and each subsequent stage, the corresponding lithography image interference vector (such as the penultimate lithography image interference vector) can be added or weighted summed with the interference information removal vector of the previous stage to obtain the corresponding addition vector, and then, based on the vector association parameters between the training first pattern semantic vector and the addition vector, the addition vector is optimized to obtain the corresponding optimization vector, and, based on the vector association parameters between the training first pattern semantic vector and the optimization vector, the optimization vector is optimized to obtain the interference information removal vector of the current stage, until the interference information removal vector of the last stage is obtained. It should be noted that the interference removal operation is partially different from the joint mining operation in the aforementioned steps, but the main calculation process is the same. Therefore, after training, only a simple adaptive adjustment of the joint mining subunit is required.

[0115] Step S105 , performing a pattern restoration operation on the interference information removal vector through the first output branch in the process evaluation subunit, and outputting a corresponding first restored photolithography pattern.

[0116] In an embodiment of the present invention, the interference information removal vector can be subjected to a pattern restoration operation through the first output branch in the process evaluation subunit, and a corresponding first restored photolithography pattern can be output. The first output branch can be a decoding neural network, and image decoding can be performed based on the interference information removal vector to obtain a first restored photolithography pattern. In addition, the process evaluation subunit also includes a second output branch, and the second output branch is used to map and output the etching process joint vector to form corresponding local process evaluation data, as described above.

[0117] Step S106: performing semantic mining operation on the first restored photolithography pattern through the other photolithography pattern mining unit, and outputting a corresponding first restored pattern semantic vector.

[0118] In an embodiment of the present invention, after obtaining the first restored lithography pattern, the other lithography pattern mining unit can perform semantic mining operations on the first restored lithography pattern (as mentioned above, it can be convolution and other processing) to output the corresponding first restored pattern semantic vector.

[0119] Step S107: determining a corresponding first error index according to the trained first pattern semantic vector and the first restored pattern semantic vector.

[0120] In an embodiment of the present invention, after obtaining the first restored pattern semantic vector, a corresponding first error index can be determined based on the training first pattern semantic vector and the first restored pattern semantic vector, wherein the first error index can be used to reflect the difference between the training first pattern semantic vector and the first restored pattern semantic vector.

[0121] Step S108: updating the joint mining subunit according to the first error indicator.

[0122] In the embodiment of the present invention, after obtaining the first error indicator, the joint mining subunit may be updated according to the first error indicator, such as adjusting corresponding network parameters.

[0123] In the above step S107, the specific method of determining the corresponding first error index is not limited and can be selected according to actual needs. For example, in an alternative implementation, in order to ensure that the determined first error index has a high reliability, the above step S107 can further include the following contents:

[0124] First, the corresponding error adjustment parameters can be analyzed according to the Q stages, wherein the error adjustment parameters have a negatively correlated corresponding relationship with the Q stages; illustratively, since the interference information removal vectors of multiple stages can be obtained, multiple first restored lithography patterns and multiple first restored pattern semantic vectors can be obtained, and for each stage, the ratio between the current stage and Q can be calculated, and then the difference between the target value and the ratio is calculated to obtain the corresponding error adjustment parameter, wherein the specific numerical value of the target value is not limited, for example, the target value can be 1;

[0125] Secondly, a vector difference index between the training first pattern semantic vector and the first restored pattern semantic vector may be determined; illustratively, a mean square error or a root mean square error between the training first pattern semantic vector and the first restored pattern semantic vector may be calculated;

[0126] Finally, the vector difference index may be adjusted based on the error adjustment parameter to output a corresponding first error index; illustratively, a weighted sum calculation may be performed on each vector difference index based on the error adjustment parameter to obtain the corresponding first error index.

[0127] In the above step S108, the specific method of updating the joint mining subunit is not limited and can be selected according to actual needs. For example, in an alternative implementation, in order to ensure that the updating of the joint mining subunit has a high reliability, the above step S108 can further include the following contents:

[0128] The network parameters of the lithography pattern mining unit and the process evaluation sub-unit are maintained unchanged, and the network parameters of the joint mining sub-unit are updated according to the first error index so that the first error index converges, that is, only the network parameters of the joint mining sub-unit are updated, that is, the network parameters of the joint mining sub-unit are updated in the direction of reducing the first error index, such as updating the included parameter matrix.

[0129] It should be noted that, in the above-mentioned implementation, the joint mining subunit is trained accordingly, and the specific method of training other network structures is not limited. For example, in an alternative implementation, the network parameters of the lithography pattern mining unit and the process evaluation subunit can be updated to achieve corresponding training, thereby ensuring that the etching process evaluation network can reliably achieve the corresponding functions. The specific process of training is as follows:

[0130] First, a training second lithography pattern can be determined;

[0131] Secondly, the photolithography pattern mining unit may perform semantic mining operations on the training second photolithography pattern to output a corresponding second photolithography pattern semantic vector;

[0132] Then, the second photolithography pattern semantic vector may be subjected to a pattern restoration operation through the first output branch in the process evaluation subunit, and a corresponding second restored photolithography pattern may be output;

[0133] Secondly, a second error index, such as a mean square error or a root mean square error, may be obtained based on the training second lithography pattern and the second restored lithography pattern;

[0134] Finally, based on the second error index, the network parameters of the lithography pattern mining unit and the process evaluation subunit can be updated so that the second error index converges; for example, the network parameters of the lithography pattern mining unit and the process evaluation subunit can be updated in the direction of reducing the second error index.

[0135] It should be noted that the training process in the above two stages is actually unsupervised training, which can reduce the number of annotations for training data and improve efficiency. However, in some alternative implementations, in order to ensure that the etching process evaluation network formed by training can have reliable joint mining and evaluation capabilities, the step of training the etching process evaluation network can also include the following contents (in some other implementations, training can also be directly based on the following contents without the above two stages of unsupervised training):

[0136] First, a training etching process, a training third lithography pattern, and a third lithography pattern semantic vector corresponding to the training third lithography pattern can be determined, wherein the third lithography pattern semantic vector is formed by the lithography pattern mining unit performing a semantic mining operation on the training third lithography pattern, as described above;

[0137] Then, the vector space conversion unit may perform a space conversion operation on the third lithography pattern semantic vector to output a corresponding third lithography pattern conversion vector, as described above;

[0138] Further, for each local training etching process among the multiple local training etching processes included in the training etching process, the following operations are performed on the local training etching process:

[0139] The joint mining subunit can perform M-stage joint mining operations on the local training etching process according to the third lithography pattern conversion vector, and output the joint vector of the training etching process at the Mth stage; the process evaluation subunit can perform a mapping output operation on the training etching process joint vector to form local process evaluation data corresponding to the local training etching process, as described above;

[0140] Afterwards, after obtaining the local process evaluation data corresponding to each of the local training etching processes, the training process evaluation data corresponding to the training etching process may be output based on the local process evaluation data corresponding to each of the local training etching processes, as described above;

[0141] Finally, based on the target error index between the training process evaluation data and the actual process evaluation data corresponding to the training etching process (for example, when the training process evaluation data belongs to the classification result, the corresponding cross entropy loss can be calculated as the target error index), the network parameters of the lithography pattern mining unit, the vector space conversion unit, the joint mining sub-unit and the process evaluation sub-unit can be updated so that the target error index converges; that is, the network parameters can be updated in the direction of reducing the target error index.

[0142] It should be noted that, in the above-mentioned implementations, convergence means that the corresponding error index is reduced to the configured threshold or the reduction is less than the preset value, or the number of updates is greater than the preset number, which can be selected according to actual conditions.

[0143] In summary, the etching process evaluation method and system based on artificial intelligence provided by the present invention firstly determines the semantic vector of the lithography pattern; secondly, the vector space conversion unit included in the etching process evaluation network performs a space conversion operation on the semantic vector of the lithography pattern, and outputs the corresponding lithography pattern conversion vector; then, according to the lithography pattern conversion vector, the etching process evaluation unit included in the etching process evaluation network performs a joint mining and evaluation operation on the various local etching processes included in the target etching process, and outputs the target process evaluation data corresponding to the target etching process. Based on the above content, on the one hand, the powerful learning ability and data analysis ability of the neural network can be used to perform process evaluation, so that compared with the conventional technical solution based on manual evaluation, it can have higher efficiency and reliability. On the other hand, since the target etching process is jointly mined and evaluated in combination with the lithography pattern conversion vector, the basis of the evaluation operation not only has the semantic information of the target etching process, but also has the semantic information of the target lithography pattern or the training lithography pattern. There is a certain correlation between the etching process and the lithography pattern. Therefore, the basis of the evaluation operation has richer and more reliable semantic information, so that the reliability of the target process evaluation data obtained by evaluation can be higher. Therefore, based on the solution provided by the present invention, the problem of relatively low efficiency and reliability of etching process evaluation in the prior art can be improved.

[0144] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0145] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0146] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0147] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An etching process evaluation method based on artificial intelligence, characterized in that: include: Determine a photolithography pattern semantic vector, wherein the photolithography pattern semantic vector is formed based on a photolithography pattern semantic space or formed by a photolithography pattern mining unit in an etching process evaluation network performing a semantic mining operation on a target photolithography pattern, the photolithography pattern semantic space is a space where semantic vectors formed by the photolithography pattern mining unit performing a semantic mining operation on each training photolithography pattern after training is completed, and the target photolithography pattern is a pattern formed by an exposure process and is a pattern that needs to be etched by a target etching process to be evaluated; Performing a space conversion operation on the photolithography pattern semantic vector through a vector space conversion unit included in the etching process evaluation network, and outputting a corresponding photolithography pattern conversion vector; For each of the multiple local etching processes included in the target etching process, the following operations are performed on the local etching process: an etching process semantic vector corresponding to the local etching process is determined, wherein the etching process semantic vector is obtained by performing a word embedding operation on the local etching process; based on the photolithography pattern conversion vector and the etching process semantic vectors corresponding to other local etching processes, the etching process semantic vector corresponding to the local etching process is subjected to M-stage joint mining operations by a joint mining subunit included in an etching process evaluation unit included in the etching process evaluation network, and an etching process joint vector of the Mth stage is output, wherein data processed by the joint mining operation of the n+1th stage includes the etching process joint vector of the nth stage, and n+1 is less than or equal to M; the etching process joint vector is mapped and output by a process evaluation subunit included in the etching process evaluation unit to form local process evaluation data corresponding to the local etching process; after obtaining the local process evaluation data corresponding to each of the local etching processes, the target process evaluation data corresponding to the target etching process is output based on the local process evaluation data corresponding to each of the local etching processes.

2. The etching process evaluation method based on artificial intelligence according to claim 1, characterized in that: The step of performing M-stage joint mining operations on the etching process semantic vector corresponding to the local etching process based on the photolithography pattern conversion vector and the etching process semantic vectors corresponding to other local etching processes by the joint mining subunit included in the etching process evaluation unit included in the etching process evaluation network, and outputting the etching process joint vector of the Mth stage includes: In the joint mining operation of the first stage, the center vector of the etching process semantic vector corresponding to each other local etching process is determined, and based on the vector association parameter between the center vector and the etching process semantic vector corresponding to the local etching process, the etching process semantic vector corresponding to the local etching process is optimized to form the etching process optimization vector of the first stage, and the etching process optimization vector is optimized by using the vector association parameter between the photolithography pattern conversion vector and the etching process optimization vector to form the etching process joint vector of the first stage; In the joint mining operation of the second stage and each subsequent stage, internal semantic optimization operation is performed on the internal optimization vector of the previous stage of each other local etching process, and the internal optimization vector of the current stage of each other local etching process is output, and internal semantic optimization operation is performed on the photolithography pattern optimization vector of the previous stage, and the photolithography pattern optimization vector of the current stage is output, wherein the internal optimization vector of the first stage is the corresponding etching process semantic vector, and the photolithography pattern optimization vector of the first stage is the photolithography pattern conversion vector; Determine the center vector of the internal optimization vector of each other local etching process in the current stage, and optimize the etching process joint vector based on the vector association parameters between the center vector and the etching process joint vector of the previous stage to form the etching process optimization vector of the current stage, and optimize the etching process optimization vector using the vector association parameters between the lithography pattern optimization vector of the current stage and the etching process optimization vector to form the etching process joint vector of the current stage.

3. The etching process evaluation method based on artificial intelligence according to claim 1, characterized in that: The etching process evaluation method based on artificial intelligence also includes: Determine a training first lithography pattern, a training first pattern semantic vector corresponding to the training first lithography pattern, and Q stages configured for the training first lithography pattern, wherein the training first pattern semantic vector is formed by other lithography pattern mining units performing semantic mining operations on the training first lithography pattern, the other lithography pattern mining units do not belong to the etching process evaluation network, and Q is less than or equal to M; By means of the lithography pattern mining unit, a semantic mining operation is performed on the trained first lithography pattern, and a corresponding first lithography pattern semantic vector is output; By means of the joint mining subunit, the first lithography pattern semantic vector is subjected to Q stages of interference application operations, and a corresponding lithography pattern interference vector is output; Through the joint mining subunit, the lithography pattern interference vector is subjected to Q stages of interference removal operations according to the trained first pattern semantic vector, and a corresponding interference information removal vector is output, wherein the interference removal operation refers to performing a joint mining operation on the trained first pattern semantic vector and the lithography pattern interference vector to remove interference information; Through the first output branch in the process evaluation subunit, the interference information removal vector is subjected to a pattern restoration operation, and a corresponding first restored photolithography pattern is output, wherein the process evaluation subunit further includes a second output branch, and the second output branch is used to perform a mapping output operation on the etching process joint vector to form corresponding local process evaluation data; By means of the other photolithography pattern mining unit, the first restored photolithography pattern is subjected to a semantic mining operation, and a corresponding first restored pattern semantic vector is output; Determining a corresponding first error index according to the trained first pattern semantic vector and the first restored pattern semantic vector; The joint mining subunit is updated according to the first error indicator.

4. The etching process evaluation method based on artificial intelligence according to claim 3 is characterized in that: The step of determining a corresponding first error index based on the trained first pattern semantic vector and the first restored pattern semantic vector comprises: Analyzing corresponding error adjustment parameters according to the Q stages, wherein there is a negative correlation between the error adjustment parameters and the Q stages; Determining a vector difference index between the trained first pattern semantic vector and the first restored pattern semantic vector; Based on the error adjustment parameter, the vector difference indicator is adjusted and a corresponding first error indicator is output.

5. The etching process evaluation method based on artificial intelligence according to claim 3 is characterized in that: The step of updating the joint mining subunit according to the first error indicator includes: The network parameters of the lithography pattern mining unit and the process evaluation sub-unit are maintained unchanged, and the network parameters of the joint mining sub-unit are updated according to the first error index so that the first error index converges.

6. The etching process evaluation method based on artificial intelligence according to claim 3 is characterized in that: Before the step of performing semantic mining operation on the trained first lithography pattern by the lithography pattern mining unit and outputting the corresponding first lithography pattern semantic vector, the etching process evaluation method based on artificial intelligence includes: Determining a training second lithography pattern; By means of the lithography pattern mining unit, the training second lithography pattern is subjected to a semantic mining operation, and a corresponding second lithography pattern semantic vector is output; Performing a pattern restoration operation on the second photolithography pattern semantic vector through the first output branch in the process evaluation subunit, and outputting a corresponding second restored photolithography pattern; Obtaining a second error indicator based on the trained second photolithography pattern and the second restored photolithography pattern; Based on the second error indicator, network parameters of the lithography pattern mining unit and the process evaluation subunit are updated so that the second error indicator converges.

7. The etching process evaluation method based on artificial intelligence according to claim 3 is characterized in that: After the step of updating the joint mining subunit according to the first error indicator, the etching process evaluation method based on artificial intelligence further includes: Determine a training etching process, a training third photolithography pattern, and a third photolithography pattern semantic vector corresponding to the training third photolithography pattern, wherein the third photolithography pattern semantic vector is formed by performing a semantic mining operation on the training third photolithography pattern by the photolithography pattern mining unit; Performing a space conversion operation on the third lithography pattern semantic vector by the vector space conversion unit, and outputting a corresponding third lithography pattern conversion vector; For each local training etching process among the multiple local training etching processes included in the training etching process, the following operations are performed on the local training etching process: By means of the joint mining subunit, the local training etching process is subjected to M stages of joint mining operations according to the third photolithography pattern conversion vector, and a joint vector of the training etching process of the Mth stage is output; Through the process evaluation subunit, the training etching process joint vector is mapped and outputted to form local process evaluation data corresponding to the local training etching process; After obtaining the local process evaluation data corresponding to each of the local training etching processes, outputting the training process evaluation data corresponding to the training etching process based on the local process evaluation data corresponding to each of the local training etching processes; Based on the target error index between the training process evaluation data and the actual process evaluation data corresponding to the training etching process, the network parameters of the lithography pattern mining unit, the vector space conversion unit, the joint mining sub-unit and the process evaluation sub-unit are updated so that the target error index converges.

8. An etching process evaluation system based on artificial intelligence, characterized in that: include: Memory for storing computer programs; A processor connected to the memory is used to execute a computer program stored in the memory to implement the artificial intelligence-based etching process evaluation method described in any one of claims 1 to 7.

Images