CN115598937B - Photoetching mask shape prediction method and device and electronic equipment - Google Patents
Photoetching mask shape prediction method and device and electronic equipment Download PDFInfo
- Publication number
- CN115598937B CN115598937B CN202211592524.5A CN202211592524A CN115598937B CN 115598937 B CN115598937 B CN 115598937B CN 202211592524 A CN202211592524 A CN 202211592524A CN 115598937 B CN115598937 B CN 115598937B
- Authority
- CN
- China
- Prior art keywords
- light source
- mask
- neuron
- source data
- wafer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G03—PHOTOGRAPHY; CINEMATOGRAPHY; ANALOGOUS TECHNIQUES USING WAVES OTHER THAN OPTICAL WAVES; ELECTROGRAPHY; HOLOGRAPHY
- G03F—PHOTOMECHANICAL PRODUCTION OF TEXTURED OR PATTERNED SURFACES, e.g. FOR PRINTING, FOR PROCESSING OF SEMICONDUCTOR DEVICES; MATERIALS THEREFOR; ORIGINALS THEREFOR; APPARATUS SPECIALLY ADAPTED THEREFOR
- G03F7/00—Photomechanical, e.g. photolithographic, production of textured or patterned surfaces, e.g. printing surfaces; Materials therefor, e.g. comprising photoresists; Apparatus specially adapted therefor
- G03F7/70—Microphotolithographic exposure; Apparatus therefor
- G03F7/70483—Information management; Active and passive control; Testing; Wafer monitoring, e.g. pattern monitoring
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention discloses a photoetching mask shape prediction method and device and electronic equipment. The photoetching mask shape prediction method provided by the invention utilizes a pulse neural network based on brain-like calculation to combine self-adaptive optimization to carry out photoetching mask shape prediction; a light source generated by a photoetching system irradiates the convergent lens through projection type exposure, so that the convergent lens is exposed and developed on a mask and then is projected onto a wafer through the mask; by predicting the shape of the photoetching mask before the light beam reaches the wafer, the corresponding adjustment and optimization of the chip design pattern and the adjustment and optimization of the initial irradiation light source can be preferentially carried out in the layout design stage, so that the light finally irradiated on the wafer meets the photoetching requirement.
Description
Technical Field
The invention relates to the technical field of semiconductor design and manufacture, in particular to a photoetching mask shape prediction method and device and electronic equipment.
Background
The photolithography technology is a core process in the chip manufacturing process, and before the photolithography process starts, a chip design pattern needs to be copied onto a mask plate through a specific device, and then light with a specific wavelength is generated through the photolithography device to copy the chip design pattern on the mask plate onto a wafer for producing chips. However, due to the non-linear effect in the optical system, the mask and the photoresist system, a pattern distortion phenomenon occurs during the process of transferring the chip design pattern to the wafer, an optical proximity effect is generated, and if the pattern distortion phenomenon is not eliminated, the whole manufacturing technology may fail.
In the prior art, optimization of a chip design pattern is mainly performed based on conventional Optical or Optical and chemical reactions, including Optical Proximity Correction (OPC), source Mask Optimization (SMO), sub-Resolution-Assist-Feature (SRAF), and the like, to maximally reduce Optical Proximity effects caused by nonlinear effects such as optics. For example, optimization at the optical level, or optimization at the physical level (addition of inert gas slows down the reaction of certain steps or slows down the passing light/electrons, etc.); or by means of some chemical additives to slow down the diffraction/superposition effects of light. These techniques are dependent on (or limited to) the practical experience of the relevant practitioner or the precision/maintenance/advancement of the manufacturing equipment to some extent, and require a large amount of auxiliary optimization with reference to the existing template (which additionally requires many parameters), and the shortcomings of the prior art are highlighted in the era of the increasingly updated moore law, particularly in the process of 10nm or less.
In the industry, there are also some techniques for predicting the shape of a Lithography Mask (Lithography Mask) by machine learning or deep learning, and then optimizing a chip design pattern. But instead. This is achieved by means of machine learning (or deep learning, reinforcement learning), which requires a very large number of data sets (training sets, test sets, validation sets), which are the most lacking in the lithography field (mask, etch, OPC, etc.), so that the resulting prediction model is lacking in terms of accuracy or speed.
Disclosure of Invention
The invention aims to provide a photoetching mask shape prediction method, a photoetching mask shape prediction device and electronic equipment, wherein photoetching mask shape prediction is completed through a pulse neural network and a self-adaptive optimization mode, generation and training of a prediction model are not needed to be performed by more training data sets, prediction accuracy and speed can be ensured, and the photoetching mask shape prediction is realized so as to well optimize a chip design graph.
In order to achieve the above object, the present invention provides a method for predicting a shape of a lithography mask, comprising the steps of: 1) Receiving a test pulse signal from a light source and collecting test light source data of different stages, wherein the test light source data of different stages comprises: first light source data when the light source passes through the convergent lens and does not reach the mask, second light source data when the light source passes through the convergent lens and does not reach the wafer, third light source data when the light source passes through the mask and does not reach the wafer, and fourth light source data when the light source passes through the mask and irradiates the wafer; 2) Inputting the first light source data into neurons of a spiking neural network; 3) Acquiring different neuron modes in a pulse time sequence by adopting a self-adaptive optimization learning method based on the second light source data, the third light source data and the fourth light source data; 4) Receiving a test pulse signal from a light source for multiple times, and iteratively executing the steps 1) to 3) to obtain a neuron mode of a corresponding iteration round number; and 5) predicting the shape of the photoetching mask after the target light source irradiates the mask based on all the acquired neuron modes.
In order to achieve the above object, the present invention further provides a lithographic mask shape prediction apparatus, comprising: the data acquisition module is used for receiving test pulse signals from the light source and acquiring test light source data of different stages, wherein the test light source data of different stages comprise: first light source data when the light source passes through the convergent lens and does not reach the mask, second light source data when the light source passes through the convergent lens and does not reach the wafer, third light source data when the light source passes through the mask and does not reach the wafer, and fourth light source data when the light source passes through the mask and irradiates the wafer; the data input module is used for inputting the first light source data into neurons of a pulse neural network; a first pattern obtaining module, configured to obtain different neuron patterns in a pulse time sequence by using a self-adaptive optimization learning method based on the second light source data, the third light source data, and the fourth light source data; the second mode acquisition module is used for receiving a test pulse signal from a light source for multiple times, and iteratively executing the calling of the data acquisition module, the data input module and the first mode acquisition module to acquire the neuron mode of the corresponding iteration round number; and the prediction module is used for predicting the shape of the photoetching mask after the mask is irradiated by the target light source based on all the obtained neuron modes.
To achieve the above object, the present invention further provides an electronic device, which includes a memory, a processor and a computer executable program stored in the memory and running on the processor, wherein the processor executes the computer executable program to implement the steps of the method for predicting the shape of the lithography mask according to the present invention.
The invention adopts a method of combining a pulse neural network based on a brain-like intelligent computing framework with a self-adaptive optimization technology to complete the acquisition of a neuron mode so as to predict the shape of various light sources irradiated on a wafer, and then predicts the pattern irradiated on a mask, so that the corresponding adjustment optimization can be preferentially carried out on a chip design graph and the adjustment optimization can be carried out on an initial irradiation light source in a layout design stage, and the light finally irradiated on the wafer meets the photoetching requirement. The invention omits the traditional OPC, SMO, SRAF and other degrees depending on the experience of related practitioners or the precision/maintenance/advancement and the like of preparation equipment, and can play the role of artificial intelligence in the scene with smaller process. The invention solves the technical problem that the accuracy or speed of the obtained prediction model is deficient due to the lack of a large amount of data sets in the traditional photoetching mask shape prediction method based on machine learning or deep learning; meanwhile, the calculation overhead and the performance overhead are saved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that are required to be used in the description of the embodiments will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 is a schematic diagram illustrating a method for predicting a shape of a lithography mask according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a light source at various stages according to an embodiment of the invention;
fig. 3 is a schematic diagram of an impulse neural network combined with an ant colony optimization algorithm according to an embodiment of the present invention;
fig. 4 is a block diagram of a lithography mask shape prediction apparatus according to an embodiment of the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It is to be understood that the disclosed embodiments are merely exemplary of the invention, and are not intended to be exhaustive or exhaustive. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
An embodiment of the invention provides a method for predicting a shape of a lithography mask.
Fig. 1 to fig. 3 are also shown, in which fig. 1 is a schematic diagram illustrating steps of a photolithography mask shape prediction method according to an embodiment of the present invention, fig. 2 is a schematic diagram illustrating different stages of a light source according to an embodiment of the present invention, and fig. 3 is a schematic diagram illustrating a pulse neural network and an ant colony optimization algorithm according to an embodiment of the present invention.
As shown in fig. 1, the method of this embodiment includes the following steps: s1, receiving a test pulse signal from a light source and collecting test light source data of different stages, wherein the test light source data of different stages comprise: first light source data when the light source passes through the convergent lens and does not reach the mask, second light source data when the light source passes through the convergent lens and does not reach the wafer, third light source data when the light source passes through the mask and does not reach the wafer, and fourth light source data when the light source passes through the mask and irradiates the wafer; s2, inputting the first light source data into a neuron of a pulse neural network; s3, acquiring different neuron modes in a pulse time sequence by adopting a self-adaptive optimization learning method based on the second light source data, the third light source data and the fourth light source data; s4, receiving a test pulse signal from a light source for multiple times, and iteratively executing the steps S1 to S3 to obtain a neuron mode of a corresponding iteration round number; and S5, predicting the shape of the photoetching mask after the mask is irradiated by the target light source based on all the obtained neuron modes.
Regarding step S1, test pulse signals from the light sources are received and test light source data at different stages are collected. Specifically, the test light source data of the different stages includes: first Light source data (CL-Light-Before, CL-LB) Before the Light source 20 passes through the converging lens (ConvergentLens, CL) 21, second Light source data (CL-Light-After, CL-LA) when the Light source passes through the converging lens 21 and does not reach the Mask (Mask) 22, third Light source data (Light-After-Mask, LA-M) when the Light source passes through the Mask 22 and does not reach the Wafer 29, and fourth Light source data (Light-At-Wafer, LA-W) when the Light source passes through the Mask 22 and irradiates the Wafer 29, as shown in fig. 2. In fig. 2, two sets of projection lenses between wafer 29 and mask 22: a projection lens 23 adjacent to the mask 22 and a projection lens 24 adjacent to the wafer 29. A light source generated by a photoetching system irradiates the convergent lens through projection type exposure, so that the convergent lens is exposed and developed on a mask and then is projected onto a wafer through the mask. In this embodiment, four stages of light source data are collected.
In some embodiments, the light source is a DUV pulsed signal generated by an excimer laser of a DUV lithography system. The light source can be a DUV pulse signal with the wavelength of 193nm, and the DUV pulse signal irradiates the wafer after passing through the mask in the form of a beam-converging light. The excimer laser may be an ArF excimer laser. In an ArF excimer laser, ar + and F-ions are first generated therein, and then pulses of high voltage are applied to the Ar + and F-ions so that the ions are bonded together to form an ArF excimer. When the atomic states in the excimer decay from the excited state to the ground state, the DUV emits energy outward. At this time, the excimer laser is emitted in a short pulse form (projected onto the mask). The firing pulse emission will continue to generate a pulsed signal as long as the voltage is sufficient and continuous. Converting the pulse signal into a corresponding phase spectrogram, where the phase spectrogram signal includes information in a specific time range or in a specific operation scenario, for example, including: phase, initial shape (concave/convex, etc.), etc. Based on the acquisition of the test light source data at different stages, a first batch of training data participating in the training of the impulse neural network may be obtained. In general, in order to complete the irradiation of a mask pattern onto a wafer, many light source irradiation, development, light diffraction/interference/superposition, standing wave effect elimination, and the like are required, which requires the light source to continuously irradiate many times. Optionally, the wafer requires N lithographic shots, where N >0 and N ∈ positive integer. In other embodiments, the excimer laser may also be another KrF-based excimer laser.
In some embodiments, the first light source data CL-LB, the second light source data CL-LA, the third light source data LA-M, and the fourth light source data LA-W each include: focal depth, resolution, detection control chip, alignment precision, numerical aperture of optical system, exposure wavelength, pulse signal, light source type, and light source irradiation mode; the second light source data CL-LA and the third light source data LA-M further include: diffraction/interference deformation; the fourth light source data LA-W further includes: diffraction/interference deformation, the wavelength and focal length of two groups of projection lenses between the wafer and the mask, the distance between the two groups of projection lenses, the specification of the wafer, the size of the photoresist on the wafer, and the distance between the photoresist and the adjacent projection lenses.
In the first stage, as shown in fig. 2, the light source is collected before passing through the lithography mask, specifically before passing through the converging lens 21. The light source 20 is a DUV pulsed signal generated by an excimer laser of a DUV lithography system. Before the light source 20 passes through the converging lens 21, the light source parameters that can be collected include: depth of focus (DOF), resolution, various detection control chips (Monitor Chip), overlay accuracy, numerical Aperture (NA) of an optical system, exposure wavelength (λ) I, pulse signal I (time domain/frequency domain, in this case, initial pulse signal), light source type, light source irradiation method, and the like. The light source illumination mode includes on-axis illumination, off-axis illumination (OAI), and the like.
In the second stage, the light source passing through the converging lens 21 and reaching the mask 22 is collected. When the light source 20 passes through the converging lens 21 and does not reach the mask 22, the light source parameters that can be collected include: depth of focus (DOF), resolution, various detection control chips (Monitor Chip), overlay accuracy, numerical Aperture (NA) of an optical system, exposure wavelength (λ) II, pulse signal II (time domain/frequency domain, in this case, pulse signal after passing through a converging lens), light source type, light source irradiation method, and the like. The light source illumination mode includes on-axis illumination, off-axis illumination (OAI), and the like. In addition, after passing through the converging lens 21, the light source 20 generates light source data subjected to deformation such as diffraction and interference; thus, the second light source data acquired has more diffractive/interferometric deformation II than the first light source data.
In the third stage, the light source is collected when it has passed through the mask 22 and has not reached the wafer 29. When the light source 20 passes through the mask 22 and does not reach the wafer 29, the light source parameters that can be collected include: depth of focus (DOF), resolution, various detection control chips (Monitor Chip), overlay accuracy, numerical Aperture (NA) of an optical system, exposure wavelength (λ) III, pulse signal III (time domain/frequency domain, in this case, pulse signal after mask), light source type, light source irradiation method, diffraction/interference distortion III, and the like. The light source illumination mode includes on-axis illumination, off-axis illumination (OAI), and the like. That is, the kind of the light source data collected in the second stage and the third stage is substantially the same.
In the fourth stage, the light source irradiated to the wafer 29 after passing through the mask 22 is collected. In this embodiment, after passing through the mask 22, the light source 20 passes through two sets of projection lenses 23 and 24, and then irradiates the wafer 29, and the light source parameters that can be collected at this stage include: depth of focus (DOF), resolution, various detection control chips (Monitor Chip), overlay accuracy, numerical Aperture (NA) of an optical system, exposure wavelength (λ) IV, pulse signal IV (time domain/frequency domain, in this case, pulse signal irradiated onto a wafer via a projection lens), light source type, light source irradiation method, diffraction/interference deformation IV, and the like. The light source illumination mode includes on-axis illumination, off-axis illumination (OAI), and the like. Before the light source 20 irradiates the photoresist (not shown) on the wafer 29, it passes through two sets of projection lenses 23, 24; therefore, compared with the third light source data, the acquired fourth light source data has more wavelengths and focal lengths of two groups of projection lenses between the wafer and the mask, a distance between the two groups of projection lenses, a specification of the wafer, a size of a photoresist on the wafer, and a distance between the photoresist and an adjacent projection lens.
With respect to step S2, the first light source data is input to neurons of a spiking neural network. Specifically, the first illuminant data of the first stage of the data acquisition stage is transmitted into neurons of a Spiking Neural Network (SNN), and since the SNN is a time-series based (with spatiotemporal characteristics) pulse series, different neurons store different information (wavelength, phase, concave/convex shape, etc.). Once a neuron is stimulated by external stimulus (e.g., input of a pulse signal) in a certain time sequence, the neuron continuously generates waveforms with certain or different wavelengths, and a range of nerve synapses of the neuron are stimulated to deform, so that pulse sequences forming a certain intrinsic pattern (pattern) are gathered together to distinguish different pulse neurons (classification of neurons). As a result, neurons with a particular kind of information are fired by weight (w) linking of these synapses, as shown in FIG. 3. Different neurons carry different information.
With respect to step S3, different neuron patterns in the pulse time series are acquired using an adaptive optimization learning method based on the second light source data, the third light source data, and the fourth light source data. Specifically, according to the input data (first to fourth light source data) of the current round, the corresponding neuron mode is found through the reaction between the activated neuron and synapse and then combined with the adaptive optimization learning method. The above-described patterns correspond to labels in conventional machine learning/deep learning.
The learning methods in the current SNN algorithms are mainly classified into unsupervised learning, supervised learning, reinforcement learning and the like, but these methods are methods which need to initialize learning information (such as initial gradient, learning rate and the like). In the application of predicting the shape of the mask, the information carried by these neuron patterns is different each time with great probability, and if these learning methods are used, the generalization capability of the model may be insufficient, i.e. the model is not very general. Based on this, due to the sparsity of the pulse signal, the present embodiment adopts a learning method based on adaptive optimization to search for each neuron pattern in the time series. Different from the traditional machine learning/deep learning method which needs to establish a model by more data and needs to divide the data into a training set, a testing set and a verification set, the neuron mode is obtained in the embodiment, and the data does not need to be more data and does not need to be divided into the training set, the testing set and the verification set.
In some embodiments, the step of obtaining different neuron patterns in the pulse time series by using the adaptive optimization learning method further comprises: different neuron modes in a preliminary pulse time sequence are obtained through activated neuron and synaptic response learning by adopting a self-adaptive optimization learning method, wherein the neuron modes store combinations of different discrete signals (discretized parameters such as focal depth, wavelength and the like), and the weight and parameter values of the discrete signals among the different neuron modes are different; different neurons receiving data of the same neuron pattern form a common neuron pattern via synapses. Specifically, when the light source reaches the wafer position, an adaptive optimization learning method is adopted to search the pattern of each neuron in the time series (namely, neuron classification is carried out). The neurons can be in an activated state or a suppressed state, and the neurons in the activated state are classified, so that a certain neuron can be identified as belonging to a certain class of neuron patterns. Some preliminary patterns in the pulse time sequence are obtained in the spiking neural network, and then among synapses in each neuron, the same synapse in the pattern is triggered, i.e., different neurons receiving data of the same neuron pattern form a common neuron pattern through the synapses. The light source reaches the photoresist on the wafer to obtain the developing shape engraved on the wafer, and a whole set of light source data and parameters thereof are collected through the steps S1-S3.
In some embodiments, the step of obtaining different neuron patterns in the pulse time series by using the adaptive optimization learning method further comprises: searching each neuron mode in a time sequence by adopting an ant colony optimization algorithm; wherein, the neuron mode stores the combination of different discrete signals, and the weight and parameter value of the discrete signal are different between different neuron modes. Based on the novel impulse neural network, the corresponding output mode of the impulse neural network is predicted according to input data (or signals, frequency spectrograms and the like) in the current impulse neural network. Preferably, the embodiment adopts Ant Colony Optimization (ACO for short). The order in which the ant colony optimization algorithm is implemented is shown in fig. 3.
In connection with the foregoing embodiment, the step of using the ant colony optimization algorithm further includes: 1) Setting algorithm parameters of a current impulse neural network at a certain time, wherein the algorithm parameters comprise: the number of neurons, the number of synapses of the neurons, the time dimension of a certain moment, the distance between synapses, the amount of pheromones carried by a current neuron node and generated by the light source, the volatilization rate of the pheromones along with time and the probability that different pheromone concentration paths are selected in the follow-up process; 2) And substituting the algorithm parameters into an ant colony optimization algorithm, and gradually iterating until an optimal path is found, wherein the optimal path is a path for receiving information carried by the pulse signal.
For example, let a current spiking neural network at some time:
number of # neurons: n, each neuron corresponds to an agent;
number of neuronal synapses: m;
time dimension at time # time: t;
# inter-synaptic distance Dij;
amount of # pheromone: q (information such as the phase generated by the aforementioned light source, carried by the current neuron node);
time-dependent volatilization rate of pheromone: rho, rho is a constant less than 1;
probability of different pheromone concentration pathways being selected subsequently: p-c, concentration probability for short; a path with a large pheromone concentration (concentration) has a higher probability of being subsequently selected.
And substituting the listed parameters into an ant colony optimization algorithm, and gradually iterating until an optimal path is found. After the learning of the ant colony optimization algorithm, the pulse neural network learns some preliminary neuron patterns in the pulse time sequence; then, in each neuron, the same synapse pattern is triggered in the synapse.
The processing operation on data in the ant colony algorithm may specifically be: 1) Initializing relevant parameters, including: ant colony scale, pheromone factor, heuristic function factor, pheromone volatilization factor, pheromone constant, maximum iteration number and the like; and reading the data into a program, and preprocessing: such as converting coordinate information of neurons into a distance matrix between neurons. 2) Ants were randomly placed at different starting points and their next access neurons were calculated for each ant until all neurons were visited by any ant. 3) And calculating the path length of each ant, recording the optimal solution of the current iteration times, and updating the concentration of the pheromone on the path. 4) Judging whether the iteration times are reached, if not, returning to the step 2); if yes, the procedure is ended, and the best path is found.
And S4, receiving the test pulse signal from the light source for multiple times, and iteratively executing the steps S1 to S3 to obtain the neuron mode of the corresponding iteration rounds. In general, in order to complete the irradiation of the mask pattern onto the wafer, many light source irradiation, development, diffraction/interference/superposition of light, elimination of standing wave effect, and the like are required, which requires the light source to irradiate continuously for many times. Optionally, the wafer requires N lithographic shots, where N >0 and N ∈ positive integer. And the number of the iteration rounds is the number N of times of photoetching irradiation required by the wafer.
The neuron continuously receives the pulse signal from the light source, so that a whole set of data and parameters thereof are collected in each round of collection, and the obtained neuron mode of each iteration round is obtained.
Regarding step S5, a lithographic mask shape after the mask is irradiated by the target light source is predicted based on all the acquired neuron patterns. Specifically, the shape of the photoetching mask irradiated on the mask can be predicted by utilizing the mode obtained in the steps, so that the chip design graph can be preferentially and correspondingly adjusted in the layout design stage, and the initial irradiation light source can be adjusted and optimized, so that the light finally falling on the wafer meets the photoetching requirement.
In some embodiments, the step of predicting the shape of the lithography mask after the mask is illuminated by the target light source based on all the obtained neuron patterns further comprises: and based on all the obtained neuron modes, gradually carrying out backward extrapolation along each forward path to obtain a mask pattern and a chip design pattern, thereby predicting the shape of the mask irradiated by the light source. By predicting the shape of the light source after irradiating the mask, the chip design graph in the layout can be corrected in advance.
Based on the same inventive concept, the invention also provides a photoetching mask shape prediction device. The photoetching mask shape predicting device can predict the shape of the mask irradiated by the light source by adopting the photoetching mask shape predicting method shown in figure 1, so that the chip design graph in the layout can be corrected in advance, and the yield of products is improved.
Please refer to fig. 4, which is a block diagram illustrating a lithographic mask shape prediction apparatus according to an embodiment of the present invention. As shown in fig. 4, the lithography mask shape predicting apparatus includes: a data acquisition module 41, a data input module 42, a first mode acquisition module 43, a second mode acquisition module 44, and a prediction module 45.
Specifically, the data acquisition module 41 is configured to receive a test pulse signal from a light source and acquire test light source data of different stages, where the test light source data of different stages includes: first light source data when the light source passes through the convergent lens and does not reach the mask, second light source data when the light source passes through the convergent lens and does not reach the wafer, third light source data when the light source passes through the mask and does not reach the wafer, and fourth light source data when the light source passes through the mask and irradiates the wafer. The data input module is used for inputting the first light source data into the neurons of the pulse neural network. The first pattern obtaining module 43 is configured to obtain different neuron patterns in the pulse time sequence by using an adaptive optimization learning method based on the second light source data, the third light source data, and the fourth light source data. The second mode obtaining module 44 is configured to receive a test pulse signal from a light source for multiple times, and iteratively execute calling of the data acquiring module, the data input module, and the first mode obtaining module to obtain a neuron mode with a corresponding iteration number. The prediction module 45 is configured to predict a shape of the lithography mask after the mask is irradiated by the target light source based on all the obtained neuron patterns. The working modes of the modules can refer to the descriptions of the corresponding steps in the photolithographic mask shape prediction method shown in fig. 1, and are not described herein again.
In some embodiments, the first pattern obtaining module 43 is further configured to find each neuron pattern in the time series by using an ant colony optimization algorithm; the neuron mode stores different combinations of discrete signals, and the weights and parameter values of the discrete signals in different neuron modes are different. The implementation steps adopting the ant colony optimization algorithm comprise: 1) Setting algorithm parameters of a current impulse neural network at a certain time, wherein the algorithm parameters comprise: the number of neurons, the number of synapses of the neurons, the time dimension of a certain moment, the distance between synapses, the amount of pheromones carried by the current neuron node and generated by the light source, the volatilization rate of the pheromones along with time and the probability of different pheromone concentration paths being selected in the follow-up process; 2) And substituting the algorithm parameters into an ant colony optimization algorithm, and gradually iterating until an optimal path is found, wherein the optimal path is a path for receiving information carried by the pulse signal.
According to the photoetching mask shape prediction method and device provided by the embodiment of the invention, the photoetching mask shape prediction is carried out by combining self-adaptive optimization with a pulse neural network based on brain-like calculation; a light source generated by a photoetching system irradiates the convergent lens through projection type exposure, so that the convergent lens is exposed and developed on a mask and then is projected onto a wafer through the mask; by predicting the shape of the photoetching mask before the light beam reaches the wafer, the corresponding adjustment and optimization of the chip design graph and the adjustment and optimization of the initial irradiation light source can be preferentially carried out in the layout design stage, so that the light finally irradiated on the wafer meets the photoetching requirement. The embodiment of the invention omits the traditional degrees of OPC, SMO, SRAF and the like depending on the experience of related personnel or the precision/maintenance/advancement and the like of preparation equipment, and can play the role of artificial intelligence in the scene with smaller process. The invention solves the technical problem that the accuracy or speed of the obtained prediction model is deficient due to the lack of a large amount of data sets in the traditional photoetching mask shape prediction method based on machine learning or deep learning; meanwhile, the calculation overhead and the performance overhead are saved.
Based on the same inventive concept, the invention also provides an electronic device, which comprises a memory, a processor and a computer executable program, wherein the computer executable program is stored on the memory and can run on the processor; the processor, when executing the computer executable program, implements the steps of the lithography mask shape prediction method as shown in fig. 1.
It is within the scope of the inventive concept that embodiments may be described and illustrated in terms of modules that perform one or more of the described functions. These modules may be physically implemented by analog and/or digital circuits, such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hardwired circuits, etc., and may optionally be driven by firmware and/or software. The circuitry may be implemented in one or more semiconductor chips, for example. The circuitry making up the modules may be implemented by dedicated hardware, or by a processor (e.g., one or more programmed microprocessors and associated circuitry), or by a combination of dedicated hardware to perform some of the functions of the module and a processor to perform other functions of the module. Each module of an embodiment may be physically separated into two or more interactive and discrete modules without departing from the scope of the inventive concept. Likewise, the modules of the embodiments may be physically combined into more complex modules without departing from the scope of the inventive concept.
Generally, terms may be understood at least in part from their usage in context. For example, the term "one or more" as used herein may be used in a singular sense to describe a feature, structure, or characteristic, or may be used in a plural sense to describe a feature, structure, or combination of features, at least in part, depending on the context. Additionally, the term "based on" may be understood as not necessarily intended to convey an exclusive set of factors, but may instead allow for the presence of other factors not necessarily expressly described, again depending at least in part on the context.
It is noted that the terms "comprises" and "comprising," and variations thereof, as used herein, are intended to cover a non-exclusive inclusion. The terms "first," "second," and the like are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order, unless otherwise clearly indicated by the context, and it is to be understood that the data so used is interchangeable under appropriate circumstances. In addition, the embodiments and features of the embodiments in the present invention may be combined with each other without conflict. Moreover, in the foregoing description, descriptions of well-known components and techniques are omitted so as to not unnecessarily obscure the concepts of the present invention. In the above embodiments, each embodiment is described with emphasis on differences from other embodiments, and the same/similar parts among the embodiments may be referred to each other.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (10)
1. A method for predicting the shape of a lithography mask is characterized by comprising the following steps: 1) Receiving a test pulse signal from a light source and collecting test light source data of different stages, wherein the test light source data of different stages comprises: first light source data when the light source passes through the convergent lens and does not reach the mask, second light source data when the light source passes through the convergent lens and does not reach the wafer, third light source data when the light source passes through the mask and does not reach the wafer, and fourth light source data when the light source passes through the mask and irradiates the wafer; 2) Inputting the first light source data into neurons of a spiking neural network; 3) Acquiring different neuron modes in a pulse time sequence by adopting a self-adaptive optimization learning method based on the second light source data, the third light source data and the fourth light source data; 4) Receiving a test pulse signal from a light source for multiple times, and iteratively executing the step 1) to the step 3) to obtain a neuron mode of a corresponding iteration round number; and 5) predicting the shape of the photoetching mask after the target light source irradiates the mask based on all the acquired neuron modes.
2. The method of claim 1, wherein the light source is a DUV pulsed signal with a wavelength of 193nm generated by an excimer laser of a DUV lithography system; the number of the iteration rounds is the number N of times of photoetching irradiation needed by the wafer, N is greater than 0, and N belongs to a positive integer.
3. The method of claim 1, wherein the first light source data, the second light source data, the third light source data, and the fourth light source data each comprise: focal depth, resolution, detection control chip, alignment precision, numerical aperture of optical system, exposure wavelength, pulse signal, light source type, and light source irradiation mode; the second light source data and the third light source data further include: diffraction/interference deformation; the fourth light source data further includes: diffraction/interference deformation, the wavelength and focal length of two groups of projection lenses between the wafer and the mask, the distance between the two groups of projection lenses, the specification of the wafer, the size of the photoresist on the wafer, and the distance between the photoresist and the adjacent projection lenses.
4. The method of claim 1, wherein the step of obtaining different neuron patterns in the temporal sequence of impulses using an adaptive optimization learning method further comprises: searching each neuron mode in a time sequence by adopting an ant colony optimization algorithm; the neuron mode stores different combinations of discrete signals, and the weights and parameter values of the discrete signals in different neuron modes are different.
5. The method of claim 4, wherein the step of employing an ant colony optimization algorithm further comprises: setting algorithm parameters of a current impulse neural network at a certain time, wherein the algorithm parameters comprise: the number of neurons, the number of synapses of the neurons, the time dimension of a certain moment, the distance between synapses, the amount of pheromones carried by the current neuron node and generated by the light source, the volatilization rate of the pheromones along with time and the probability of different pheromone concentration paths being selected in the follow-up process; and substituting the algorithm parameters into an ant colony optimization algorithm, and gradually iterating until an optimal path is found, wherein the optimal path is a path for receiving information carried by the pulse signal.
6. The method of claim 1, wherein the step of obtaining different neuron patterns in the temporal sequence of impulses using an adaptive optimization learning method further comprises: different neuron modes in a preliminary pulse time sequence are obtained through activated neuron and synaptic response learning by adopting a self-adaptive optimization learning method, wherein combinations of different discrete signals are stored in the neuron modes, and the weights and parameter values of the discrete signals among the different neuron modes are different; different neurons receiving data of the same neuron pattern form a common neuron pattern via synapses.
7. The method of claim 1, wherein the step of predicting the shape of the lithographic mask after the mask is illuminated by the target light source based on all of the obtained neuron patterns further comprises: and based on all the obtained neuron modes, gradually carrying out backward extrapolation along each forward path to obtain a mask pattern and a chip design pattern, thereby predicting the shape of the mask irradiated by the light source.
8. A lithographic mask shape prediction apparatus, comprising: the data acquisition module is used for receiving test pulse signals from the light source and acquiring test light source data of different stages, wherein the test light source data of different stages comprise: first light source data when the light source passes through the convergent lens and does not reach the mask, second light source data when the light source passes through the convergent lens and does not reach the wafer, third light source data when the light source passes through the mask and does not reach the wafer, and fourth light source data when the light source passes through the mask and irradiates the wafer; the data input module is used for inputting the first light source data into neurons of a pulse neural network; a first pattern obtaining module, configured to obtain different neuron patterns in a pulse time sequence by using a self-adaptive optimization learning method based on the second light source data, the third light source data, and the fourth light source data; the second mode acquisition module is used for receiving a test pulse signal from a light source for multiple times, and iteratively executing the calling of the data acquisition module, the data input module and the first mode acquisition module to acquire a neuron mode of a corresponding iteration round number; and the prediction module is used for predicting the shape of the photoetching mask after the mask is irradiated by the target light source based on all the obtained neuron modes.
9. The apparatus of claim 8, wherein the first pattern obtaining module is further configured to search for each neuron pattern in a time series by using an ant colony optimization algorithm; the neuron mode stores different combinations of discrete signals, and the weights and parameter values of the discrete signals in different neuron modes are different.
10. An electronic device comprising a memory, a processor and a computer-executable program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for predicting the shape of a lithography mask according to any one of claims 1 to 7 when executing the computer-executable program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211592524.5A CN115598937B (en) | 2022-12-13 | 2022-12-13 | Photoetching mask shape prediction method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211592524.5A CN115598937B (en) | 2022-12-13 | 2022-12-13 | Photoetching mask shape prediction method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115598937A CN115598937A (en) | 2023-01-13 |
| CN115598937B true CN115598937B (en) | 2023-04-07 |
Family
ID=84852573
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211592524.5A Active CN115598937B (en) | 2022-12-13 | 2022-12-13 | Photoetching mask shape prediction method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115598937B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116306452B (en) * | 2023-05-17 | 2023-08-08 | 华芯程(杭州)科技有限公司 | Photoresist parameter acquisition method and device and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107797391A (en) * | 2017-11-03 | 2018-03-13 | 上海集成电路研发中心有限公司 | Optical adjacent correction method |
| WO2018125219A1 (en) * | 2016-12-30 | 2018-07-05 | Intel Corporation | Systems, methods, and apparatuses for implementing geometric kernel based machine learning for reducing opc model error |
| CN113892059A (en) * | 2019-03-25 | 2022-01-04 | Asml荷兰有限公司 | Method for determining patterns in a patterning process |
| CN114063383A (en) * | 2020-07-29 | 2022-02-18 | 三星电子株式会社 | Method for forming shape on mask based on deep learning, and mask manufacturing method using the method for forming shape on mask |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2571054B2 (en) * | 1987-04-28 | 1997-01-16 | キヤノン株式会社 | Exposure apparatus and element manufacturing method |
| JP2007273560A (en) * | 2006-03-30 | 2007-10-18 | Toshiba Corp | Light intensity distribution simulation method |
| JP2010044101A (en) * | 2008-08-08 | 2010-02-25 | Toshiba Corp | Pattern predicting method, program, and apparatus |
| JP2011138967A (en) * | 2009-12-28 | 2011-07-14 | Nikon Corp | Generating method, program, light source for exposure, and exposure apparatus |
| JP2011187597A (en) * | 2010-03-08 | 2011-09-22 | Nikon Corp | Illuminating light source evaluating method, illuminating light source setting method, exposure method and device manufacturing method, and program |
| CN102096336A (en) * | 2010-12-31 | 2011-06-15 | 清华大学 | Method for determining illumination intensity distribution of light source of photoetching process |
| CN102169295A (en) * | 2011-06-02 | 2011-08-31 | 清华大学 | Method for determining illumination intensity distribution of light source and mask pattern of photo-etching process |
| US20200380362A1 (en) * | 2018-02-23 | 2020-12-03 | Asml Netherlands B.V. | Methods for training machine learning model for computation lithography |
| KR20210010897A (en) * | 2018-06-15 | 2021-01-28 | 에이에스엠엘 네델란즈 비.브이. | Machine Learning Based Backlight Proximity Correction and Process Model Calibration |
-
2022
- 2022-12-13 CN CN202211592524.5A patent/CN115598937B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018125219A1 (en) * | 2016-12-30 | 2018-07-05 | Intel Corporation | Systems, methods, and apparatuses for implementing geometric kernel based machine learning for reducing opc model error |
| CN107797391A (en) * | 2017-11-03 | 2018-03-13 | 上海集成电路研发中心有限公司 | Optical adjacent correction method |
| CN113892059A (en) * | 2019-03-25 | 2022-01-04 | Asml荷兰有限公司 | Method for determining patterns in a patterning process |
| CN114063383A (en) * | 2020-07-29 | 2022-02-18 | 三星电子株式会社 | Method for forming shape on mask based on deep learning, and mask manufacturing method using the method for forming shape on mask |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115598937A (en) | 2023-01-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102499656B1 (en) | Deep Learning for Semantic Segmentation of Patterns | |
| KR102644214B1 (en) | Methods for training machine learning model for computation lothography | |
| KR102408319B1 (en) | Generate simulated output for your sample | |
| JP7064080B2 (en) | Photosensitized chemical amplification resist (PS-CAR) model calibration | |
| KR102471849B1 (en) | Photosensitized chemically amplified resist (PS-CAR) simulation | |
| TW202043909A (en) | Method and apparatus for layout pattern selection | |
| CN115598937B (en) | Photoetching mask shape prediction method and device and electronic equipment | |
| US20220335290A1 (en) | Method for increasing certainty in parameterized model predictions | |
| KR102589702B1 (en) | Method and apparatus for evaluating statistically distributed measured values in the inspection of elements of a photolithographic process | |
| WO2019219826A1 (en) | Method and apparatus for evaluating an unknown effect of defects of an element of a photolithography process | |
| US20220129775A1 (en) | Prediction and metrology of stochastic photoresist thickness defects | |
| CN112347838A (en) | Road map fusion | |
| TW202147035A (en) | Systems and methods for process metric aware process control | |
| CN117313640A (en) | Training method, device, equipment and storage medium for lithography mask generation model | |
| Wang et al. | A2-ILT: GPU accelerated ILT with spatial attention mechanism | |
| CN116342983B (en) | Method, electronic device and computer readable medium for generating and using graphic model | |
| Benaddy et al. | Recurrent neural network for software failure prediction | |
| Fahey et al. | SwiftSpike: An efficient software framework for the development of spiking neural networks | |
| Zach | Neural-network-based approach to resist modeling and OPC | |
| TWI814370B (en) | Causal convolution network for process control | |
| EP4075341A1 (en) | Modular autoencoder model for manufacturing process parameter estimation | |
| Younger et al. | A physical machine based on a super-turing compu-tational model | |
| EP4075340A1 (en) | Modular autoencoder model for manufacturing process parameter estimation | |
| EP4075339A1 (en) | Modular autoencoder model for manufacturing process parameter estimation | |
| Jiang et al. | Exploring Rule-Free Layout Decomposition via Deep Reinforcement Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |