A reconfigurable mask, reconfiguration method and reconfiguration system

Abstract

The application provides a reconfigurable mask and a reconfiguration method. The reconfigurable mask comprises: a mask array comprising at least one light modulation unit; and a driving module connected to each light modulation unit, the driving module being configured to provide a driving voltage for each light modulation unit. The light modulation unit comprises: a light modulation component comprising a phase modulation layer, the phase modulation layer being configured to change the phase of incident light based on the input driving voltage; and an output, the output being configured to output the incident light with the changed phase. Based on the reconfigurable mask, the target light field pattern can be iteratively learned and optimized based on the back propagation principle of the light diffraction neural network model, and the voltage parameters of the light diffraction neural network model can be optimized and configured through the electro-optic phase modulation technology to generate the driving voltage. In this way, through a modulable mask, the target light field pattern is taken as the iterative learning target, different wafer patterns are realized, the reconfigurable mask is multiplexed, and the manufacturing cost is saved.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a reconfigurable photomask, a reconfiguration method, and a reconfiguration system. Background Technology

[0002] A photomask is an image master used to transfer patterns in photolithography. It can be formed by an opaque light-blocking film on a transparent substrate to create a photomask pattern structure, and then the pattern information is transferred to the product substrate through an exposure process.

[0003] In related technologies, photomasks can be fabricated by electron beam direct writing of wafer layouts. However, the fabrication cost of photomasks is high, and each photomask can only correspond to one specific layout pattern, thus increasing the trial-and-error costs during the wafer fabrication process. Therefore, photomask fabrication is an indispensable step in the wafer fabrication process.

[0004] Furthermore, as the requirements for wafer processing precision continue to increase, the number of trial and error attempts for the mask can be increased by adjusting the mask fabrication parameters, thereby improving the pattern transfer effect. This, however, leads to an increase in mask fabrication costs. Summary of the Invention

[0005] This application provides a reconfigurable mask, a reconfiguration method, and a reconfiguration system, which can solve the problem of high fabrication cost of mask in the prior art.

[0006] In a first aspect, a reconfigurable photomask is provided, comprising: a photomask array, the photomask array including at least one light modulation unit arranged in an array, incident light passing through the photomask array to form a target light field pattern; a driving module, the driving module being connected to each of the light modulation units respectively, the driving module being used to provide a driving voltage to each of the light modulation units; wherein, the light modulation unit includes: a light modulation component, the light modulation component including a phase modulation layer, the phase modulation layer being used to change the phase of the incident light based on the input driving voltage; and an output unit, outputting the phase-changed incident light.

[0007] In some embodiments of this application, the reconfigurable mask further includes: a control module, which is connected to each of the light modulation units respectively, and is used to provide a selection control voltage to each of the light modulation units; the light modulation unit further includes: a voltage selection component, a first terminal of which is connected to the driving module, a second terminal of which is connected to the control module, and a third terminal of which is connected to the light modulation component, and the voltage selection component is used to control whether the driving voltage is input to the light modulation component according to the received selection control voltage.

[0008] In some embodiments of this application, the reconfigurable mask further includes a voltage generation module, which is connected to the driving module and the control module respectively. The voltage control unit is used to generate the driving voltage and the selection control voltage according to the wafer layout and the optical diffraction neural network model.

[0009] In some embodiments of this application, the voltage selection component includes a transistor, the source of which is connected to the driving module, and the gate of which is connected to the control module.

[0010] In some embodiments of this application, the optical modulation component includes a first substrate layer, a first conductive layer, a phase modulation layer, a second conductive layer, and a first protective layer stacked sequentially; the first substrate layer is made of a light-transmitting material; the first conductive layer is in contact with the drain of the transistor, and the second conductive layer is grounded; the phase modulation layer changes the phase of the incident light based on the output voltage of the drain of the transistor; the first protective layer is made of a light-transmitting material.

[0011] In some embodiments of this application, the first substrate layer includes a first region and a second region, the first region being used to form the voltage selection component and the second region being used to form the optical modulation component.

[0012] In some embodiments of this application, the material of the first conductive layer is indium tin oxide, and the material of the second conductive layer is indium tin oxide.

[0013] In some embodiments of this application, the wavelength of the incident light is from 10 nm to 400 nm.

[0014] In some embodiments of this application, the output element is a first metal layer including at least one through-hole for outputting the incident light with a changed phase.

[0015] Secondly, a reconstruction method for a reconstructable photomask is provided, applied to the aforementioned reconstructable photomask, comprising: receiving a wafer layout; if the wafer layout is in an unreconstructed state, generating a target light field pattern based on the wafer layout, and training a photodiffraction neural network model based on the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output, and obtaining a set of driving voltage parameters for the trained photodiffraction neural network model; the preset condition is that the electric field difference between the real-time light field pattern and the target light field pattern is less than a preset threshold; if the wafer layout is in a reconstructed state, directly obtaining the set of driving voltage parameters for the photodiffraction neural network model corresponding to the wafer layout; and generating a driving voltage and a control voltage based on the set of driving voltage parameters.

[0016] In some embodiments of this application, the step of generating a target light field pattern based on the wafer layout, and training a light diffraction neural network model based on the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output; obtaining the set of driving voltage parameters of the trained light diffraction neural network model includes: obtaining a preset voltage matrix that matches the mask array, wherein the number of elements in the preset voltage matrix is ​​positively correlated with the number of optical modulation components in the mask array; inputting the incident light phase, and using the preset initial conditions as model parameters, the light diffraction neural network model outputs a real-time light field pattern corresponding to the mask array. The preset initial conditions include the incident light phase and the preset voltage matrix; the electric field difference between the real-time optical diffraction field and the target diffraction field is estimated, the real-time optical diffraction field corresponds to the real-time light field pattern, and the target diffraction field corresponds to the target light field pattern; when the electric field difference value is greater than or equal to the preset threshold, the element values ​​in the preset voltage matrix are updated according to the electric field difference value; when the electric field difference value is less than the preset threshold, the training of the optical diffraction neural network model is terminated, the preset voltage matrix is ​​determined as the voltage parameter set of the trained optical diffraction neural network model, and the voltage parameter set is obtained.

[0017] In some embodiments of this application, the step of inputting the incident light phase and outputting the real-time light field pattern corresponding to the mask array using the optical diffraction neural network model with the preset voltage matrix as model parameters includes: calculating the real-time light field sub-pattern corresponding to each optical modulation component based on the photoelectric phase modulation model, the waveguide diffraction model, and the preset voltage matrix. The physical model corresponding to the optical modulation component includes the photoelectric phase modulation model and the waveguide diffraction model. The waveguide diffraction model is positively correlated with the incident field of the incident light, and the photoelectric phase modulation model is positively correlated with the liquid crystal thickness and refractive index in the optical modulation component. The real-time light field pattern corresponding to the mask array is obtained based on the optical diffraction neural network model and the real-time light field sub-pattern.

[0018] In some embodiments of this application, updating the element values ​​in the preset voltage matrix based on the electric field difference value includes: determining the error gradient value of the optical modulation component by using the chain rule and the electric field difference value, when the error estimation function is minimized; updating the element values ​​in the preset voltage matrix based on the error gradient value according to a preset update rule, wherein the preset update rule is: the updated voltage value is the sum of the product of the error gradient value and the preset learning rate and the voltage value.

[0019] In some embodiments of this application, generating a driving voltage and a control voltage based on the set of voltage parameters includes: determining the set of voltage parameters as the driving voltage of the optical modulation component, and determining the control voltage of the voltage selection component based on the control parameters of the voltage selection component.

[0020] Thirdly, a reconfigurable mask reconstruction system is provided, including a driving module configured to: receive a wafer layout; if the wafer layout is in an unreconstructed state, generate a target light field pattern, and train a light diffraction neural network model based on the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output, and obtain a set of driving voltage parameters for the trained light diffraction neural network model; the preset condition is that the electric field difference between the real-time light field pattern and the target light field pattern is less than a preset threshold; if the wafer layout is in a reconstructed state, directly obtain the set of driving voltage parameters for the light diffraction neural network model corresponding to the wafer layout; and generate a driving voltage and a control voltage based on the set of driving voltage parameters.

[0021] In some embodiments of this application, the step of generating a target light field pattern based on the wafer layout, and training a light diffraction neural network model based on the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output; obtaining the set of driving voltage parameters of the trained light diffraction neural network model includes: obtaining a preset voltage matrix that matches the mask array, wherein the number of elements in the preset voltage matrix is ​​positively correlated with the number of optical modulation components in the mask array; inputting the incident light phase, and using the preset initial conditions as model parameters, the light diffraction neural network model outputs a real-time light field pattern corresponding to the mask array. The preset initial conditions include the incident light phase and the preset voltage matrix; the electric field difference between the real-time optical diffraction field and the target diffraction field is estimated, the real-time optical diffraction field corresponds to the real-time light field pattern, and the target diffraction field corresponds to the target light field pattern; when the electric field difference value is greater than or equal to the preset threshold, the element values ​​in the preset voltage matrix are updated according to the electric field difference value; when the electric field difference value is less than the preset threshold, the training of the optical diffraction neural network model is terminated, the preset voltage matrix is ​​determined as the voltage parameter set of the trained optical diffraction neural network model, and the voltage parameter set is obtained.

[0022] In some embodiments of this application, the step of inputting the incident light phase and outputting the real-time light field pattern corresponding to the mask array using the optical diffraction neural network model with the preset voltage matrix as model parameters includes: calculating the real-time light field sub-pattern corresponding to each optical modulation component based on the photoelectric phase modulation model, the waveguide diffraction model, and the preset voltage matrix. The physical model corresponding to the optical modulation component includes the photoelectric phase modulation model and the waveguide diffraction model. The waveguide diffraction model is positively correlated with the incident field of the incident light, and the photoelectric phase modulation model is positively correlated with the liquid crystal thickness and refractive index in the optical modulation component. The real-time light field pattern corresponding to the mask array is obtained based on the optical diffraction neural network model and the real-time light field sub-pattern.

[0023] In some embodiments of this application, updating the element values ​​in the preset voltage matrix based on the electric field difference value includes: determining the error gradient value of the optical modulation component by using the chain rule and the electric field difference value, when the error estimation function is minimized; updating the element values ​​in the preset voltage matrix based on the error gradient value according to a preset update rule, wherein the preset update rule is: the updated voltage value is the sum of the product of the error gradient value and the preset learning rate and the voltage value.

[0024] In some embodiments of this application, generating a driving voltage and a control voltage based on the set of voltage parameters includes: determining the set of voltage parameters as the driving voltage of the optical modulation component, and determining the control voltage of the voltage selection component based on the control parameters of the voltage selection component.

[0025] The technical solution provided in this application, based on an embodiment of this application, provides a reconfigurable photomask. The reconfigurable photomask includes densely arranged and tunable light modulation units. These units transmit incident light downwards via waveguides and output light in a diffractive manner, forming a specific light field intensity pattern on the output layer (photoresist layer). Corresponding to the reconfigurable photomask, this application also provides a reconfiguration method and system for the reconfigurable photomask. Based on the backpropagation principle of a light diffraction neural network model, iterative learning and optimization of the target light field pattern are performed. The voltage parameters of the light diffraction neural network model are optimized using photoelectric phase modulation technology, so that the light field intensity pattern output by the output layer of the light diffraction neural network model is the target light field pattern. Thus, using a single tunable photomask, with the target light field pattern as the iterative learning target, different wafer layouts can be achieved, reusing the reconfigurable photomask and saving manufacturing costs. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the structure of a light modulation unit according to some embodiments of this application;

[0028] Figure 2 This is a schematic diagram of the structure of a reconfigurable mask according to some embodiments of this application;

[0029] Figure 3 This is a flowchart illustrating a reconfigurable mask reconstruction method according to some embodiments of this application;

[0030] Figure 4 This is a flowchart illustrating a method for obtaining a set of voltage parameters of a trained optical diffraction neural network model according to some embodiments of this application;

[0031] Figure 5 This is a schematic diagram of the phases of liquid crystal molecules according to some embodiments of this application;

[0032] Figure 6 These are schematic diagrams of the photolithography process according to some embodiments of this application;

[0033] Explanation of reference numerals in the attached diagram: 11-transistor, 111-source, 112-gate, 113-drain, 114-second region, 115-second protective layer, 116-second metal layer, 12-optical modulation component, 121-first region, 122-first conductive layer, 123-phase modulation layer, 124-second conductive layer, 125-first protective layer, 126-first metal layer, and 127-via. Detailed Implementation

[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0035] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0036] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, such that a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0037] In related technologies, photomasks can be fabricated by electron beam direct writing of wafer layouts. However, the fabrication cost of photomasks is high, and each photomask can only correspond to one specific layout pattern, thus increasing the trial-and-error costs during the wafer fabrication process. Therefore, photomask fabrication is an indispensable step in the wafer fabrication process.

[0038] Furthermore, as the requirements for wafer processing precision continue to increase, the number of trial and error attempts for the mask can be increased by adjusting the mask fabrication parameters, thereby improving the pattern transfer effect. This, however, leads to an increase in mask fabrication costs.

[0039] To solve the above technical problems, such as Figure 1 As shown, this application embodiment provides a reconfigurable mask, including:

[0040] A mask array, comprising at least one light modulation unit arranged in an array, wherein incident light passes through the mask array to form a target light field pattern;

[0041] A driving module is connected to each of the light modulation units respectively, and the driving module is used to provide a driving voltage to each of the light modulation units;

[0042] The light modulation unit includes:

[0043] An optical modulation component 12, the optical modulation component 12 including a phase modulation layer 123, the phase modulation layer 123 being used to change the phase of the incident light based on the input driving voltage;

[0044] Output device, outputs incident light with a changed phase.

[0045] In this embodiment, the array can be arranged in a matrix structure, a honeycomb structure, or a radial structure. In this embodiment, the array arrangement is not limited.

[0046] In this embodiment, the number of light modulation units in the mask array is adapted to the size and complexity of the wafer layout. If the light modulation units in the mask array are arranged in a matrix structure, the number of rows (m) and columns (n) in the matrix structure can be the same or different. For example, m=1, n=1; m=3, n=2; m=50, n=1; m=1000, n=1000, etc.

[0047] For example, the size of the light modulation unit is 2×2μm. 2 The mask array comprises 2000 rows and 2000 columns of light modulation units, and the size of the mask array is 4×4cm. 2 The size of the target light field pattern is 1×1cm 2 .

[0048] In this embodiment, each of the light modulation units can output corresponding outgoing light after being irradiated by incident light. When the outgoing light irradiates the wafer surface, it forms a light field pattern. Therefore, the mask array, including at least one of the light modulation units, can output a target light field pattern (formed by mutual interference of at least one light field image) corresponding to the wafer layout after being irradiated by incident light. After the target light field pattern is applied to the photoresist, the target light field pattern is fixed on the photoresist by exposure and development. Then, the wafer covered with photoresist is etched to transfer the target light field pattern onto the wafer.

[0049] In this embodiment, the driving voltage provided by the driving module acts directly on the phase modulation layer so that the incident light can change its phase after passing through the phase modulation layer.

[0050] In some embodiments of this application, the reconfigurable mask provided in this application further includes:

[0051] A control module is connected to each of the light modulation units respectively, and the control module is used to provide a selection control voltage for each of the light modulation units;

[0052] The light modulation unit further includes:

[0053] A voltage selection component is provided, wherein a first end of the voltage selection component is connected to the driving module, a second end of the voltage selection component is connected to the control module, and a third end of the voltage selection component is connected to the optical modulation component. The voltage selection component is used to control whether the driving voltage is input to the optical modulation component according to the received selection control voltage.

[0054] In this embodiment, the control module is connected to each of the optical modulation units, i.e., the control module is connected to each of the optical modulation components. The driving voltage provided by the driving module and the selection control voltage provided by the control module work together on the optical modulation component through a voltage selection component. This controls whether the driving voltage is input to the optical modulation component.

[0055] In this embodiment, the voltage selection component may include a transistor. Based on the physical structure of the transistor, such as... Figure 1 As shown, the source 111 and gate 112 of transistor 11 are connected to the driving module and the control module, respectively; the phase modulation layer 123 is connected to the drain 113 of transistor 11, and the optical modulation component 12 outputs outgoing light corresponding to the incident light based on the output voltage of the drain 113 of transistor 11. Hereafter, transistor 11 may refer to the voltage selection component.

[0056] In this embodiment, the control module may include m×n voltage output terminals, each of which can output a different voltage value. Based on the connection relationship between each light modulation unit and the voltage output terminals, a control voltage is provided to each light modulation unit.

[0057] In this embodiment of the application, the control module may further include a voltage output terminal. In this case, the control voltage of the gate 112 may be the same voltage value to control whether the source 111 and the drain 113 are turned on.

[0058] In some embodiments of this application, the reconfigurable mask further includes a voltage generation module, which is connected to the driving module and the control module respectively. The voltage control unit is used to generate the driving voltage and the selection control voltage according to the wafer layout and the optical diffraction neural network model.

[0059] In a first example, the voltage generation module can control multiple voltage output terminals to output the drive voltage and the selection control voltage with different voltage values ​​via a microcontroller or single-chip microcomputer.

[0060] In the second example, the voltage generation module can control the duty cycle of the pulse width modulation (PWM) signal through a microcontroller or single-chip microcomputer, and then convert the PWM signal into an analog voltage signal through a filter circuit, thereby realizing the driving voltage and the selection control voltage with different voltage values ​​output by multiple voltage output terminals.

[0061] In the third example, the voltage generation module can control the analog-to-digital converter (DAC) via a microcontroller or single-chip microcomputer to convert the digital signal into an analog voltage signal, thereby enabling multiple voltage output terminals to output the driving voltage and the selection control voltage with different voltage values.

[0062] In this embodiment, the light modulation unit is an element of the mask array, wherein the transistor 11 is used to control whether to transmit control voltage to the light modulation component 12. The transistor 11 can be a P-type transistor or an N-type transistor.

[0063] It is understandable that, such as Figure 2 As shown, in the mask array (comprising 20 light modulation units arranged in 4 rows and 5 columns), the gate 112 of all the light modulation units is connected to the voltage output terminal of the driving module via a scan line, and the source 111 of all the light modulation units in the mask array is connected to the voltage output terminal of the driving module via a data line. Thus, the driving module can determine which light modulation unit to provide the driving voltage to via the data line and the scan line.

[0064] In this embodiment, the transistor 11 can control the on / off state of the source 111 and the drain 113 using a selection control voltage received by the control module from the gate 112. When the source 111 and the drain 113 are on, the transistor 11 is used to transmit the driving voltage to the optical modulation component 12. When the source 111 and the drain 113 are not on, the transistor 11 is used not to transmit the driving voltage to the optical modulation component 12.

[0065] In some embodiments of this application, such as Figure 1 As shown, the optical modulation component 12 includes a first substrate layer, a first conductive layer 122, a phase modulation layer 123, a second conductive layer 124, and a first protective layer 125 stacked sequentially.

[0066] The material of the first substrate layer is a light-transmitting material;

[0067] The first conductive layer 122 is in contact with the drain 113 of the transistor 11, and the second conductive layer 124 is grounded;

[0068] The phase modulation layer 123 changes the phase of the incident light based on the output voltage of the drain 113 of the transistor 11;

[0069] The material of the first protective layer 125 is a light-transmitting material.

[0070] In some embodiments of this application, the output element is a first metal layer 126 including at least one through-hole 127 for outputting the incident light with a changed phase.

[0071] In this embodiment of the application, the material of the first substrate layer can be glass.

[0072] In this embodiment, when the source 111 and drain 113 are turned on, the transistor 11 transmits the driving voltage to the first conductive layer 122. As the magnitude of the driving voltage changes, the voltage between the first conductive layer 122 and the second conductive layer 124 also changes. The first conductive layer 122 and the second conductive layer 124 act as two electrode plates, and the electric field between them changes, causing a change in the light transmission direction in the phase modulation layer 123 located between the first conductive layer 122 and the second conductive layer 124. Ultimately, the phase modulation layer 123 can change the phase of the incident light in response to changes in the output voltage of the drain 113 of the transistor 11.

[0073] It is understood that, based on the output voltage of the drain 113 of the transistor 11, the voltage difference between the first conductive layer 122 and the second conductive layer 124 is controlled, thereby controlling the electric field environment of the phase modulation layer 123, so that the optical modulation component 12 can output outgoing light corresponding to the incident light.

[0074] In some embodiments of this application, the first conductive layer 122 is made of indium tin oxide (ITO), and the second conductive layer 124 is also made of ITO. ITO has high conductivity and high transmittance, allowing for the modulation of the electric field within the first conductive layer 122 and the second conductive layer 124 without affecting light transmission.

[0075] In this embodiment, the phase modulation layer 123 can be a liquid crystal layer. The liquid crystal layer, based on the properties of liquid crystal materials, controls the arrangement of liquid crystal molecules by voltage to change the light transmission state, thereby achieving image display. Liquid crystal is a composite organic material composed of long, rod-shaped liquid crystal molecules. In its natural state, the liquid crystal molecules in the phase modulation layer 123 are arranged in an orderly manner, with the long axes of the liquid crystal molecules approximately parallel and perpendicular to the upper and lower first conductive layers 122 and second conductive layers 124. After incident light passes through the first polarizer, its polarization direction is parallel to the long axis of the liquid crystal molecules, and the light cannot pass through the second polarizer. At this time, the incident light is filtered by the polarizer and no light is output.

[0076] When different driving voltages are applied, the liquid crystal molecules in the phase modulation layer 123 rotate under the action of the electric field, causing the liquid crystal molecules to rearrange perpendicular to the direction of the electric field, resulting in different phase orientations of the liquid crystal molecules. When light passes through the liquid crystal layer, phase delay occurs and the polarization state changes. After the incident light passes through the first polarizer, its polarization direction is not completely parallel to the long axis of the liquid crystal molecules. The light can be output through the transmission axis of the second polarizer. At this time, the incident light can be output after being filtered by the polarizer.

[0077] In this embodiment, the material of the first protective layer 125 may be silicon dioxide.

[0078] In this embodiment, the first metal layer 126 may include one or more vias 127. Each via 127 can output outgoing light corresponding to the incident light.

[0079] In some embodiments of this application, the aperture of the through-hole 127 is adapted to the wavelength of the incident light, and the length of the through-hole 127 is positively correlated with the wavelength of the incident light.

[0080] In some embodiments of this application, the incident light wavelength is from 10 nm to 400 nm. The incident light wavelength corresponds to the wavelength of the lithography machine's light source.

[0081] In this embodiment, after the driving module provides a driving voltage to each of the light modulation units and the control module provides a selection control voltage to each of the light modulation units, the voltage selection component, based on the driving voltage of the source 111 and the selection control voltage of the gate 112, if the first conductive layer 122 receives the driving voltage of the drain 113, then the phase modulation layer 123 changes the alignment direction of the liquid crystal molecules therein according to the driving voltage. At this time, the incident light first irradiates the first substrate layer, then penetrates the first substrate layer and the first conductive layer 122. Next, the incident light irradiates the liquid crystal molecules of the phase modulation layer 133 and changes the phase of the incident light. Then, the phase-changed incident light exits the phase modulation layer 133, penetrates the second conductive layer 124, irradiates the first metal protective layer 126, and finally, the phase-changed incident light is emitted from the via 127.

[0082] In some embodiments of this application, the material of the first metal layer 126 is one of the elements selected from gold, silver, copper, aluminum, chromium, nickel, or titanium. It is understood that the first metal layer 126 is opaque, and the outgoing light corresponding to the incident light with a changed phase is emitted from the through-hole 127.

[0083] In some embodiments of this application, such as Figure 1As shown, the first substrate layer includes a first region 121 and a second region 114, the first region 121 being used to form the transistor 11, and the second region 114 being used to form the optical modulation component 12.

[0084] In some embodiments of this application, in order to facilitate the fabrication of the light modulation unit, improve the lifespan of the light modulation unit and the mask array, and improve the stability of the light modulation unit, a second protective layer 115 can be provided on the basis of forming the transistor 11 in the first region. The second protective layer 115 is in direct contact with the other side of the transistor 11 and is connected to the first protective layer 125. The first thickness and the second thickness are the same. The first thickness is the sum of the thicknesses of the first conductive layer 122, the second conductive layer 124, the phase modulation layer 123, and the first protective layer 125, and the second thickness is the sum of the thicknesses of the transistor 11 and the second protective layer 115. A second metal layer 116 is connected to the first metal layer 126, and the second metal layer 116 has the same thickness as the first metal layer 126.

[0085] In this embodiment of the application, the material of the first substrate layer can be glass, and the first substrate layer can be formed in one piece.

[0086] In this embodiment, the material of the second protective layer 115 may be silicon dioxide.

[0087] In some embodiments of this application, the material of the first metal layer is one of the elements selected from gold, silver, copper, aluminum, chromium, nickel, or titanium.

[0088] In this embodiment of the application, the light modulation unit (including the transistor 11 and the light modulation component 12) can be formed through a fabrication process, and the mask array (including at least one of the light modulation units) can also be formed.

[0089] To address the high manufacturing cost of photomasks, in the embodiments of this application, such as... Figure 3 As shown, this application provides a reconstruction method for a reconfigurable photomask, applied to the aforementioned reconfigurable photomask (where the number of light modulation units is sufficient to form a target light field pattern), comprising:

[0090] Step S301: Receive wafer layout.

[0091] Step S302: If the wafer layout is in an unreconstructed state, a target light field pattern is generated according to the wafer layout, and the optical diffraction neural network model is trained according to the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output, and the set of driving voltage parameters of the trained optical diffraction neural network model is obtained.

[0092] The preset condition is that the difference in electric field between the real-time light field pattern and the target light field pattern is less than a preset threshold.

[0093] In this embodiment, the wafer layout can be a part of a layout file (DGS file). The target light field pattern corresponds to the exposure effect of the wafer layout. The target light field pattern can be identified using a two-dimensional light intensity matrix, where the matrix elements can be 0 or 1. Matrix element 1 (light intensity 1) is used to identify non-exposed areas, and matrix element 0 (light intensity 0) is used to identify exposed areas.

[0094] For example, the wafer layout size is 1.2μm × 0.9μm.

[0095] In this embodiment, different wafer layouts can be marked with identification identifiers, and the reconfiguration status of the wafer layouts can be distinguished by preset markers. Identification identifiers can be used for layout numbers and layout areas. Layout areas can be identified using a horizontal layout ruler and a vertical layout ruler.

[0096] In this embodiment of the application, after obtaining the voltage parameter set of the trained optical diffraction neural network model, the wafer layout can be marked as reconstructed by a preset marker, and the voltage parameter set corresponding to the wafer layout can be stored.

[0097] In the embodiments of this application, such as Figure 4 As shown, step S302 can be implemented by the following steps S401 to S405.

[0098] Step S401: Obtain a preset voltage matrix that matches the mask array.

[0099] The number of elements in the preset voltage matrix is ​​positively correlated with the number of optical modulation components in the mask array. The mask array is part of a reconfigurable mask. Depending on the control method of the optical modulation units, the number of elements in the preset voltage matrix can be twice the number of optical modulation units (capable of storing m×n×2 voltage values), or it can be one more than the number of optical modulation units (capable of storing m×n+1 voltage values).

[0100] In the embodiments of this application, the elements in the preset voltage matrix have the same initial voltage value, such as 0V, 1V, and 3V, etc.

[0101] Step S402: Input the incident light phase, and output the real-time light field pattern corresponding to the mask array using the optical diffraction neural network model with preset initial conditions as model parameters.

[0102] The preset initial conditions include the incident light phase and the preset voltage matrix.

[0103] In this embodiment, the optical diffraction neural network model, as an optical computing architecture, is used to complete various intelligent learning tasks. This includes highly dense control units that can passively manipulate light waves to propagate within the outgoing optical neural network, achieving computational and inference functions through light diffraction and interference.

[0104] It should be noted that the optical diffraction neural network model in this application embodiment is a physical model of light propagation obtained from a reconfigurable mask.

[0105] In this embodiment, the basic units of the optical diffraction neural network model include a photoelectric phase modulation model and a waveguide diffraction model. The physical structure corresponding to the photoelectric phase modulation model includes a first substrate layer, a first conductive layer 122, a phase modulation layer 123, a second conductive layer 124, and a first protective layer 125. The photoelectric phase modulation model can, based on the photoelectric effect, modulate the phase and refractive index of the liquid crystal molecules in the phase modulation layer 123 through voltage regulation, thereby controlling the photon phase of the incident light and achieving wavefront phase modulation before entering the first metal layer 126. The physical structure corresponding to the waveguide diffraction model includes a first metal layer 126 with vias. It is understood that the waveguide diffraction model varies with the number of vias included in the first metal layer 126.

[0106] In this embodiment of the application, the phase θ of the liquid crystal molecules in the phase modulation layer 123 LC (like Figure 5 As shown, the photoelectric phase modulation model can be expressed as: in, This is the voltage-phase modulation function. The phase change of light passing through the liquid crystal region is modulated by the voltage of the liquid crystal region, where V is the voltage between the first conductive layer 122 and the second conductive layer 124, λ is the wavelength of light in the liquid crystal region, and h LC n is the thickness of the liquid crystal phase modulation layer 123. LC is the refractive index of the liquid crystal in phase modulation layer 123.

[0107] In this embodiment of the application, the input to the waveguide diffraction model is the incident electric field E. in The output is an optical diffraction field E out The waveguide diffraction mode can be represented as Where ε is the model fitting coefficient, and i is the identifier of the imaginary part of the complex number. The simulation fitting coefficients (used to identify the propagation phase of photons and the phase difference under interface mismatch) are used, where r is the distance between the first metal layer 126 and the wafer.

[0108] In some embodiments of this application, the implementation of step S402 may include: calculating the real-time light field sub-pattern corresponding to each of the optical modulation components based on the photoelectric phase modulation model, the waveguide diffraction model, and the preset voltage matrix. The physical model corresponding to the optical modulation component includes the photoelectric phase modulation model and the waveguide diffraction model. The waveguide diffraction model is positively correlated with the incident field of the incident light, and the photoelectric phase modulation model is positively correlated with the liquid crystal thickness and refractive index in the optical modulation component. The real-time light field pattern is obtained based on the optical diffraction neural network model and the real-time light field sub-pattern.

[0109] In the embodiments of this application, such as Figure 6 As shown, the input layer of the optical diffraction neural network model takes incident light (coherent light of equal phase) as input. The diffraction layer in the optical diffraction neural network model is used to achieve mutual interference between the outgoing light output from the optical modulation component. The output layer in the optical diffraction neural network model is used to form the target optical field pattern at the intermediate layer position of the photoresist layer on the wafer. The optical diffraction neural network model utilizes a backpropagation mechanism to optimize the optical field distribution in the output layer (and photoresist layer).

[0110] In this embodiment, the forward propagation (i.e., optical diffraction propagation) of the optical diffraction neural network model: the light field intensity at the j-th optical modulation component in the output layer can be expressed as:

[0111]

[0112] Among them, O out,j For real-time optical diffraction field, i is the position of the optical modulation component, and E out,i→j (V i (j) represents the diffraction electric field of the i-th optical modulation component at output layer j. j is the location marker of the output layer (i.e., the wafer photoresist layer), which needs to cover the area of ​​the desired wafer layout; for example, j covers an area of ​​2×2 cm. 2 .

[0113] Step S403: Estimate the electric field difference between the real-time optical diffraction field and the target diffraction field.

[0114] Among them, the real-time optical diffraction field corresponds to the real-time light field pattern, and the target diffraction field corresponds to the target light field pattern.

[0115] In this embodiment of the application, the method for estimating the electric field difference value can be as follows:

[0116]

[0117] Where Error is the electric field difference value, O out,j For real-time optical diffraction field, T out,j The target diffraction field is defined by j, which is the location marker of the output layer (i.e., the wafer photoresist layer). This j must cover the area of ​​the desired wafer layout; for example, j covers a range of 2 × 2 μm. 2 The upper limit k of the wafer position identifier j depends on the size of the optical neural network, i.e., the size of the output layer, and can be 10, 100, 1000, etc. A larger value for k results in lower output layer precision, but also a larger optical neural network and a greater computational load. Generally, the value of k is approximately... This ensures that the distance between adjacent positions in the output layer is equal to the wavelength of light. Based on the above estimation method, the difference in electric field between the real-time optical diffraction field and the target diffraction field can be estimated.

[0118] In summary, the optical diffraction neural network model optimizes and updates the preset voltage matrix (the voltage configuration parameter of the optical network in the optical diffraction neural network model) through forward propagation, error estimation, and back propagation. After continuous iteration and optimization, its error value will approach a small stable value, so that the degree pattern output by the output layer of the optical diffraction neural network model can be the target light field pattern according to the updated preset voltage matrix. That is, a feedback is realized through the operation mechanism of the optical diffraction neural network model.

[0119] Step S404: When the electric field difference value is greater than or equal to a preset threshold, update the element values ​​in the preset voltage matrix according to the electric field difference value.

[0120] In some embodiments of this application, the implementation of step S404 above may include using the backpropagation mechanism of the optical diffraction neural network model: determining the error gradient value of the optical modulation component by using the chain rule and the electric field difference value when the error estimation function is minimized; updating the element values ​​in the preset voltage matrix based on the error gradient value according to the preset update rule, wherein the preset update rule is: the updated voltage value is the sum of the product of the error gradient value and the preset learning rate and the voltage value.

[0121] In this embodiment of the application, after updating the element values ​​in the preset voltage matrix, the above steps S402 and S403 are repeated to re-estimate the electric field difference value. In this way, the electric field difference value is gradually reduced through iteration until the electric field difference value is less than the preset threshold (the real-time light field pattern output by the output layer of the optical diffraction neural network model is the target light field pattern).

[0122] Step S405: When the electric field difference value is less than a preset threshold, the training of the optical diffraction neural network model is terminated, the preset voltage matrix is ​​determined as the voltage parameter set of the trained optical diffraction neural network model, and the voltage parameter set is obtained.

[0123] In this embodiment of the application, when the electric field difference value is less than a preset threshold, the optimization of the model parameters in the optical diffraction neural network model is completed. At this time, the real-time light field pattern output by the output layer of the optical diffraction neural network model is the target light field pattern.

[0124] It is understandable that when at least one optical modulation component in the mask matrix receives the driving voltage corresponding to the voltage parameter set, the reconfigurable mask can form the target light field pattern under the illumination of the incident light.

[0125] In this embodiment of the application, according to the model parameter storage rules of the optical diffraction neural network model, the preset voltage matrix is ​​obtained and the preset voltage matrix is ​​determined as a voltage parameter set (that is, each element value in the preset voltage matrix is ​​determined as a voltage parameter, and all element values ​​in the preset voltage matrix are determined as a voltage parameter set).

[0126] Thus, by training the optical diffraction neural network model, a set of voltage parameters is obtained, which facilitates the acquisition of voltage parameter sets corresponding to different wafer layouts.

[0127] Step S303: If the wafer layout is in a reconstructed state, directly obtain the set of driving voltage parameters of the optical diffraction neural network model corresponding to the wafer layout.

[0128] In this embodiment, the wafer layout is marked as reconstructed by a preset marker, and the set of driving voltage parameters corresponding to the wafer layout can be directly searched and obtained.

[0129] It is understandable that if the wafer layout is in a reconstructed state, then the optical diffraction neural network model has been trained based on the wafer layout, and the set of voltage parameters corresponding to the wafer layout has been stored.

[0130] Step S304: Generate driving voltage and control voltage based on the set of driving voltage parameters.

[0131] In some embodiments of this application, step S304 may include: determining the voltage parameter set as the driving voltage of the optical modulation component, and determining the control voltage of the voltage selection component based on the control parameters of the voltage selection component.

[0132] Thus, the light modulation unit based on the reconfigurable mask with the same physical structure provides different driving voltages and control voltages. After being illuminated by incident light, the reconfigurable mask can output different target light field patterns, thereby realizing the reconstruction of the reconfigurable mask.

[0133] Based on the reconfigurable photomask provided in this application embodiment, the reconfigurable photomask includes densely arranged and tunable light modulation units. These light modulation units transmit incident light downwards via waveguides and output light in the form of light diffraction, forming a specific light field intensity pattern on the output layer (photoresist layer). Corresponding to the reconfigurable photomask, this application embodiment also provides a reconfiguration method for the reconfigurable photomask. Based on the backpropagation principle of a light diffraction neural network model, the method iteratively learns and optimizes the target light field pattern. It optimizes the voltage parameters of the light diffraction neural network model using photoelectric phase modulation technology, so that the light field intensity pattern output by the output layer of the light diffraction neural network model is the target light field pattern. Thus, using a single tunable photomask, with the target light field pattern as the iterative learning target, different wafer layouts can be achieved, reusing the reconfigurable photomask and saving manufacturing costs.

[0134] In this embodiment of the application, a reconfigurable mask reconstruction system is also provided, including a voltage generation module. The voltage generation module is configured to: receive a wafer layout; if the wafer layout is in an unreconstructed state, generate a target light field pattern, and train a light diffraction neural network model based on the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output, and obtain a set of driving voltage parameters for the trained light diffraction neural network model; the preset condition is that the electric field difference between the real-time light field pattern and the target light field pattern is less than a preset threshold; if the wafer layout is in a reconstructed state, directly obtain the set of driving voltage parameters for the light diffraction neural network model corresponding to the wafer layout; and generate a driving voltage and a control voltage based on the set of driving voltage parameters.

[0135] In some embodiments of this application, the voltage generation module is configured to: generate a target light field pattern based on the wafer layout, and train a light diffraction neural network model according to the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output; obtain a set of driving voltage parameters for the trained light diffraction neural network model, including: obtaining a preset voltage matrix that matches the mask array, wherein the number of elements in the preset voltage matrix is ​​positively correlated with the number of light modulation components in the mask array; input the incident light phase, and the light diffraction neural network model with the preset initial conditions as model parameters outputs the real-time light field pattern corresponding to the mask array. The light field pattern, wherein the preset initial conditions include the incident light phase and the preset voltage matrix; the electric field difference between the real-time optical diffraction field and the target diffraction field is estimated, wherein the real-time optical diffraction field corresponds to the real-time light field pattern and the target diffraction field corresponds to the target light field pattern; when the electric field difference value is greater than or equal to the preset threshold, the element values ​​in the preset voltage matrix are updated according to the electric field difference value; when the electric field difference value is less than the preset threshold, the training of the optical diffraction neural network model is terminated, and the preset voltage matrix is ​​determined as the voltage parameter set of the trained optical diffraction neural network model, and the voltage parameter set is obtained.

[0136] In some embodiments of this application, the voltage generation module is configured to: input the incident light phase, and output the real-time light field pattern corresponding to the mask array using the optical diffraction neural network model with the preset voltage matrix as model parameters, including: calculating the real-time light field sub-pattern corresponding to each optical modulation component based on the photoelectric phase modulation model, the waveguide diffraction model, and the preset voltage matrix, wherein the physical model corresponding to the optical modulation component includes the photoelectric phase modulation model and the waveguide diffraction model, the waveguide diffraction model is positively correlated with the incident field of the incident light, and the photoelectric phase modulation model is positively correlated with the liquid crystal thickness and refractive index in the optical modulation component; and obtaining the real-time light field pattern corresponding to the mask array based on the optical diffraction neural network model and the real-time light field sub-pattern.

[0137] In some embodiments of this application, the voltage generation module is configured to: update the element values ​​in the preset voltage matrix according to the electric field difference value, including: determining the error gradient value of the optical modulation component by using the chain rule and the electric field difference value, when the error estimation function is minimized; updating the element values ​​in the preset voltage matrix based on the error gradient value according to a preset update rule, wherein the preset update rule is: the updated voltage value is the sum of the product of the error gradient value and the preset learning rate and the voltage value.

[0138] In some embodiments of this application, the voltage generation module is configured to: generate a driving voltage and a control voltage according to the voltage parameter set, including: determining the voltage parameter set as the driving voltage of the optical modulation component, and determining the control voltage of the voltage selection component based on the control parameters of the voltage selection component.

[0139] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0140] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A reconfigurable photomask, characterized in that, include: A mask array, comprising at least one light modulation unit arranged in an array, wherein incident light passes through the mask array to form a target light field pattern; A driving module is connected to each of the light modulation units respectively, and the driving module is used to provide a driving voltage to each of the light modulation units; The light modulation unit includes: An optical modulation assembly, the optical modulation assembly including a phase modulation layer, the phase modulation layer being used to change the phase of the incident light based on the input driving voltage; Output device, outputs incident light with a changed phase.

2. The reconfigurable mask according to claim 1, characterized in that, The reconfigurable mask also includes: A control module is connected to each of the light modulation units respectively, and the control module is used to provide a selection control voltage for each of the light modulation units; The light modulation unit further includes: A voltage selection component is provided, wherein a first end of the voltage selection component is connected to the driving module, a second end of the voltage selection component is connected to the control module, and a third end of the voltage selection component is connected to the optical modulation component. The voltage selection component is used to control whether the driving voltage is input to the optical modulation component according to the received selection control voltage.

3. The reconfigurable mask according to claim 2, characterized in that, The reconfigurable mask also includes: A voltage generation module is connected to both the driving module and the control module. The voltage control unit is used to generate the driving voltage and the selection control voltage based on the wafer layout and the optical diffraction neural network model.

4. The reconfigurable mask according to claim 2, characterized in that, The voltage selection component includes a transistor, the source of which is connected to the driving module, and the gate of which is connected to the control module.

5. The reconfigurable mask according to claim 1, characterized in that, The optical modulation component comprises a first substrate layer, a first conductive layer, a phase modulation layer, a second conductive layer, and a first protective layer stacked sequentially. The material of the first substrate layer is a light-transmitting material; The first conductive layer is connected to the drain of the transistor, and the second conductive layer is grounded; The phase modulation layer changes the phase of the incident light based on the output voltage of the transistor drain. The material of the first protective layer is a light-transmitting material.

6. The reconfigurable mask according to claim 5, characterized in that, The first substrate layer includes a first region and a second region, the first region being used to form the voltage selection component and the second region being used to form the optical modulation component.

7. The reconfigurable mask according to claim 5, characterized in that, The first conductive layer is made of indium tin oxide, and the second conductive layer is made of indium tin oxide.

8. The reconfigurable mask according to claim 5, characterized in that, The incident light wavelength is from 10 nm to 400 nm.

9. The reconfigurable mask according to claim 5, characterized in that, The output element is a first metal layer including at least one through hole for outputting the incident light with a changed phase.

10. A method for reconstructing a reconfigurable mask, characterized in that, Applied to the reconfigurable mask according to any one of claims 1 to 9, comprising: Receive wafer layout; If the wafer layout is in an unreconstructed state, a target light field pattern is generated based on the wafer layout, and the optical diffraction neural network model is trained according to the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output, and the set of driving voltage parameters of the trained optical diffraction neural network model is obtained; the preset condition is that the electric field difference between the real-time light field pattern and the target light field pattern is less than a preset threshold. If the wafer layout is in a reconstructed state, then the set of driving voltage parameters of the optical diffraction neural network model corresponding to the wafer layout is directly obtained; Based on the set of driving voltage parameters, the driving voltage and control voltage are generated.

11. The method for reconstructing a reconfigurable mask according to claim 10, characterized in that, The target light field pattern is generated according to the wafer layout, and the optical diffraction neural network model is trained according to the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output. Obtain the set of driving voltage parameters for the trained optical diffraction neural network model, including: Obtain a preset voltage matrix that matches the mask array, wherein the number of elements in the preset voltage matrix is ​​positively correlated with the number of optical modulation components in the mask array; The incident light phase is input, and the optical diffraction neural network model with the preset initial conditions as model parameters outputs the real-time light field pattern corresponding to the mask array. The preset initial conditions include the incident light phase and the preset voltage matrix. Estimate the electric field difference between the real-time optical diffraction field and the target diffraction field, wherein the real-time optical diffraction field corresponds to the real-time light field pattern, and the target diffraction field corresponds to the target light field pattern; When the electric field difference value is greater than or equal to the preset threshold, the element values ​​in the preset voltage matrix are updated according to the electric field difference value. When the electric field difference value is less than the preset threshold, the training of the optical diffraction neural network model is terminated, the preset voltage matrix is ​​determined as the voltage parameter set of the trained optical diffraction neural network model, and the voltage parameter set is obtained.

12. The method for reconstructing a reconfigurable mask according to claim 11, characterized in that, The step of inputting the incident light phase and using the preset voltage matrix as model parameters, the optical diffraction neural network model outputs the real-time light field pattern corresponding to the mask array, including: Based on the photoelectric phase modulation model, the waveguide diffraction model, and the preset voltage matrix, the real-time light field sub-pattern corresponding to each of the optical modulation components is calculated. The physical model corresponding to the optical modulation component includes the photoelectric phase modulation model and the waveguide diffraction model. The waveguide diffraction model is positively correlated with the incident field of the incident light, and the photoelectric phase modulation model is positively correlated with the liquid crystal thickness and the liquid crystal refractive index in the optical modulation component. Based on the optical diffraction neural network model and the real-time optical field sub-pattern, the real-time optical field pattern corresponding to the mask array is obtained.

13. The method for reconstructing a reconfigurable mask according to claim 11, characterized in that, The step of updating the element values ​​in the preset voltage matrix based on the electric field difference value includes: By using the chain rule and the electric field difference value, the error gradient value of the optical modulation component is determined when the error estimation function is minimized. According to the preset update rule, the element values ​​in the preset voltage matrix are updated based on the error gradient value. The preset update rule is: the updated voltage value is the sum of the product of the error gradient value and the preset learning rate and the voltage value.

14. The method for reconstructing a reconfigurable mask according to claim 10, characterized in that, The step of generating the driving voltage and control voltage based on the voltage parameter set includes: The set of voltage parameters is determined as the driving voltage of the optical modulation component, and the control voltage of the voltage selection component is determined based on the control parameters of the voltage selection component.

15. A reconfigurable mask reconstruction system, characterized in that, Includes a voltage control unit, which is configured to: Receive wafer layout; If the wafer layout is in an unreconstructed state, a target light field pattern is generated, and the optical diffraction neural network model is trained based on the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output, and the set of driving voltage parameters of the trained optical diffraction neural network model is obtained; the preset condition is that the electric field difference between the real-time light field pattern and the target light field pattern is less than a preset threshold. If the wafer layout is in a reconstructed state, then the set of driving voltage parameters of the optical diffraction neural network model corresponding to the wafer layout is directly obtained; Based on the set of driving voltage parameters, the driving voltage and control voltage are generated.

16. The reconfigurable mask reconfiguration system according to claim 15, characterized in that, The target light field pattern is generated according to the wafer layout, and the optical diffraction neural network model is trained according to the target light field pattern under preset initial conditions until a real-time light field pattern that meets the preset conditions is output. Obtain the set of driving voltage parameters for the trained optical diffraction neural network model, including: Obtain a preset voltage matrix that matches the mask array, wherein the number of elements in the preset voltage matrix is ​​positively correlated with the number of optical modulation components in the mask array; The incident light phase is input, and the optical diffraction neural network model with the preset initial conditions as model parameters outputs the real-time light field pattern corresponding to the mask array. The preset initial conditions include the incident light phase and the preset voltage matrix. Estimate the electric field difference between the real-time optical diffraction field and the target diffraction field, wherein the real-time optical diffraction field corresponds to the real-time light field pattern, and the target diffraction field corresponds to the target light field pattern; When the electric field difference value is greater than or equal to the preset threshold, the element values ​​in the preset voltage matrix are updated according to the electric field difference value. When the electric field difference value is less than the preset threshold, the training of the optical diffraction neural network model is terminated, the preset voltage matrix is ​​determined as the voltage parameter set of the trained optical diffraction neural network model, and the voltage parameter set is obtained.

17. The reconfigurable mask reconfiguration system according to claim 16, characterized in that, The step of inputting the incident light phase and using the preset voltage matrix as model parameters, the optical diffraction neural network model outputs the real-time light field pattern corresponding to the mask array, including: Based on the photoelectric phase modulation model, the waveguide diffraction model, and the preset voltage matrix, the real-time light field sub-pattern corresponding to each of the optical modulation components is calculated. The physical model corresponding to the optical modulation component includes the photoelectric phase modulation model and the waveguide diffraction model. The waveguide diffraction model is positively correlated with the incident field of the incident light, and the photoelectric phase modulation model is positively correlated with the liquid crystal thickness and the liquid crystal refractive index in the optical modulation component. Based on the optical diffraction neural network model and the real-time optical field sub-pattern, the real-time optical field pattern corresponding to the mask array is obtained.

18. The reconfigurable mask reconfiguration system according to claim 16, characterized in that, The step of updating the element values ​​in the preset voltage matrix based on the electric field difference value includes: By using the chain rule and the electric field difference value, the error gradient value of the optical modulation component is determined when the error estimation function is minimized. According to the preset update rule, the element values ​​in the preset voltage matrix are updated based on the error gradient value. The preset update rule is: the updated voltage value is the sum of the product of the error gradient value and the preset learning rate and the voltage value.

19. The reconfigurable mask reconfiguration system according to claim 15, characterized in that, The step of generating the driving voltage and control voltage based on the voltage parameter set includes: The set of voltage parameters is determined as the driving voltage of the optical modulation component, and the control voltage of the voltage selection component is determined based on the control parameters of the voltage selection component.

Images