Adjustable mode converter based on liquid crystal and its optimization method

Abstract

The present invention discloses an optimization method for an adjustable mode converter based on liquid crystal, which belongs to the field of integrated optics and optical communications. The optimization method of the present invention achieves a faster convergence speed by constructing an optimization model of an adjustable mode converter based on liquid crystal, utilizing an adjoint algorithm, and mainly updating the gradient directional structure of the parameter space based on the calculated objective function. Logical judgment is used to minimize the consumption of computing resources to improve the efficiency of generating the device while ensuring that the target function of the device meets the preset performance requirements. The present invention optimizes an adjustable mode converter based on liquid crystal, which converts the fundamental mode light introduced from the input waveguide into multiple high-order mode lights through the adjustable mode converter. It is mainly used for the exchange of data of different channels in optical communication links, and can also be used for on-chip multi-channel sensor chips to cooperate with intelligent algorithms for flexible data processing, and is combined with a mode demultiplexer to realize the upload and download of multi-channel signals.

Description

Technical Field

[0001] The present invention relates to an on-chip multifunctional optical device optimization method, in particular to a liquid crystal-based adjustable mode converter, and belongs to the fields of integrated optics and optical communications. Background Art

[0002] To meet the ever-increasing demand for data bandwidth, multiple multiplexing technologies are needed to expand bandwidth, such as time division multiplexing, wavelength division multiplexing, and polarization multiplexing. However, current multiplexing technologies are still unable to meet the growing bandwidth requirements. Mode multiplexing (MDM) can expand multiple channels using a single wavelength laser, which helps reduce the cost and size of optical communication systems. Tunable mode converters (TMCs) enable flexible signal switching between multiple channels, fully utilizing optical bandwidth, and combined with mode demultiplexers (DMs) can achieve optical routing.

[0003] Mode converters can convert the fundamental transverse mode into any higher-order mode. Currently, there are tunable mode converters based on a variety of principles, such as subwavelength plasmonic waveguides, slit photonic crystal waveguides, tilted Bragg gratings and thermo-optic effects, lithium niobate crystals, and phase-change materials (GST).

[0004] Switching speed is crucial for tunable mode converters. The previously mentioned tunable platforms all have their drawbacks, such as slow modulation rates due to the thermo-optical effect, large lithium niobate crystals, and high losses in phase-change materials. Therefore, liquid crystals are used as tunable platforms, enabling tunability by controlling the angle of the liquid crystal director. Appropriate optimization algorithms for multifunctional devices can greatly improve design efficiency. Heuristic algorithms, such as particle swarm optimization and genetic algorithms, can search most parameter spaces. However, as the design space expands, their computational complexity increases significantly, consuming significant computational resources. In recent years, neural networks have been introduced into optical device design, resulting in numerous high-performance devices. However, neural network algorithms require full-wave simulation software for database preparation. Furthermore, due to the nature of optical devices, multiple device structures may correspond to a single optical response. This multi-solution nature makes neural network convergence difficult. Summary of the Invention

[0005] One of the main purposes of the present invention is to provide an optimization method for an adjustable mode converter based on liquid crystal, which is used to optimize the adjustable mode converter based on liquid crystal. By constructing an optimization model of the adjustable mode converter based on liquid crystal, using an adjoint algorithm, the gradient directional update structure of the structural parameters is mainly based on the calculated objective function to achieve a faster convergence speed, and using logical judgment to minimize computing resource consumption to improve the efficiency of generating devices while ensuring that the target function of the device meets the preset performance requirements.

[0006] The second main purpose of the present invention is to optimize the liquid crystal-based adjustable mode converter according to the optimization method of the liquid crystal-based adjustable mode converter. The liquid crystal-based adjustable mode converter converts the fundamental mode light introduced from the input waveguide into multiple high-order mode lights through the adjustable mode converter. It is mainly used for exchanging data of different channels in optical communication links. It can also be used for on-chip multi-channel sensor chips to cooperate with intelligent algorithms for flexible data processing. It is combined with a mode demultiplexer to realize the uploading and downloading of multiple signals. It has the advantages of miniaturization, high switching speed, multi-function and high robustness.

[0007] In order to achieve the above objectives, the following technical solutions are implemented:

[0008] The present invention discloses a method for optimizing a liquid crystal-based adjustable mode converter, comprising the following steps:

[0009] Step 1: Construct a liquid crystal-based tunable mode converter model. The liquid crystal-based tunable mode converter is formed by wafer etching. The etched wafer includes a waveguide and a mode conversion region. The mode conversion region is divided on the wafer, with a length of L0 and a width of W0. Hyperparameters are pre-set based on the wafer etching process and computing resource constraints. These hyperparameters include the length of L0, the width of W0, the height of H0, and the minimum feature size of the mode conversion region. The height of the tunable mode converter is the height of the wafer device layer. The mode conversion region is evenly divided into n*m unit cuboids in three dimensions. The material state of each unit cuboid is either tunable material or wafer device layer. The number of unit cuboids, n*m, is determined by setting the length and width of the mode conversion region and the size of the unit cuboid. The height of the unit cuboid is equal to the height of the tunable mode converter. The unit cuboid structural parameters are represented by density ρ, which ranges from 0 to 1, where 0 represents the unit cuboid as tunable material, 1 represents the device layer material, and intermediate values ​​are invalid. The unit cuboid represented by 0 is a unit cubic space etched out of the mode conversion area on the wafer and filled with adjustable material. The mode conversion area array formed by the device layer unit cuboid and the adjustable material unit cuboid serves as an adjustable mode converter.

[0010] Step 2: Constrain and process the unit cube structure parameters. Use the Gaussian blur algorithm to filter the high-frequency information in the density distribution ρ, that is, small cuboids or holes, sharp corners. Adding Gaussian blur processing reduces the difficulty of device processing and increases the robustness of the device; use binary mapping to judge the parameters in the middle value as adjustable materials or device layers, and eliminate the invalid parameters that cannot be produced during the iteration of the unit cube structure parameters; Perform opening operation to directly erase the tiny cuboid and adjust the density Perform closing operation to fill the tiny holes. Gaussian blur processing, opening and closing operations are performed to improve the device quality factor and robustness, reduce the difficulty of device processing, and reduce the strong anchoring effect of the tiny structure on the liquid crystal molecules, which will affect the adjustable characteristics of the liquid crystal.

[0011] According to formula (1), the Gaussian blur algorithm is used to eliminate the elongated cuboids or tiny holes in the structure; Parameters representing the density distribution after Gaussian operation:

[0012]

[0013]

[0014] Where D is the set of unit cuboids with Gaussian blur radius R around the i-th unit cuboid, r i and r j is the position of the i-th and j-th unit cubes, h ij Represents weight.

[0015] According to formula (3), the binary mapping is used to judge the parameters in the middle value as adjustable materials or device layers, eliminating the invalid parameters that cannot be produced during the iteration of the unit cube structure parameters; the density after binarization is used express:

[0016]

[0017] Here, θ controls the binarization strength, and η is the center of the curve. When the binarization strength is high, density less than η is mapped to 0, while density greater than η is mapped to 1. As the number of iterations increases, the binarization strength is gradually increased to avoid a sudden decrease in convergence speed due to a sudden increase in the binarization strength.

[0018] In order to avoid affecting the convergence speed, preferably, the center η of the curve is selected as 0.5.

[0019] In order to reduce the limitation of Gaussian blur on the degree of freedom of the device, a smaller Gaussian blur radius is used and the density Perform opening operation to directly erase the tiny cuboid and adjust the density Perform closing operation to fill the tiny holes. Gaussian blur processing, opening operation, and closing operation are performed to improve the quality factor and robustness of the device, reduce the difficulty of device processing, and reduce the strong anchoring effect of the tiny structure on the liquid crystal molecules.

[0020] Step 3: Liquid crystals offer the advantages of low optical loss and fast switching speeds, so they are used as the tunable material. Tunable devices control the relative dielectric constant by controlling the angle of the liquid crystal director. A higher birefringence index results in a higher average device figure of merit (FOM).

[0021] Step 4: Taking the maximum average quality factor FOM of the device as the optimization goal and the Helmholtz equation as the constraint condition, construct a model optimization problem of the adjustable mode converter based on liquid crystal; use the adjoint algorithm, mainly based on the electric field distribution data obtained by forward simulation and backward simulation to obtain the gradient of the objective function with respect to the structural parameters, use the gradient to update the structural parameters in a directed manner to achieve faster convergence speed, add logical judgment to ensure that the target function of the device meets the preset performance requirements, and minimize the consumption of computing resources to improve the efficiency of device generation.

[0022] The electric field distribution in the mode conversion region obtained by forward and backward simulations is used to calculate the gradient of the objective function relative to the structural parameters. Taking the maximum average quality factor (FOM) of the device as the optimization goal and the Helmholtz equation as the constraint, the optimization problem of the liquid crystal-based adjustable mode converter model is constructed as shown in formula (4):

[0023]

[0024] The electric field distribution E in the device is obtained by the Helmholtz equation with constraints S , the superscript S represents the state of the device. Multiple angles of the liquid crystal director correspond to multiple states; ω is the operating wavelength of the device; ε represents the relative dielectric constant distribution of the device. After interpolation, the relative permittivity distribution is obtained, where represents the density distribution of structural parameters after binarization, Gaussian blur and morphological operators; J S is the current source used to excite TE mode light in the input waveguide. Maximizing the FOM is achieved through the gradient update structure constructed in step 5. The gradient is solved using the simulation data from the forward and backward simulations.

[0025] The forward simulation excites the TE0 mode in the device's input waveguide and calculates the XY plane electric field distribution data in the mode conversion region, located at half the device layer thickness. The electric field distribution of the output and input waveguides is used to calculate the device's figure of merit (FOM) for this structural parameter. A larger value indicates better device performance and is between 0 and 1. It is defined as follows:

[0026]

[0027] Where E represents the electric field distribution in the output waveguide XY plane obtained by simulation, represents the desired electric field distribution in the output waveguide XY plane, and the superscript S represents the Sth state in the multifunctional device.

[0028] The back-propagation simulation is to place the formula at the output waveguide of the device. The current source excites the light field and calculates the XY plane electric field distribution data of the mode conversion region located at half the thickness of the device layer. The electric field data obtained through forward simulation and backward simulation are processed to obtain the gradient of the objective function with respect to the structural parameters:

[0029]

[0030] in, Represents the device electric field distribution data of the forward simulation, Represents the device electric field distribution data of the backward simulation.

[0031] After completing the forward simulation of each state to obtain the electric field distribution data and quality factor FOM, a logical branching structure is added. By comparing the quality factor FOM of each state, the liquid crystal state that needs to be reversely simulated is determined to avoid the quality factor of a certain state of the adjustable device being much lower than that of other states, thereby improving the average quality factor FOM of the adjustable device. The method of adding a logical branching structure to determine the liquid crystal state that needs to be reversely simulated is as follows: compare the sizes of multiple quality factor FOMs. If the quality factor FOM of a certain state is much larger than that of other states, the reverse simulation of that state will not be run, and only the gradients of other states will be solved for structural updates. Only when the quality factor FOMs of all states are roughly the same will the gradients of each state be solved simultaneously to update the unit cube structure parameters, thereby avoiding the quality factor of a certain state of the adjustable device being much lower than that of other states, that is, improving the average quality factor of the adjustable device.

[0032] Preferably, COMSOL is selected as the optical simulation software.

[0033] Step 5: Gradient data processing and structural parameter update. After obtaining the gradient in step 4, the gradient data is processed using the ADAM optimizer and the parameters are updated to better converge the device to the local optimal solution. The optimized adjustable mode converter structural parameters are obtained, thus achieving the optimization of the liquid crystal-based adjustable mode converter.

[0034] Use formula (7) to update the parameters:

[0035]

[0036] Among them, lr represents the learning rate, which is a very important hyperparameter and has a great influence on the convergence of the device; Represents the gradient data of each state processed by the ADAM optimizer.

[0037] Step 6: Optimize the tunable mode converter structural parameters obtained in Step 5. Etch the tunable mode converter onto the wafer, and then etch the waveguide and grating coupler onto the wafer. Liquid crystal is then used to fill the mode conversion region and input and output waveguides of the device, completing the fabrication of the liquid crystal cell. The resulting tunable mode converter has a quality factor that meets the pre-determined performance requirements, significantly reducing crosstalk between optical signals in different modes and increasing the switching speed of optical signals in different modes.

[0038] The liquid crystal-based adjustable mode converter is obtained according to the liquid crystal-based adjustable mode converter optimization method. The adjustable mode converter is used to convert the fundamental mode light introduced from the input waveguide into multiple high-order mode lights. The conversion into multiple high-order mode lights is mainly used for data exchange between different channels in optical communication links. It can also be used for on-chip multi-channel sensor chips to cooperate with intelligent algorithms for flexible data processing. The adjustable mode converter is combined with a mode demultiplexer to realize the upload and download of multiple signals, significantly reducing the crosstalk between different mode optical signals and improving the switching speed of different mode optical signals.

[0039] Beneficial effects:

[0040] 1. The liquid crystal-based adjustable mode converter and its optimization method disclosed in the present invention construct a liquid crystal-based adjustable mode converter optimization model, utilize an adjoint algorithm, and mainly update the structure of the gradient of the structural parameters based on the calculated objective function to achieve a faster convergence speed. Logical judgment is used to minimize computing resource consumption to improve the efficiency of device generation while ensuring that the device target function meets the preset performance requirements.

[0041] 2. The present invention discloses a liquid crystal-based adjustable mode converter and an optimization method thereof. After completing the forward simulation of each state to obtain the electric field distribution data and the quality factor FOM, a logical branch structure is added to judge the liquid crystal state that needs to be reversely simulated by comparing the quality factor FOM of each state, so as to avoid the quality factor of a certain state of the adjustable device being much lower than that of other states, that is, to improve the average quality factor FOM of the adjustable device.

[0042] 3. The present invention discloses a liquid crystal-based adjustable mode converter and its optimization method, which uses a smaller Gaussian blur radius and adjusts the density Perform opening operation to directly erase the tiny cuboid and adjust the density Perform closing operation to fill the tiny holes. Gaussian blur processing, opening and closing operations are performed to improve the device's quality factor and robustness, reduce device processing difficulty, and reduce the strong anchoring effect of the microstructure on the liquid crystal molecules, reducing the restrictions on the device's degrees of freedom caused by Gaussian blur. The anchoring effect affects the tunable properties of the liquid crystal.

[0043] 4. This invention discloses a liquid crystal-based tunable mode converter and its optimization method. Liquid crystals offer the advantages of low optical loss and fast switching speeds, and the tunable material is liquid crystal. The tunable device achieves control of the relative dielectric constant by controlling the angle of the liquid crystal director. A higher birefringence of the liquid crystal results in a higher average device figure of merit (FOM).

[0044] 5. The liquid crystal-based adjustable mode converter and its optimization method disclosed in the present invention, on the basis of achieving the above-mentioned beneficial effects 1, 2, 3, and 4, is used to convert the optimized adjustable mode converter from the fundamental mode light introduced from the input waveguide into multiple high-order mode lights. The converted multiple high-order mode lights are mainly used for data exchange between different channels in optical communication links, and can also be used for on-chip multi-channel sensor chips to cooperate with intelligent algorithms for flexible data processing. The present invention can generate multifunctional, compact, high-performance on-chip optical devices, providing new components for optical communication systems and optical sensing systems. The adjustable mode converter is combined with a mode demultiplexer to realize the upload and download of multiple signals, significantly reducing the crosstalk between different mode optical signals and improving the switching speed of different mode optical signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] FIG1 is a flow chart of the liquid crystal-based adjustable mode converter optimization method disclosed in the present invention;

[0046] FIG2 is a schematic diagram of the device of the present invention on the left; FIG2 is a schematic diagram of the function of the present invention on the right;

[0047] Figure 3 is a schematic diagram showing a mode conversion region of an adjustable mode converter according to an exemplary embodiment;

[0048] Figure 4 It is the process of processing the initial values ​​of the design method and structural parameters;

[0049] in: Figure 4 -a is the initial structural parameter distribution of the design method, black represents the wafer device layer, Figure 4 -b is a local enlarged view of the structural parameters, Figure 4 -c is the structural parameter of the local magnified area after Gaussian blurring, Figure 4 -d is the structural parameter of the local amplified area after binary mapping, Figure 4 -d is the structural parameter of the local magnified area after opening and closing operations.

[0050] Figure 5 To investigate the relationship between the birefringence of liquid crystals and the average quality factor of the device, the structure of a tunable mode converter was optimized using three liquid crystal materials with different birefringence. The figure lists the normalized quality factors of each state of the three-state tunable mode converter under the three liquid crystals.

[0051] Figure 6 is the quality factor of different algorithm structures at different numbers of iterations.

[0052] Figure 7 It is the simulation result of the present invention;

[0053] in: Figure 7 -a is a schematic diagram of the structure of the mode conversion region 4 of the final device, Figure 7 -b is the XY plane light intensity distribution when the liquid crystal in liquid crystal cell 5 is in state 1 and the device is located in the middle in the Z direction, Figure 7 -c is the XY plane light intensity distribution when the device is in the middle in the Z direction when the liquid crystal is in state 2, Figure 7 -d is the XY plane light intensity distribution when the device is in the middle in the Z direction when the liquid crystal is in state 3, Figure 7 -e is the quantization of the bcd results. When the x-axis is labeled as state 1, only the TE0 mode is output, and the light of other modes is very weak. When the x-axis is labeled as state 2, only the TE1 mode is output. When the x-axis is labeled as state 3, only the TE2 mode is output.

[0054] Figure 8 It is the experimental schematic diagram and the SEM image of the actual device. Figure 8 The middle area of ​​Figure a is a SEM image of the entire system, which consists of two grating couplers, waveguides, a tunable mode converter, and a mode demultiplexer. The light source is a supercontinuum laser (SCW), and detection is performed using a spectrometer. Figure 8 -b is the SEM image of the adjustable mode converter, the core component of the system; Figure 8 -c is the SEM image of the mode demultiplexer.

[0055] Figure 9 is the system wavelength response curve obtained experimentally.

[0056] In the figure: 1—substrate, 2—input waveguide, 3—output waveguide, 4—mode conversion region, 5—liquid crystal cell. DETAILED DESCRIPTION

[0057] In order to better illustrate the purpose and advantages of the present invention, the invention is further described below with reference to the accompanying drawings.

[0058] like Figure 1 As shown, the present embodiment discloses a liquid crystal-based adjustable mode converter, and the specific implementation steps are as follows:

[0059] Step 1: Construct a liquid crystal-based adjustable mode converter model. The liquid crystal-based adjustable mode converter is formed by wafer etching. The wafer has silicon dioxide as the substrate 1 and 800nm ​​thick silicon nitride as the device layer. After etching the input waveguide 2, output waveguide 3 and mode conversion area 4 on the device layer, liquid crystal material is filled and packaged to form a liquid crystal box 5, as shown in FIG. Figure 2 The mode conversion region 4 is divided on the wafer, and the length L0 of the mode conversion region 4 is 14μm and the width W0 is 16μm. TE0 mode light is introduced into the input waveguide 2, and after passing through the mode conversion region 4, multiple high-order mode lights can be output from the output waveguide 3. The height of the adjustable mode converter is the height of the wafer device layer. Figure 3 The mode conversion region 4 is evenly divided into 140*160 unit cuboids in three-dimensional space. The length, width and height of the unit cube are 100nm*100nm*800nm. The material state of a single unit cube is divided into adjustable material or wafer device layer. The black square in the figure represents the unit cube material state as the wafer device layer, and the white square in the figure represents the unit cube material state as the adjustable material. The unit cube structural parameters are represented by density ρ, and the density ρ value is between 0-1, where: 0 represents that the unit cuboid is an adjustable material, 1 represents the device layer material, and the intermediate value is invalid. The unit cuboid represented by 0 is the adjustable material filling the unit cube space etched out in the mode conversion region on the wafer. The mode conversion region array formed by the device layer unit cuboid and the adjustable material unit cuboid serves as an adjustable mode converter.

[0060] Step 2: Use formula (1) to Gaussian blur the high-frequency information in the density distribution ρ. Adding Gaussian blur processing reduces the difficulty of device processing and increases the robustness of the device. Figure 4 There is a structure with a sudden increase in density in a, which is high-frequency information in the density distribution. Figure 4 b is the density distribution after the Gaussian blur algorithm. After processing, the structure with a sudden increase in density is transformed into a gradient structure with a lower spatial frequency. Formula (3) is used for binary mapping to judge the parameters at the intermediate value as adjustable materials or device layers, eliminating the invalid parameters that cannot be produced during the iteration of the unit cube structure parameters. Figure 4 c is the density distribution after Gaussian blur and binarization mapping. After processing, the size of the hole is reduced. Figure 4 a Hole size, after Gaussian blur and binary mapping, the hole can be closed. After Gaussian blur and binary mapping, the isolated structure of the device can be greatly reduced, and the integrity of the device layer in the device mode conversion area 4 can be increased, that is, most of the device layer unit cubes are adjacent to each other; Perform opening operation to directly erase the tiny cuboid and adjust the density Perform closing operation to fill the tiny holes. Gaussian blur processing, binary mapping, opening and closing operations are performed to improve the device quality factor and robustness, reduce the difficulty of device processing, and reduce the strong anchoring effect of these tiny structures on liquid crystal molecules, which will affect the adjustable properties of liquid crystals.

[0061] Step 3: Liquid crystal has the advantages of low optical loss and fast switching speed, so the tunable material is liquid crystal. The tunable device can adjust the relative dielectric constant by controlling the angle of the liquid crystal director. Figure 5 Using a variety of liquid crystal materials and the same optimization method, the quality factor FOM of the device in each liquid crystal state was obtained. It was found that the greater the birefringence of the liquid crystal, the larger the average quality factor FOM of the device obtained. Among them, the birefringence of E7 liquid crystal is the largest, and the average quality factor of the device obtained is the largest.

[0062] Step 4: The present invention discloses an optimization method for a liquid crystal-based adjustable mode converter. The optimization objective is to maximize the average quality factor (FOM) of the device in formula (5). The Helmholtz equation in formula (4) is used as a constraint to construct a model optimization problem for the liquid crystal-based adjustable mode converter. The adjoint algorithm is used to obtain the gradient of the objective function with respect to the structural parameters, primarily based on the electric field distribution data obtained by forward and backward simulations. This process is described by formula (6). The gradient is used to update the structural parameters in a directional manner to achieve a faster convergence speed. The electric field distribution data obtained by forward and backward simulations are located in the XY plane, with the Z-axis coordinate being half the height of the device layer when the Z-axis origin is at the interface between the substrate and the device layer. Specifically, the electric field data within the mode conversion region 4 on this plane are obtained.

[0063] Adding logical judgment ensures that the target function of the device meets the preset performance requirements, while minimizing the consumption of computing resources to improve the efficiency of generating devices. Figure 1 After completing the forward simulation of each state to obtain the electric field distribution data and quality factor FOM, a logical judgment structure is added to determine the liquid crystal state that needs to be simulated in the reverse direction by comparing the quality factor FOM of each state. When the quality factor FOM of a certain state is much larger than that of other states, the reverse simulation will not be performed on that state. By adding a logical branch structure, the liquid crystal state with a quality factor FOM that is too large cannot obtain the electric field distribution data for the reverse simulation, and it is also impossible to obtain the gradient data of this state through formula (6). After using the gradient of other states to update the structural parameters, the liquid crystal state with a quality factor FOM that is too large will cause the quality factor FOM to decrease due to the lack of gradient. Figure 6 Figure 2 shows the quality factor of the structure using different optimization methods at different iteration times. Comparing the average quality factor of the device after the two optimization methods are completed, the average quality factor of the structure with logic branches is much higher than that of the optimization method based on stochastic gradient descent.

[0064] Step 5: Process the gradient data and update the structural parameters. After obtaining the gradient in step 4, use the ADAM optimizer to process it so that the device can better converge to the local optimal solution. Use formula (7) to update the parameters. The optimized adjustable mode converter structural parameters are obtained, thus achieving the optimization of the liquid crystal-based adjustable mode converter. Figure 7 The optimized mode converter structure is divided into the middle mode conversion region 4. This structure is obtained by the optimization algorithm. The input waveguide 2 on the left is a single-mode waveguide with a waveguide width W_input of 1 μm. The output waveguide 3 on the right is a multimode waveguide with a waveguide width W_output of 2 μm. Figure 7 d, when the liquid crystal director is along the positive direction of the X-axis, the TE0 mode with a wavelength of 1550 nm is input from the input waveguide 2, converted into the TE2 mode signal through the mode conversion region 4 and output from the output waveguide 3; Figure 7 c. When the liquid crystal director rotates 33° counterclockwise, the TE0 mode with a wavelength of 1550 nm input from input waveguide 2 is converted into a TE1 mode signal through mode conversion region 4 and output from output waveguide 3. Figure 7 b When the liquid crystal director rotates 90° counterclockwise, the TE0 mode with a wavelength of 1550nm is input from the input waveguide 2, converted into a TE0 mode signal through the mode conversion region 4 and output from the output waveguide 3. Figure 7 e represents the proportion of each mode light in the output waveguide 3 of the device in three states. In state 1, TE0 light accounts for the majority of the output waveguide 3, and the crosstalk is below -11.35 dB. In state 2, TE1 light accounts for the majority of the output waveguide 3, and the crosstalk is below -13.77 dB. In state 3, TE2 light accounts for the majority of the output waveguide 3, and the crosstalk is below -20.30 dB, meeting the requirements of practical applications.

[0065] Step 6: Optimize the tunable mode converter structural parameters obtained in Step 5. Use a focused ion beam (FIB) to etch the tunable mode converter into the device layer of the wafer. Also, etch the waveguide and grating coupler onto the wafer. Liquid crystal is then used to fill the mode conversion region and input and output waveguides of the device, completing the fabrication of the liquid crystal cell. The resulting tunable mode converter has a quality factor that meets the preset performance requirements, significantly reduces crosstalk between optical signals in different modes, and improves the switching speed of optical signals in different modes. To verify the feasibility of the device, an optimized design and experiments were conducted using a filling material switch, replacing the refractive index change caused by a change in the liquid crystal director. The filling material switch was achieved using water and air. When the tunable mode converter is not filled with water, the output waveguide 3 converts the TE0 mode light introduced from the input waveguide 2 into the TE1 mode light, representing State 2 (STATE 2). When the tunable mode converter is filled with water, the output waveguide 3 converts the TE0 mode light introduced from the input waveguide 2 into the TE0 mode light, representing State 1 (STATE 1). To better observe the mode conversion effect of the tunable mode converter, a mode demultiplexer is connected to the tunable mode converter output waveguide 3. The mode demultiplexer switches the different modes of light output by the tunable mode converter to different tunable mode demultiplexer output waveguides. The tunable mode converter uses a common silicon-on-insulator wafer, meaning the device layer is made of single-crystal silicon with a thickness of 220 nm. The device operates at a wavelength of 1550 nm. Figure 8 It is a schematic diagram of the system experiment and a scanning electron microscope (SEM) image of the device. Figure 8 The center of the image (a) shows a SEM image of the entire system. The light source is a supercontinuum laser (SCW). The output laser passes through a linear polarizer to obtain linearly polarized light. The polarization direction of the optical fiber output is controlled by a three-ring polarization controller. The optical fiber output is coupled to the silicon wafer via a grating coupler. The output is also coupled out of the optical fiber via a grating coupler. The spectral response of the entire system is measured using a spectrometer. Figure 8 b is the SEM image of the mode conversion region 4 of the adjustable mode converter.

[0066] Figure 8 c is the SEM image of the mode demultiplexer. Figure 9 is the spectral response curve of the device. Figure 9 a Spectral response of state 1. When filled with water, the output port is port 1, which is the output waveguide at the upper end of the mode demultiplexer. Figure 9 b) When there is no filler, the output port is port 2, which is the output waveguide at the lower end of the mode demultiplexer. The experiment demonstrated that the output port switches under different refractive index conditions, indicating that the tunable mode converter outputs different light modes when the refractive index of the filler material changes.

[0067] According to the optimization method of the adjustable mode converter based on liquid crystal, the adjustable mode converter based on liquid crystal is obtained. The adjustable mode converter is used to convert the fundamental mode light introduced from the input waveguide into multiple high-order mode lights. The conversion into multiple high-order mode lights is mainly used in optical communication links for data exchange of different channels, and can also be used for on-chip multi-channel sensor chips to cooperate with intelligent algorithms for flexible data processing. The adjustable mode converter is combined with a mode demultiplexer to realize the uploading and downloading of multiple signals, significantly reducing the crosstalk between different mode optical signals and improving the switching speed of different mode optical signals. The optimization method of the adjustable mode converter based on liquid crystal disclosed by the present invention can generate multifunctional, compact, high-performance on-chip optical devices, providing a series of multifunctional devices based on liquid crystal or other adjustable materials for optical communication systems and optical sensing systems.

[0068] The above specific description further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for optimizing a liquid crystal-based adjustable mode converter, characterized in that: The following steps are involved: Step 1: Construct a liquid crystal-based adjustable mode converter model. The liquid crystal-based adjustable mode converter is formed by wafer etching. The etched wafer includes a waveguide and a mode conversion region. The mode conversion region is divided on the wafer, and the length of the mode conversion region is L0 and the width is W0. The hyperparameters are pre-set according to the limitations of the wafer etching process and computing resources. The hyperparameters include the length L0, width W0, height H0 and the minimum feature size of the mode conversion region. The height of the adjustable mode converter is the height of the wafer device layer. The mode conversion region is evenly divided into n *m unit cuboids, the material state of a single unit cuboid is divided into adjustable material or wafer device layer; the number of unit cuboids n*m is determined by setting the length, width and size of the mode conversion region and the unit cuboid; the height of the unit cuboid is equal to the height of the adjustable mode converter; the unit cuboid structural parameters are represented by density ρ, and the density ρ value is between 0 and 1, where: 0 represents that the unit cuboid is an adjustable material, 1 represents a device layer material, and the intermediate value is invalid; the unit cuboid represented by 0 fills the unit cuboid space etched away in the mode conversion region on the wafer; A mode conversion region array is formed by device layer unit cuboids and tunable material unit cuboids as an adjustable mode converter; Step 2: Constrain and process the unit rectangular parallelepiped structural parameters; use the Gaussian blur algorithm to filter the high-frequency information in the density distribution ρ, that is, small rectangular parallelepipeds or holes, sharp corners, and add Gaussian blur processing to reduce the difficulty of device processing and increase the robustness of the device; use binary mapping to judge the parameters in the middle value as adjustable materials or device layers, and eliminate the invalid parameters that cannot be produced during the iteration of the unit rectangular parallelepiped structural parameters; Perform opening operation to directly erase the tiny cuboid and adjust the density Perform closing operations to fill tiny holes; Step 3: Liquid crystal materials are selected as the tunable material; the tunable device adjusts the relative dielectric constant by controlling the angle of the liquid crystal director; the greater the birefringence of the liquid crystal, the greater the average quality factor (FOM) of the device obtained; Step 4: With the device average quality factor (FOM) as the optimization goal and the Helmholtz equation as the constraint, a model optimization problem for a liquid crystal-based adjustable mode converter is constructed. Using the adjoint algorithm, the gradient of the objective function with respect to the structural parameters is obtained based on the electric field distribution data obtained from forward and backward simulations. The structural parameters are updated in a directed manner using the gradient to achieve faster convergence, and logical judgment is added. Step 5: Gradient data processing and structural parameter update: After obtaining the gradient in step 4, the gradient data is processed using the ADAM optimizer and the parameters are updated to converge the device to a local optimal solution. The optimized adjustable mode converter structural parameters are obtained, thus achieving the optimization of the liquid crystal-based adjustable mode converter. Step 6: Optimize the structural parameters of the adjustable mode converter according to step 5, etch the adjustable mode converter on the wafer, and etch the waveguide and grating coupler on the wafer; use liquid crystal to fill the mode conversion area and input and output waveguides of the device with liquid crystal, and then complete the production of the liquid crystal box; obtain an adjustable mode converter with a quality factor that meets the preset performance requirements.

2. The method for optimizing a liquid crystal-based adjustable mode converter according to claim 1, wherein: Step 2 is implemented as follows: According to formula (1), the Gaussian blur algorithm is used to eliminate the elongated cuboids or tiny holes in the structure; Parameters representing the density distribution after Gaussian operation: Where D is the set of unit cuboids with Gaussian blur radius R around the i-th unit cuboid, r i and r j is the position of the i-th and j-th unit cuboids, h ij Represents weight; According to formula (3), the binary mapping is used to judge the parameters in the middle value as adjustable materials or device layers, eliminating the invalid parameters that cannot be produced during the iteration of the unit rectangular parallelepiped structure parameters; the density after binarization is used express: Among them, θ controls the intensity of binarization, and η is the center of the curve. When the binarization intensity is large, density less than η will be mapped to 0, while density greater than η will be mapped to 1. Use a small Gaussian blur radius and adjust the density Perform opening operation to directly erase the tiny cuboid and adjust the density Perform closing operations to fill tiny holes; Perform Gaussian blur processing, opening operation, and closing operation.

3. The optimization method of a liquid crystal-based adjustable mode converter according to claim 2, wherein: Step 4 is implemented as follows: The electric field distribution in the mode conversion region obtained by forward simulation and backward simulation is used to calculate the gradient of the objective function relative to the structural parameters according to the electric field distribution. Taking the maximum average quality factor (FOM) of the device as the optimization goal and the Helmholtz equation as the constraint condition, the optimization problem of the adjustable mode converter model based on liquid crystal is constructed as shown in formula (4): The electric field distribution E in the device is obtained by the Helmholtz equation with constraints S , the superscript S represents the state of the device, and multiple angles of the liquid crystal director correspond to multiple states; ω is the angular frequency; ε represents the relative dielectric constant distribution of the device, through After interpolation, the relative permittivity distribution is obtained, where represents the density distribution of structural parameters after binarization, Gaussian blur and morphological operators; J S is a current source used to excite TE mode light in the input waveguide; maximizing the quality factor FOM is achieved through the gradient update structure constructed in the subsequent step 5, and the gradient is solved using the simulation data of the forward simulation and the backward simulation; The forward simulation excites the TE0 mode in the input waveguide of the device and calculates the XY plane electric field distribution data of the mode conversion region located at half the device layer height. The electric field distribution of the output waveguide and the input waveguide are used to obtain the quality factor (FOM) of the device at this structural parameter. The figure of merit is between 0 and 1 and is defined as follows: Where E represents the electric field distribution in the output waveguide XY plane obtained by simulation, represents the desired electric field distribution in the output waveguide XY plane, and the superscript S represents the Sth state in the multifunctional device; The back-propagation simulation is to place the formula at the output waveguide of the device. The current source excites the light field and calculates the XY plane electric field distribution data of the mode conversion region located at half the device layer height. The electric field data obtained through forward simulation and backward simulation are processed to obtain the gradient of the objective function with respect to the structural parameters: in, Represents the device electric field distribution data of the forward simulation, Represents the device electric field distribution data of the backward simulation; After completing the forward simulation of each state to obtain the electric field distribution data and quality factor FOM, a logical branch structure is added; the quality factor FOM of each state is compared to determine the liquid crystal state that needs to be backward simulated; the method of adding a logical branch structure to determine the liquid crystal state that needs to be backward simulated is as follows: compare the sizes of multiple quality factor FOMs. If the quality factor FOM of a certain state is much larger than that of other states, the backward simulation of this state will not be run, and only the gradients of other states will be solved for structural update; only when the quality factor FOMs of all states are roughly the same, the gradients of each state will be solved at the same time to update the unit cuboid structure parameters, so as to avoid the quality factor of a certain state of the adjustable device being much lower than that of other states, that is, to improve the average quality factor of the adjustable device.

4. The method for optimizing a liquid crystal-based adjustable mode converter according to claim 3, wherein: In step 5, Use formula (7) to update the parameters: Among them, lr represents the learning rate, which is a very important hyperparameter; Represents the gradient data of each state processed by the ADAM optimizer.

5. The method for optimizing a liquid crystal-based adjustable mode converter according to claim 2, 3 or 4, wherein: The center η of the curve is chosen to be 0.

5.

6. The method for optimizing a liquid crystal-based adjustable mode converter according to claim 2, 3 or 4, wherein: COMSOL is the optical simulation software.

7. A liquid crystal-based adjustable mode converter, characterized in that: The optimization method of the liquid crystal-based tunable mode converter according to claim 2, 3 or 4 is implemented, wherein the tunable mode converter is used to convert fundamental mode light introduced from an input waveguide into a plurality of high-order mode lights.

Images