A power beam splitter based on phase change material, design method and image recognizer

Abstract

The present invention belongs to the field of optical devices, and specifically relates to a power beam splitter based on phase change material, a design method, and an image recognizer. The power beam splitter comprises an input waveguide, a coupling region, and two output waveguides connected in sequence. The coupling region comprises a phase change material array region and an optimization region. The two output waveguides are respectively connected to the phase change material array region and the optimization region. The optimization region is divided into a plurality of units, each unit having two states, and the state of each unit is determined according to a direct binary search algorithm. The present invention significantly reduces device size, reduces loss and crosstalk, and achieves high-precision and wide power distribution.

Description

Technical Field

[0001] The present invention belongs to the field of optical devices, and in particular relates to a power beam splitter based on phase change material, a design method and an image recognizer. Background Art

[0002] In recent years, many theoretical and experimental studies have been devoted to the miniaturization of optical devices. As a very basic component of integrated photonic circuits, power beam splitters play a key role in the design of advanced signal processing applications and are widely used in various applications, such as high-speed optical communications and advanced radar systems. In the past, multimode interferometers (MMIs) have been the first choice for achieving this function due to their small size, large bandwidth and good manufacturing tolerance. However, traditional MMIs can only distribute power equally to each output channel based on the principle of self-imaging. Obviously, power beam splitters with arbitrary power ratio design freedom are more attractive because they will have a wider range of applications, such as feedback circuits, tap port power monitoring, power balancing or power imbalance splitting circuits, etc. For example, Ke Xu et al. designed a 1×4 multi-channel arbitrary power beam splitter with a power splitting ratio of 1:2:4:8 by cascading a 1×3 multi-channel MMI coupler with a phase shifter and a 4×4 multi-channel MMI coupler [Li Z, Liu Y, Guan H, et al. Ultra-compact low-loss 1×4 optical power splitter with splitting ratio of 1:2:4:8based on two-stage cascaded MMI couplers. Optics Letters, 2019, 44: 5622-5625.]. However, the design method of the arbitrary power beam splitter above uses manual parameter adjustment technology, which is limited to a few parameters intuitively selected based on design experience. The device structure is large (ranging from tens of microns to hundreds of microns) and is costly, which is not conducive to dense photonic integrated circuits. Therefore, finding an arbitrary power beam splitter with excellent performance and low cost is a current problem that needs to be solved.

[0003] With the rapid development of algorithmic optical devices in recent years, they can flexibly design the refractive index distribution of devices. This allows for the manipulation of light fields within subwavelength ranges without the need to predict the initial distribution, facilitating the design of ultra-compact, high-performance devices. Many algorithmic power splitting devices have broken through the symmetric structures of traditional devices, achieving superior performance. Ke Xu et al. first used the direct binary search (DBS) algorithm to design an arbitrary power splitter and achieved an 80% transmission efficiency [K.Xu, L.Liu, X.Wen, W.Sun, N.Zhang, N.Yi, S.Sun, S.Xiao, and Q.Song, “Integrated photonic power divider with arbitrary power ratios,” Optics Lett. 42, 855 (2017).]. However, these silicon-based photonic integrated circuit devices exhibit the drawback of being uncontrollable: a single device structure can only achieve a single power splitting ratio. To achieve a different power splitting ratio, a new device must be fabricated, hindering the integrated development of photonic integrated circuits. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a power beam splitter based on phase change material, a design method and an image recognizer, which significantly reduces the device size, reduces loss and crosstalk, and achieves high-precision and wide power distribution.

[0005] An embodiment of the present invention provides a power beam splitter based on phase change material, comprising an input waveguide, a coupling region, and two output waveguides connected in sequence, wherein the coupling region comprises a phase change material array region and an optimized region, and the two output waveguides are respectively connected to the phase change material array region and the optimized region;

[0006] The optimization region is divided into a plurality of units, each unit has two states, and the state of each unit is determined according to a direct binary search algorithm.

[0007] The input waveguide, coupling region and output waveguide of the present invention respectively include a substrate (generally silicon dioxide) and a silicon material deposited on the substrate, such as a standard SOI waveguide with a 220 nm thick silicon layer and a 3 μm thick silicon dioxide layer.

[0008] Optionally, the phase change material array region includes a plurality of phase change blocks composed of phase change material and arranged at intervals and embedded in the coupling region. Preferably, the phase change blocks are in the shape of long strips.

[0009] Optionally, the arrangement direction of the phase change blocks is perpendicular to the transmission direction of light.

[0010] Preferably, the phase change material is Sb2Se3.

[0011] The main body of the coupling region of the present invention is made of silicon dioxide and silicon material deposited on the silicon dioxide. The meaning of embedding the phase change block in the coupling region is to etch a hole in the silicon material (generally, the etching stops at the upper surface of the substrate), and the phase change material is poured into the hole to obtain the phase change block. The phase change material is embedded in the silicon layer. When it is in the C state (the phase change material is crystalline), it can effectively guide light to be output from the upper channel (corresponding to the output waveguide connected to the phase change material array area) from the specified track. In the A state (the phase change material is amorphous), the refractive index is almost the same as that of Si, so that the light will not be constrained by the phase change material array area as in the C state, and will enter the optimized area and be output from the lower channel. Compared with the phase change material directly deposited on the surface of the coupling region, it can effectively reduce loss and crosstalk and increase the power distribution width.

[0012] Optionally, the status of each unit is determined as follows:

[0013] Initialize the state of each unit; then change the state of a unit and compare the transmittance of light before and after the change. If the transmittance after the state change is greater than the transmittance before the state change, retain the state of the changed unit. Repeat the above method for all units until they become stable and obtain the state of each unit.

[0014] Optionally, the transmittance is the sum of two transmittance differences, wherein one transmittance difference is the absolute value of the difference in transmittance of light passing through two output waveguides when all phase change materials are crystalline; and the other transmittance difference is the absolute value of the difference in transmittance of light passing through two output waveguides when all phase change materials are amorphous.

[0015] The objective function of transmittance is expressed by the following formula:

[0016] FOM c-Sb2Se3 =T upper -T lower

[0017] FOM a-Sb2Se3 =T lower -T upper

[0018] FOM total =FOM c-Sb2Se3 +FOM a-Sb2Se3

[0019] Where T upper 、T lowerThey are the TE0 mode transmittance of the upper and lower channels (the upper channel is the output waveguide corresponding to the phase change material array area, and the lower channel is the output waveguide corresponding to the optimization area) in the wavelength range of 1530-1570nm.

[0020] The stability means that the ratio of the difference in transmittance before and after the state change to the transmittance before or after the state change is lower than a certain value, such as lower than 0.1%, which means that it is stable and the objective function is convergent.

[0021] Optionally, the two states of the unit are an etching state and a non-etching state, and the etching state is to etch the center of the unit into an air circular hole.

[0022] The present invention provides a design method for a power beam splitter based on phase change materials. The optimized area is obtained by optimizing the following steps: initializing the state of each unit; then changing the state of a certain unit, comparing the transmittance of light before and after the change; if the transmittance after the state change is greater than the transmittance before the state change, retaining the state of the unit after the change; and repeatedly editing all units in the above manner until they tend to be stable, thereby obtaining the state of each unit.

[0023] The present invention provides an image recognizer, comprising the power beam splitter based on phase change material.

[0024] The beneficial effects of the present invention are that the phase-change material Sb2Se3, as a material that can change its own optical properties, can exist in two states: crystalline and amorphous. These two states are reversible, and can be switched between the crystalline state (c-Sb2Se3) and the amorphous state (a-Sb2Se3) by applying external light pulses or electrical pulses, thereby changing its optical properties. In addition, Sb2Se3 is non-volatile and maintains a constant optical state in the absence of power input.

[0025] This paper utilizes the DBS algorithm (direct binary search algorithm) to propose a non-volatile programmable power beam splitter based on phase-change materials. By changing the state distribution of the phase-change material (e.g., an Sb2Se3 array), different power distribution ratios can be achieved with high precision in a very compact structure (only approximately 2.34 microns in width and 4.92 microns in length). This device has the advantages of being extremely small, non-volatile, and digitally programmable, which is beneficial for true optoelectronic fusion chips. In addition, this device is used to implement a compact and scalable optical computing unit in the optical convolutional neural network architecture, and is used to achieve deep learning tasks with high speed, low power consumption, and a small footprint, greatly reducing the computational cost of matrix multiplication in the convolutional layer.

[0026] The present invention uses a phase change material array and an optimized area (commonly an air hole array) in combination. Compared with the use of only phase change material arrays or only optimized areas, it can better control the effective refractive index and change the light field distribution in a limited size space, thereby achieving high-precision and wide power distribution. At the same time, it also significantly reduces the device size, reduces loss and crosstalk. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the top structure of the present invention.

[0028] Figure 2 It is a side structural schematic diagram of the present invention.

[0029] Figure 3 The light field distribution (left) and output channel transmission curve (right) of the phase change material array under full a-Sb2Se3.

[0030] Figure 4 The light field distribution (left) and output channel transmission curve (right) of the phase change material array under full c-Sb2Se3.

[0031] Figure 5 The power contrast τ map of the two output channels is controlled at 128 different levels for the present invention.

[0032] Figure 6 The handwriting recognition accuracy diagram of the present invention on a 64-bit computer (left figure) and OCNN (right figure).

[0033] In the figure, 1 is input waveguide, 2 is phase change material array area, 3 is optimization area, 31 is air hole, 4 is upper output waveguide, and 5 is lower output waveguide. DETAILED DESCRIPTION

[0034] like Figure 1-2 As shown, a power beam splitter based on phase change material includes an input waveguide 1, a coupling region and two output waveguides connected in sequence, wherein the coupling region includes a phase change material array region 2 and an optimized region 3, and the two output waveguides are respectively connected to the phase change material array region 2 and the optimized region 3;

[0035] The optimization region is divided into a plurality of units, each unit has two states, and the state of each unit is determined according to a direct binary search algorithm.

[0036] The two output waveguides include an upper output waveguide 4 and a lower output waveguide 5 . The upper output waveguide 4 is connected to the phase change material array area 2 , and the lower output waveguide 5 is connected to the optimization area 3 .

[0037] The input waveguide 1 and output waveguides are both standard SOI waveguides constructed with a 220nm-thick silicon layer and a 3μm-thick silicon dioxide layer. The input waveguide 1 is 620nm wide, the upper output waveguide 4 is 620nm wide, and the lower output waveguide 5 is 720nm wide. The optimized region has a width of W = 2.34μm and a length of L = 4.92μm.

[0038] The silicon photonic waveguide coupling region is divided into two upper and lower regions. The upper coupling region is the phase change material array region 2, namely the Sb2Se3 array, with a total of 30 units, each unit is 90nm long and 620nm wide; the lower coupling region is the optimized region 3, namely the air hole array, with a total of 40×12 units, each cell is a 120nm×120nm square, and the center is filled with an air hole. The radius of the air hole is 45nm and the hole depth is 220nm.

[0039] The present invention optimizes the device's circular air hole array using a direct binary search algorithm. The state of each circular air hole cell is determined by calculating the state of each cell based on a set objective function to maximize the objective function. Each cell in the coupled region is sequentially optimized, the state of the scanned cell is changed, the current objective function is calculated, and the current objective function is compared with the objective function value when the cell state remains unchanged. If the current objective function value improves, the new state of the scanned cell is retained; otherwise, the cell is restored to its original state.

[0040] To achieve the specified power splitting function, we set the goal that when the Sb2Se3 array is in the full a-Sb2Se3 (OFF) state (amorphous state), TE0 light is output from the lower output channel; when the Sb2Se3 array is in the full c-Sb2Se3 (ON) state, TE0 light is output from the upper output channel. Therefore, during the device optimization process, the objective function (FOM) is defined as:

[0041] FOM c-Sb2Se3 =T upper -T lower

[0042] FOM a-Sb2Se3 =T lower -T upper

[0043] FOM total =FOM c-Sb2Se3 +FOM a-Sb2Se3

[0044] Where T upper 、T lowerThey are the TE0 mode transmittance of the upper and lower channels in the wavelength range of 1530-1570 nm, respectively. The DBS algorithm is used to optimize the drilling by column scanning until the objective function reaches the target and converges.

[0045] like Figure 3-4 The figure shows the optical field distribution of the Sb2Se3 array device in different states and the spectral response of different output channels in the wavelength range of 1530-1570nm. Within the 40nm bandwidth, the average insertion loss of the device is -0.64dB and the average crosstalk value is -32.06dB when the device is in the full a-Sb2Se3 state; the average insertion loss of the device is -0.46dB and the average crosstalk value is -14.33dB when the device is in the full c-Sb2Se3 state.

[0046] Our device requires embedding a phase-change material into the silicon layer and placing it in close contact with the light field to provide a high refractive index change. However, conventional phase-change materials, Ge2Sb2Te5 and Ge2Sb2Se4Te1, both cause significant transmission losses, thereby reducing device performance. Compared to the former two, the new phase-change material, Sb2Se3, has complex refractive indices of 3.285+0i and 4.05+0i in its amorphous and crystalline states, respectively, and its corresponding extinction coefficients are both 0, significantly reducing the device's transmission losses. Because the refractive index of amorphous Sb2Se3 (a-Sb2Se3) is very close to that of Si, the transmission of coded and controlled light primarily relies on c-Sb2Se3. This ultra-low extinction coefficient means it can be used to construct ultra-compact, efficient power beam-splitting devices. TE0 mode light enters the coupling region from the input waveguide. When the Sb2Se3 array is in the all-a-Sb2Se3 state, the device is in the OFF state and outputs from the lower output channel. Conversely, when the Sb2Se3 array is in the all-c-Sb2Se3 state, the device is in the ON state and outputs from the upper output channel. In addition, the Sb2Se3 array of this device is programmable. By controlling the crystalline or amorphous distribution of each unit in the Sb2Se3 array, the phase of each Sb2Se3 unit in the array will change significantly, which means that the device can dynamically achieve different arbitrary power distribution ratios in the two output channels.

[0047] In addition, the present invention can encode and regulate the state distribution (conversion between crystalline and amorphous states) of the Sb2Se3 array by applying external electrical pulses or optical pulses, thereby controlling the power ratio between the two output channels, and has extremely high performance, small size and low cost.

[0048] Taking the wavelength of 1550nm as an example, we determine the power contrast τ by measuring the power of the upper and lower output channels and calculating their difference. Here we define the power contrast τ as:

[0049] τ=(Tupper -T lower ) / (T lower +T upper ).

[0050] Where T upper and T lower are the transmission powers of the upper and lower output channels at a wavelength of 1550nm, respectively. The ideal range of τ is (-1, 1), while our device can achieve a power contrast ratio τ of (-0.9995, 0.9274), which is close to the ideal case. When the Sb2Se3 array is in a completely a-Sb2Se3 state, the τ value is -0.9995; when the Sb2Se3 array is in a completely c-Sb2Se3 state, the τ value is 0.9274. By encoding the state distribution of 30 Sb2Se3 units, the device can achieve any contrast ratio between -0.9995 and 0.9274. The phase change material array in the device is programmed using a control pulse sequence of a dedicated integrated circuit, programming Sb2Se3 from the amorphous state (0) to the crystalline state (1) or from the crystalline state (1) to the amorphous state (0). Figure 5 The device demonstrates multi-level programmability and achieves 128 distinguishable power contrast levels at 1550 nm, demonstrating its low-loss and high-precision power distribution.

[0051] Here, we use this device to construct the convolution kernel, the core computing unit in the convolutional neural network architecture, and use it to implement deep learning recognition tasks (encoding the distribution of the optical phase change material array, and representing each element in the convolution kernel with multiple target power contrasts τ). Taking the MNIST dataset as an example, it contains 10 handwritten digits ranging from 0 to 9 (the training set and test set have 60,000 and 10,000 images, respectively). The network architecture consists of a convolution layer, a maximum pooling layer, and a fully connected layer. The network is constructed using an open source machine learning framework in Python, and the network is trained using stochastic gradient descent to obtain the desired output. After determining the framework of the optical convolutional neural network, the convolution kernel matrix can be programmed and reprogrammed in the system to complete a large number of complex target recognition tasks.

[0052] Figure 6Figure 3 shows the recognition accuracy of the MNIST test set for a 64-bit computer and our designed optical convolutional neural network. Our optical convolutional neural network (OCNN) achieved an accuracy of 90.74%, while the 64-bit computer achieved an accuracy of 91.38%. As can be seen, the weights obtained using the photonic device are close to those obtained by training on a 64-bit computer, demonstrating that optical network architectures can be used for deep learning recognition tasks. Furthermore, compared to traditional MZI-based optical neural networks, it does not suffer from the inability to store weights in the absence of power. The non-volatility of this device facilitates the integration of storage and computation in neural networks, providing a viable solution for implementing large-scale photonic neural network architectures.

[0053] The present invention can achieve different power distribution ratios by coding and controlling the non-volatile phase change material Sb2Se3 array on the device, with the advantages of high efficiency, high ratio accuracy and low cost. Compared with traditional power beam splitting devices, the device structure size is several orders of magnitude smaller, only 4.92×2.34μm 2 , and the structure is simple and easy to implement. It is only necessary to fill part of the Sb2Se3 array in the upper half of the coupling region, and based on the programmable characteristics of the DBS algorithm to control the state of the units in the optimized area, 128 different levels of control of the splitting ratio can be achieved in the wavelength range of 1530nm-1570nm. Taking advantage of the programmable control of this device, it is also possible to build an optical convolutional neural network through this device and realize deep learning tasks of handwritten digits, greatly reducing the computational cost of matrix multiplication of the convolution layer.

[0054] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0055] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A power beam splitter based on phase change material, characterized in that: It includes an input waveguide, a coupling region and two output waveguides connected in sequence, wherein the coupling region includes a phase change material array region and an optimization region, and the two output waveguides are respectively connected to the phase change material array region and the optimization region; The optimization area is divided into a plurality of cells, each cell has two states, and the state of each cell is determined according to a direct binary search algorithm; The phase change material array region includes a plurality of phase change blocks composed of phase change material and arranged at intervals and embedded in the coupling region; The phase change block is in the shape of a long strip; The arrangement direction of the phase change blocks is perpendicular to the transmission direction of light.

2. The power beam splitter based on phase change material as claimed in claim 1, characterized in that the phase change The material is Sb2Se3.

3. The power beam splitter based on phase change material according to any one of claims 1 to 2, characterized in that: The status of each unit is determined as follows: Initialize the state of each unit; then change the state of a unit and compare the transmittance of light before and after the change. If the transmittance after the state change is greater than the transmittance before the state change, retain the state of the changed unit. Repeat the above method for all units until they become stable and obtain the state of each unit.

4. The power beam splitter based on phase change material according to claim 3, wherein: The transmittance is the sum of two transmittance differences, wherein one transmittance difference is the absolute value of the difference in transmittance of light passing through the two output waveguides when all phase change materials are crystalline; and the other transmittance difference is the absolute value of the difference in transmittance of light passing through the two output waveguides when all phase change materials are amorphous.

5. The power beam splitter based on phase change material according to any one of claims 1 to 2, characterized in that: The two states of the unit are an etched state and a non-etched state. The etched state is where the center of the unit is etched into an air circular hole.

6. A method for designing a power beam splitter based on phase change material according to any one of claims 1 to 5, characterized in that: The optimized area is obtained by optimizing the following steps: initializing the state of each unit; then changing the state of a certain unit, comparing the transmittance of light before and after the change, and if the transmittance after the state change is greater than the transmittance before the state change, retaining the state of the unit after the change, and editing all units in the above manner until they tend to be stable, thereby obtaining the state of each unit.

7. An image recognizer, characterized in that: The power beam splitter comprising the phase change material based power beam splitter according to any one of claims 1 to 5.

Images