An optical computing chip for image edge detection

Abstract

The application belongs to the field of optoelectronic device preparation and specifically discloses an optical computing chip for image edge detection, which comprises: a first optical computing module composed of four series non-equal-arm MZMs; a second optical computing module composed of four series non-equal-arm MZMs; and an electrical signal input end of the four non-equal-arm MZMs in the first optical computing module and the four non-equal-arm MZMs in the second optical computing module, which is used for receiving four groups of electrical signals corresponding to the similarity of the adjacent four pixels of the pixel to be detected, so as to perform edge detection on the pixel to be detected; at least one waveguide arm of each non-equal-arm MZM has a waveguide non-volatile control section, and a polarization electrode is arranged on each waveguide non-volatile control section; there is a built-in electric field in the waveguide non-volatile control section; and the polarization electrode is used for controlling the static working point of the corresponding non-equal-arm MZM. According to the application, high-speed pixel electrical signals can be matched in the optical domain under the condition of no static holding power consumption, and full-optical computation of image edge detection is realized.

Description

Technical Field

[0001] This application belongs to the field of optical computing, and more specifically, relates to the field of optoelectronic device manufacturing, and relates to an optical computing chip for image edge detection. Background Technology

[0002] Image edge detection is a fundamental preprocessing step in computer vision, autonomous driving, industrial inspection, medical imaging, and intelligent sensing. Edge detection algorithms such as Sobel, Canny, and SUSAN are publicly available and widely used. Their core typically includes local neighbor pixel comparison, thresholding, and convolution or logical combination. For high-resolution and high-speed image streams, traditional electronic processing methods require frequent data transfer between memory and processor, which is limited by clock frequency, memory bandwidth, and power consumption.

[0003] Existing research has attempted to implement edge detection or image convolution operations using optical systems or photonic integrated chips, such as free-space optical filtering, micro-ring arrays, interferometer grids, or silicon photonic modulator arrays to achieve some analog computation functions. However, existing solutions typically suffer from problems such as insufficient high-speed electro-optic modulation capabilities, the need for continuous thermal tuning of the operating point, difficulty in maintaining phase and threshold drift over long periods, increased power consumption with increasing array size, and insufficient accuracy in optical analog computation.

[0004] Lithium tantalate (LTOI) thin-film insulator platforms possess strong linear electro-optic effects, low optical transmission loss, and material properties suitable for high-frequency traveling-wave electrode integration, making them suitable for high-speed electro-optic modulation. On the other hand, the LTOI bulk material exhibits ferroelectric domain-flipping characteristics. By applying a polarization electric field or polarization pulse near the optical waveguide that satisfies the ferroelectric domain-flipping condition, a non-volatile effective refractive index change can be formed in the local waveguide region, thereby maintaining the phase, operating point, or coupling state after the write power is removed.

[0005] Therefore, a novel LTOI image edge detection optical chip architecture is needed to map the disclosed edge detection criteria to on-chip persistent optical operating points and utilize LTOI high-speed electro-optic modulators to receive pixel electrical signals, enabling local matching judgments in edge detection to be performed in parallel in the optical domain. The innovation of this application lies not in proposing a new edge detection algorithm, but in proposing a chip structure, writing method, and system implementation for non-volatile storage of edge detection criteria in an LTOI dual-wavelength cascaded high-speed modulator array. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this application is to provide an optical computing chip for image edge detection, aiming to solve the problems that electronic and optical edge detection schemes are limited by insufficient computing speed and excessive chip power consumption, respectively, and lack a non-volatile hardware implementation scheme that can be directly used for high-speed optical modulator arrays.

[0007] To achieve the above objectives, this application provides an optical computing chip for image edge detection, comprising: The first optical computing module includes: a first non-equilateral Mach-Zehnder modulator MZM, a second non-equilateral MZM, a third non-equilateral MZM, and a fourth non-equilateral MZM connected in series with an optical signal output terminal and an optical signal input terminal. The second optical computing module includes: a fifth non-equidistant arm unit MZM, a sixth non-equidistant arm unit MZM, a seventh non-equidistant arm unit MZM, and an eighth non-equidistant arm unit MZM connected in series with the optical signal output terminal and the optical signal input terminal. The electrical signal input terminals of the four non-equilateral MZMs in the first optical computing module and the electrical signal input terminals of the four non-equilateral MZMs in the second optical computing module are used to synchronously receive four sets of electrical signals corresponding to the similarity of four adjacent pixels of the pixel to be detected in the image, so as to perform edge detection on the pixel to be detected; the adjacent pixels are either horizontally adjacent or vertically adjacent. In this design, at least one waveguide arm of each non-equilateral MZM has a waveguide non-volatile control segment, and each waveguide non-volatile control segment is provided with a polarization electrode. The material of the waveguide non-volatile control segment is lithium tantalate crystal, and a built-in electric field exists within the lithium tantalate crystal. The polarization electrode is used to be subjected to a polarization electric field exceeding a preset electric field strength threshold, causing at least some ferroelectric domains within the lithium tantalate crystal of the corresponding waveguide non-volatile control segment to flip, thereby causing a non-volatile change in the electro-optic coefficient of the lithium tantalate crystal. Combined with the built-in electric field, the refractive index of the corresponding waveguide non-volatile control segment is non-volatilely controlled to achieve the control of the static operating point of the corresponding non-equilateral MZM. When the optical computing chip is working, the polarization electric field is removed.

[0008] In one possible implementation, the optical computing chip operates as follows: The optical signal input terminals of the first non-equidistant MZM and the fifth non-equidistant MZM are used to synchronously receive a first wavelength optical signal and a second wavelength optical signal with a preset power ratio. The electrical signal input terminals of the first non-equilateral MZM to the fourth non-equilateral MZM are used to receive four sets of electrical signals corresponding to four horizontally adjacent pixels of the pixel to be detected in the first time window, and to receive four sets of electrical signals corresponding to four vertically adjacent pixels of the pixel to be detected in the second time window. The electrical signal input terminals of the fifth non-equilateral MZM to the eighth non-equilateral MZM are used to receive four sets of electrical signals corresponding to four horizontally adjacent pixels of the pixel to be detected in the first time window, and to receive four sets of electrical signals corresponding to four vertically adjacent pixels of the pixel to be detected in the second time window. The optical signal output terminal of the fourth non-equidistant MZM outputs a first optical signal in the first time window and a second optical signal in the second time window. The optical signal output terminal of the eighth non-equidistant MZM outputs a third optical signal in the first time window and a fourth optical signal in the second time window. The power of the first wavelength light signal and the second wavelength light signal in the first to fourth light signals reflects whether the pixel to be detected is located at the edge.

[0009] In one possible implementation, the aforementioned optical computing chip further includes: The non-equilateral Mach-Zehnder interferometer (MZI) has two input terminals for receiving optical signals of the first wavelength and the second wavelength, respectively, and two output terminals for connecting to the optical signal input terminals of the first non-equilateral MZM and the fifth non-equilateral MZM, respectively. The interval between the first wavelength and the second wavelength is half of the free spectral range of the non-equilateral MZI, and they operate at the positive and negative quadrature bias points of the non-equilateral MZI, respectively.

[0010] In one possible implementation, the difference in waveguide arm length between the first non-equilateral MZM to the eighth non-equilateral MZM determines the free spectral range of the corresponding non-equilateral MZM. The free spectral ranges of the first to eighth non-equilateral MZMs are designed in combination with the first and second wavelengths, so that after the static operating points of the first to eighth non-equilateral MZMs are adjusted, the optical signal of the first wavelength operates at the negative quadrature bias point, the negative quadrature bias point, the power peak point, the power peak point, the positive quadrature bias point, the positive quadrature bias point, the positive quadrature bias point, and the negative quadrature bias point in the first to eighth non-equilateral MZMs, respectively. The optical signal of the second wavelength operates at the power peak point, the power peak point, the negative quadrature bias point, the negative quadrature bias point, the negative quadrature bias point, the negative quadrature bias point, the positive quadrature bias point, the positive quadrature bias point, and the positive quadrature bias point in the first to eighth non-equilateral MZMs, respectively.

[0011] In one possible implementation, if the absolute value of the difference between the pixel value of the pixel to be detected and the pixel value of its neighboring pixel is not greater than a preset value, the electrical signal input to the corresponding non-equilateral MZM is the first electrical signal; otherwise, the electrical signal input to the corresponding non-equilateral MZM is the second electrical signal. When the first electrical signal is input to the electrical signal input terminal of any non-equal-arm MZM, the phase difference between the two arms of the corresponding non-equal-arm MZM increases by π / 2. When the second electrical signal is input to the electrical signal input terminal of any non-equilateral MZM, the phase difference between the two arms of the corresponding non-equilateral MZM decreases by π / 2.

[0012] In one possible implementation, the aforementioned optical computing chip further includes: The first detection unit includes a first input terminal and a first output terminal; the first input terminal is connected to the optical signal output terminal of the fourth non-equidistant MZM, and the first output terminal outputs a first level signal. The second detection unit includes a second input terminal and a second output terminal; the second input terminal is connected to the optical signal output terminal of the eighth non-equidistant MZM, and the second output terminal outputs a second level signal. When the first wavelength optical signal or the second wavelength optical signal satisfies the coherent constructive condition in the first to fourth non-equidistant MZM, the output power of the first wavelength optical signal or the second wavelength optical signal in the corresponding first optical signal or the second optical signal is close to the corresponding input power, and the corresponding first level signal is high level; otherwise, it is low level. When the first wavelength optical signal or the second wavelength optical signal satisfies the coherent constructive condition in both the fifth to the eighth non-equilateral MZM, the output power of the first wavelength optical signal or the second wavelength optical signal in the corresponding third or fourth optical signal is close to the corresponding input power, and the corresponding second level signal is high level; otherwise, it is low level.

[0013] In one possible implementation, the pixel to be detected is located at an edge when the first level signal output by the first detection unit in one of the first time window and the second time window is high. When the second level signal output by the second detection unit in both the first and second time windows is high, the pixel to be detected is located at the edge.

[0014] In one possible implementation, the four horizontally adjacent pixels of the pixel to be detected include the two pixels adjacent to the left and the two pixels adjacent to the right. The four vertically adjacent pixels of the pixel to be detected include the two pixels adjacent to the top and the two pixels adjacent to the bottom.

[0015] In one possible implementation, the waveguide non-volatile modulation sections of the first to eighth non-equilateral MZMs have a first refractive index before a polarization electric field is applied, and a second refractive index after the polarization electric field is removed. When the polarization electric field is in the first direction, the second refractive index is greater than the first refractive index; the first direction is opposite to the direction of the ferroelectric domains in the lithium tantalate crystal in the initial state; When the polarization electric field is in the second direction, the second refractive index is less than the first refractive index; the second direction is the same as the direction of the ferroelectric domains in the lithium tantalate crystal in the initial state.

[0016] In one possible implementation, when the polarization electric field is in a first direction, the degree of increase in the refractive index of the corresponding non-volatile control section of the waveguide is positively correlated with the degree of increase in the polarization electric field intensity; when the polarization electric field is in a second direction, the degree of decrease in the refractive index of the corresponding non-volatile control section of the waveguide is positively correlated with the degree of increase in the polarization electric field intensity. The first positive correlation has a first positive correlation coefficient, and the second positive correlation has a second positive correlation coefficient; the first positive correlation coefficient is less than the second positive correlation coefficient.

[0017] In summary, compared with the prior art, the technical solutions conceived in this application have the following main technical advantages: This application provides an optical computing chip for image edge detection. First, the local matching criteria in edge detection are mapped to the optical transmission state of an LTOI cascaded modulator array, which reduces the amount of digital convolution computation during operation. Second, the strong electro-optic effect and traveling wave electrode structure of LTOI support high-speed pixel electrical signal input in the tens to hundreds of GHz range. Third, the non-volatile operating point is maintained by utilizing LTOI ferroelectric domain flipping, eliminating the need for continuous thermal tuning or bias maintenance. Fourth, dual-wavelength and non-equidistant arm modulators enable the same 1×4 physical cascaded array to represent two sets of judgment arrays, improving area efficiency. Fifth, a 2×2 asymmetric Mach-Zehnder interferometer simultaneously achieves dual-wavelength phase mapping and two equal-power outputs, eliminating the need for additional power distribution structures in subsequent stages, and reads out two edge criteria through two photodetectors, improving the clarity and parallelism of criterion readout. Attached Figure Description

[0018] Figure 1 This is a diagram of an optical computing chip architecture for image edge detection provided in an embodiment of this application; Figure 2 These are atomic distribution diagrams in LT unit cells under different states provided in the embodiments of this application; Figure 3 This is a graph showing the change in LT refractive index with respect to LT polarization state and a corresponding schematic diagram of ferroelectric domain states provided in the embodiments of this application. Figure 4 This is a schematic diagram of the improved SUSAN cross template provided in the embodiments of this application; Figure 5 This is a schematic diagram of two sets of 1×4 ±1 arrays corresponding to the four pixels above and below the center pixel and the four pixels to the left and right, provided in the embodiments of this application; Figure 6 This is a schematic diagram of the static operating points of dual wavelengths in eight non-equilateral MZMs provided in the embodiments of this application; Figure 7This is a schematic diagram of the mapping between the dual-wavelength phase response and the edge criterion array in the 2×2 asymmetric Mach-Zehnder interferometer provided in the embodiments of this application; Figure 8 This is the runtime sequence and edge decision logic diagram provided in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0021] Furthermore, throughout this specification, references to "an embodiment"; "an embodiment," "an example," or similar language indicate that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of this application. Therefore, the appearance of the phrase "in one embodiment;" throughout this specification, and similar language, may, but not necessarily, refer to the same embodiment.

[0022] First, the technical terms involved in the embodiments of this application will be introduced.

[0023] (1) Ferroelectric crystals Ferroelectric crystals are a class of crystal materials that exhibit spontaneous polarization, and the direction of this spontaneous polarization can be reversed under the action of an applied electric field. They belong to the category of ferroelectric materials.

[0024] (2) Ferroelectric domains Ferroelectric domains are a type of microstructure found in ferroelectric materials. Simply put, they are small regions within a ferroelectric material that are divided by different spontaneous polarization directions.

[0025] (3) Ferroelectric domain flipping Ferroelectric domain reversal refers to the process by which the spontaneous polarization direction of a ferroelectric domain in a ferroelectric material reverses under external excitation.

[0026] (4) Defect dipole A ferroelectric crystal is a microscopic electric dipole formed inside due to the presence of defects (such as impurity ions and vacancies). Simply put, it is a pair of defects with opposite charges in a ferroelectric crystal, which form a defect dipole because the centers of positive and negative charges do not coincide.

[0027] (5) Built-in electric field In the unit cell of a ferroelectric crystal, when a large number of adjacent dipoles are neatly arranged, positive and negative bound charges are induced on the two opposite surfaces of the crystal, respectively. These macroscopic bound charges, like the two plates of a parallel-plate capacitor, generate an electric field inside the crystal that points from the positive charge end to the negative charge end, i.e., a built-in electric field.

[0028] (6) Lithium tantalate of the same composition Lithium tantalate with the same composition refers to a lithium-tantalum molar ratio that is not the theoretically perfect 1:1 stoichiometric ratio, but rather a golden ratio determined experimentally. For example, the stoichiometric ratio could be... At this ratio, the crystals grown from the melt have a composition completely consistent with the melt itself, resulting in high-quality, large-sized crystals with uniform composition and no striations, making them ideal for industrial mass production. However, the aforementioned non-ideal ratio leads to numerous defects such as lithium vacancies within the crystal.

[0029] The embodiments of this application are described below with reference to the accompanying drawings.

[0030] This application aims to address at least one problem in existing electronic and photonic edge detection schemes: electronic image processing relies on digital convolution or a large number of logic operations, and data transfer and clock frequency limitations result in high latency and power consumption; existing optical edge detection chips often struggle to simultaneously achieve high-speed modulation, zero static power consumption to stabilize the modulator operating point and phase-shift waveguide; silicon or silicon nitride platforms lack strong Pockels effect and often require thermal tuning or additional materials; and the published edge detection criteria lack a non-volatile hardware implementation that can be directly mapped to an LTOI high-speed optical modulator array.

[0031] Figure 1 This is an architectural diagram of an optical computing chip for image edge detection provided in an embodiment of this application, as shown below. Figure 1 As shown, it includes: The first optical computing module includes: a first non-equilateral Mach-Zehnder modulator MZM1, a second non-equilateral Mach-Zehnder modulator MZM2, a third non-equilateral Mach-Zehnder modulator MZM3, and a fourth non-equilateral Mach-Zehnder modulator MZM4, which are connected in series with optical signal output and optical signal input terminals. The second optical computing module includes: a fifth non-equilateral arm unit MZM5, a sixth non-equilateral arm unit MZM5, a seventh non-equilateral arm unit MZM7, and an eighth non-equilateral arm unit MZM8, which are connected in series with optical signal output and optical signal input terminals. The electrical signal input terminals of the four non-equilateral MZMs in the first optical computing module and the electrical signal input terminals of the four non-equilateral MZMs in the second optical computing module are used to synchronously receive four sets of electrical signals corresponding to the similarity of four adjacent pixels of the pixel to be detected in the image, so as to perform edge detection on the pixel to be detected; the adjacent pixels are either horizontally adjacent or vertically adjacent. In this design, at least one waveguide arm of each non-equilateral MZM has a waveguide non-volatile control segment, and each waveguide non-volatile control segment is provided with a polarization electrode. The material of the waveguide non-volatile control segment is lithium tantalate crystal, and a built-in electric field exists within the lithium tantalate crystal. The polarization electrode is used to be subjected to a polarization electric field exceeding a preset electric field strength threshold, causing at least some ferroelectric domains within the lithium tantalate crystal of the corresponding waveguide non-volatile control segment to flip, thereby causing a non-volatile change in the electro-optic coefficient of the lithium tantalate crystal. Combined with the built-in electric field, the refractive index of the corresponding waveguide non-volatile control segment is non-volatilely controlled to achieve the control of the static operating point of the corresponding non-equilateral MZM. When the optical computing chip is working, the polarization electric field is removed.

[0032] In one embodiment, the waveguide non-volatile modulation sections of the first to eighth non-equilateral MZMs have a first refractive index before a polarization electric field is applied, and a second refractive index after the polarization electric field is removed. When the polarization electric field is in the first direction, the second refractive index is greater than the first refractive index; the first direction is opposite to the direction of the ferroelectric domains in the lithium tantalate crystal in the initial state; When the polarization electric field is in the second direction, the second refractive index is less than the first refractive index; the second direction is the same as the direction of the ferroelectric domains in the lithium tantalate crystal in the initial state.

[0033] In one embodiment, such as Figure 1 As shown, the aforementioned optical computing chip also includes: It is understandable that if the ferroelectric crystal is a perfect crystal without any defects, it will not possess a built-in electric field. In this case, even if the polarization electric field can cause the ferroelectric domains within the ferroelectric crystal to flip, the refractive index of the waveguide modulation section will not change.

[0034] It is understandable that the aforementioned ferroelectric crystals can be LTOI, LNOI, or other ferroelectric crystal materials that can be used in optical waveguides or integrated photonic devices.

[0035] For example, taking lithium tantalate as a ferroelectric crystal, when there are no defects in the lithium tantalate film, its interior only contains the polarization intensity P of the ferroelectric domains. S There is no defective dipole polarization intensity P D ,like Figure 2As shown in (i), when defects exist within the lithium tantalate film, the aforementioned lithium deficiency is mainly compensated by lithium vacancies and tantalum antisite defects. These defects form defect dipole clusters with polarization intensity P. D And in the thermodynamically stable state, it interacts with the ferroelectric polarization P S Same direction Figure 2 (ii) When ferroelectric domains are flipped by an applied electric field at room temperature, spontaneous polarization can be flipped with the electric field, while lithium vacancies in defect dipoles will not rearrange on the same time scale due to insufficient mobility, such as... Figure 2 (iii) Therefore, the programming region enters a metastable state, with the defect dipoles and spontaneous polarization directions opposite. After the external electric field is removed, this built-in field configuration is maintained, and a non-volatile change in local refractive index occurs through the electro-optic effect. During high-temperature baking, lithium vacancies gain sufficient mobility to rearrange themselves. After annealing and cooling, the defect dipoles are rearranged to align with the ferroelectric domains, at which point the non-volatile refractive index change disappears, as shown in the image. Figure 2 (iv)

[0036] Figure 3 The diagram illustrates the gradual change in refractive index as the degree of ferroelectric domain flipping increases. Ferroelectric domains are represented as different polygonal regions, with solid arrows indicating the domain's orientation within each region. During the gradual flipping of ferroelectric domains, the defect dipoles, indicated by dashed arrows, do not flip along with the ferroelectric domains. This phenomenon is also reflected in the hysteresis loops of LT of the same composition. The difference between the positive and negative coercive fields of LT of the same composition is precisely due to the built-in electric field caused by the orientation difference between the defect dipoles and the ferroelectric domains.

[0037] The nonvolatile refractive index variation mechanism can be described by the built-in field electro-optic modulation model of the TE mode in the x-cut LTOI waveguide. In the initial state, the defect dipole polarization is consistent with the ferroelectric polarization, and the unusual refractive index can be written as the intrinsic unusual optical refractive index superimposed on the refractive index resulting from the Pockels effect of the defect dipole's built-in electric field. : in, For intrinsic unusual refractive index, The electro-optic coefficient, This provides an effective built-in electric field for the defective dipole. The ferroelectric domains in the polarization region have flipped (the percentage of flipped ferroelectric domains is...). The percentage of unflipped units (1- The domain orientation and the defect dipole-induced built-in electric field are reversed, resulting in an inverse sign of the electro-optic coefficient. The corresponding refractive index... for: Change in refractive index for: .

[0038] It should be noted that the shape and length of the polarization electrode are matched with the shape and length of the waveguide modulation section.

[0039] In one embodiment, the optical waveguide may be a strip waveguide, a ridge waveguide, or a planar waveguide, etc.

[0040] Furthermore, it can be understood that the aforementioned polarization electrodes and waveguide modulation sections can be arranged overlappingly, adjacent or separated on the same plane, or partially overlapping and partially adjacent on the same plane, etc. Existing volatile modulation (electro-optic modulation) settings can be referenced to achieve the best refractive index non-volatile modulation effect.

[0041] Typically, conventional electro-optic modulation, also known as volatile modulation, has a relatively small modulation electric field; while non-volatile modulation has a relatively large modulation electric field, which may be 15-20 times larger than that of volatile modulation.

[0042] For example, in the initial state, the ferroelectric domains within the waveguide modulation segment maintain a consistent orientation.

[0043] For example, when the polarization electric field is in a first direction, the degree of increase in the refractive index of the waveguide modulation segment is positively correlated with the degree of increase in the polarization electric field intensity and the built-in electric field intensity; when the polarization electric field is in a second direction, the degree of decrease in the refractive index of the waveguide modulation segment is positively correlated with the degree of increase in the polarization electric field intensity and the built-in electric field intensity. Under the same built-in electric field strength, the first positive correlation has a first positive correlation coefficient, and the second positive correlation has a second positive correlation coefficient; the first positive correlation coefficient is less than the second positive correlation coefficient.

[0044] It is understandable that the first positive correlation coefficient being less than the second positive correlation coefficient indicates that the polarization electric field along the initial direction of the ferroelectric domains is more likely to adjust the refractive index. That is, to increase or decrease the same refractive index, the polarization electric field strength value required to be adjusted in the first direction is greater than the electric field strength value required to be adjusted in the second direction.

[0045] For example, the polarization electric field is applied for a duration greater than 10 ms. It is understood that after the polarization electric field is applied for a certain period of time, the ferroelectric domains in the ferroelectric crystal begin to flip. As the polarization time increases, the flipping ratio increases until, after a period of time, the amount of ferroelectric domain flipping no longer changes, at which point the refractive index remains essentially unchanged. After this, when the polarization electric field is removed, the refractive index also remains essentially unchanged.

[0046] For example, when the optical waveguide is a single-ridge waveguide, along the transverse direction of the single-ridge waveguide, the waveguide modulation section includes a modulation region, an unmodulated region, and a groove. The modulation region is the area in the middle of the single-ridge waveguide and includes the protruding structure of the single-ridge waveguide. The unmodulated region is the area on both sides of the single-ridge waveguide. The groove is disposed between the modulation region and the unmodulated region to isolate the modulation region and the unmodulated region. The polarization electrode is disposed along both sides of the modulation region and is disposed above a portion of the groove and / or a portion of the modulation region.

[0047] Experimental studies have shown that when the ferroelectric crystal is lithium tantalate of the same composition, the preset electric field strength threshold is greater than or equal to 16.6 * 10⁻⁶. 6 V / m. Therefore, the intensity range of the polarization electric field at this time can be 16.6*10 V / m. 6 V / m ~23.3*10 6 V / m. The range of refractive index change can be 0~0.55*10 -3 .

[0048] See Figure 1 As shown, the non-equilateral Mach-Zehnder interferometer (MZI) is specifically a 2×2 asymmetric Mach-Zehnder interferometer. Its two input terminals are used to receive optical signals of the first wavelength and the second wavelength, respectively, and its two output terminals are connected to the optical signal input terminals of the first non-equilateral Mach-Zehnder interferometer and the fifth non-equilateral Mach-Zehnder interferometer, respectively. The interval between the first wavelength and the second wavelength is half of the free spectral range of the non-equilateral MZI, and they operate at the positive and negative quadrature bias points of the non-equilateral MZI, respectively.

[0049] Specifically, dual-wavelength optical signals λ1 and λ2 are input to a 2×2 asymmetric Mach-Zehnder interferometer. The wavelength interval between λ1 and λ2 is half of the free spectral range of the 2×2 asymmetric Mach-Zehnder interferometer, and they operate at the positive and negative quadrature bias points of the non-equilateral MZI, respectively, so that the two wavelengths form a predetermined equal power distribution at the two output terminals.

[0050] In one embodiment, the optical computing chip operates as follows: The optical signal input terminals of the first non-equidistant MZM and the fifth non-equidistant MZM are used to synchronously receive a first wavelength optical signal and a second wavelength optical signal with a preset power ratio. The electrical signal input terminals of the first non-equilateral MZM to the fourth non-equilateral MZM (i.e. Figure 1The high-speed radio frequency electrode (shown in yellow) is used to receive four sets of electrical signals corresponding to four horizontally adjacent pixels of the pixel to be detected in the first time window, and to receive four sets of electrical signals corresponding to four vertically adjacent pixels of the pixel to be detected in the second time window. The electrical signal input terminals of the fifth non-equilateral arm MZM to the eighth non-equilateral arm MZM (i.e. Figure 1 The high-speed radio frequency electrode (shown in yellow) is used to receive four sets of electrical signals corresponding to four horizontally adjacent pixels of the pixel to be detected in the first time window, and to receive four sets of electrical signals corresponding to four vertically adjacent pixels of the pixel to be detected in the second time window. The optical signal output terminal of the fourth non-equidistant MZM outputs a first optical signal in the first time window and a second optical signal in the second time window. The optical signal output terminal of the eighth non-equidistant MZM outputs a third optical signal in the first time window and a fourth optical signal in the second time window. The power of the first wavelength light signal and the second wavelength light signal in the first to fourth light signals reflects whether the pixel to be detected is located at the edge.

[0051] In one embodiment, such as Figure 1 As shown, the aforementioned optical computing chip also includes: The first detection unit includes a first input terminal and a first output terminal; the first input terminal is connected to the optical signal output terminal of the fourth non-equidistant MZM, and the first output terminal outputs a first level signal. The second detection unit includes a second input terminal and a second output terminal; the second input terminal is connected to the optical signal output terminal of the eighth non-equidistant MZM, and the second output terminal outputs a second level signal. When the first wavelength optical signal or the second wavelength optical signal satisfies the coherent constructive condition in the first to fourth non-equidistant MZM, the output power of the first wavelength optical signal or the second wavelength optical signal in the corresponding first optical signal or the second optical signal is close to the corresponding input power, and the corresponding first level signal is high level; otherwise, it is low level. When the first wavelength optical signal or the second wavelength optical signal satisfies the coherent constructive condition in both the fifth to the eighth non-equilateral MZM, the output power of the first wavelength optical signal or the second wavelength optical signal in the corresponding third or fourth optical signal is close to the corresponding input power, and the corresponding second level signal is high level; otherwise, it is low level.

[0052] In one embodiment, such as Figure 4 As shown, the four horizontally adjacent pixels and the four vertically adjacent pixels of the pixel to be detected can come from the improved SUSAN cross template, specifically the pixel to be detected P. nThe four horizontally adjacent pixels include the two left adjacent pixels a1 and a2 and the two right adjacent pixels a3 and a4; The pixel to be detected P n The four vertically adjacent pixels include the two pixels above, a5 and a6, and the two pixels below, a7 and a8.

[0053] Specifically, the electrical signals corresponding to the similarity between adjacent pixels of the pixel to be detected. I m for: Among them, P n This represents the pixel data to be detected, such as the grayscale value; a m This represents the pixel data of the m-th adjacent pixel, such as its grayscale value. `th` represents the preset threshold. Referring to the formula above, it can be seen that when the absolute value of the pixel difference is less than or equal to the threshold `th`, I m The value is 1 when the absolute value of the pixel difference is greater than the threshold th. I m -1, see Figure 5 As shown.

[0054] Therefore, each central pixel P to be detected n Each edge has two sets of 1×4 ±1 arrays, Xn and Yn, consisting of four pixels vertically and four pixels horizontally. The corresponding two edge criteria EF1 and EF2 are: The first edge criterion EF1 is: when either Xn or Yn is -1, -1, X, X or X, X, -1, -1, the center pixel P n Located at the edge; X represents 1 or -1; see also Figure 5 The horizontal or vertical edge in the middle.

[0055] The second edge criterion EF2 is: when arrays Xn and Yn correspond to combinations of -1, 1, 1, 1 and 1, 1, 1, -1 (the order of the combinations is not restricted), or both are -1, 1, 1, 1, or both are 1, 1, 1, -1, the center pixel P is... n Located at the edge, see Figure 5 The hypotenuse in the array; specifically, the array Xn mentioned above refers to... Figure 2 The electrical signals corresponding to a1, a2, a3, and a4 in the diagram. I m The value, Yn, refers to Figure 2 The electrical signals corresponding to a5, a6, a7, and a8 in the diagram. I m value; Figure 2The order of a1, a2, a3, a4, a5, a6, a7, and a8 is not unique; you can choose from left to right, right to left, top to bottom, or bottom to top. Figure 2 This is just one example.

[0056] In one embodiment, the difference in waveguide arm length between the first non-equilateral MZM to the eighth non-equilateral MZM determines the free spectral range of the corresponding non-equilateral MZM. See Figure 6 As shown, the free spectral ranges of the first to eighth non-equilateral MZMs are designed in combination with a first wavelength and a second wavelength, so that the static operating points of the first to eighth non-equilateral MZMs are modulated. The first and second wavelengths can be 1535 nm and 1565 nm, respectively, and the optical signal at the first wavelength operates at the negative-positive quadrature bias point Q of the first to eighth non-equilateral MZMs. - Negative and positive bias points Q - Peak power point P, orthogonal bias point Q + Orthogonal bias point Q + Orthogonal bias point Q + and the negative and positive bias point Q - The second wavelength optical signal operates at the power peak point P1, power peak point P2, and negative quadrature bias point Q1 in the first to eighth non-equilateral MZMs, respectively. - Negative and positive bias points Q - Negative and positive bias points Q - Orthogonal bias point Q + Orthogonal bias point Q + and the orthogonal bias point Q + .

[0057] In one embodiment, if the absolute value of the difference between the pixel value of the pixel to be detected and the pixel value of its neighboring pixel is not greater than a preset value, the electrical signal input to the corresponding non-equilateral MZM is the first electrical signal; otherwise, the electrical signal input to the corresponding non-equilateral MZM is the second electrical signal. It should be noted that, as can be seen from the above, the first electrical signal is 1 and the second electrical signal is -1. The four sets of electrical signals corresponding to the four adjacent pixels of each pixel on each image are converted into high-frequency radio frequency signals and input to the electrical signal input terminals corresponding to MZM1-MZM8.

[0058] Specifically, when the first electrical signal is input to the electrical signal input terminal of any non-equidistant MZM, the phase difference between the two arms of the corresponding non-equidistant MZM increases by π / 2. When the second electrical signal is input to the electrical signal input terminal of any non-equilateral MZM, the phase difference between the two arms of the corresponding non-equilateral MZM decreases by π / 2.

[0059] It is understandable that when a certain wavelength of optical signal is at the negative quadrature bias point Q of the static operating point of a non-equilateral MZM, - If the phase difference between the two arms of a non-equilateral MZM decreases by π / 2, the power of the output optical signal of that wavelength will decrease to near zero. If the phase difference between the two arms of a non-equilateral MZM increases by π / 2, the power of the output optical signal of that wavelength will double.

[0060] When a certain wavelength optical signal has a power peak point P at the static operating point of a non-equilateral MZM, if the phase difference between the two arms of the non-equilateral MZM decreases by π / 2, the power of the output optical signal of that wavelength will be reduced to half. If the phase difference between the two arms of the non-equilateral MZM increases by π / 2, the power of the output optical signal of that wavelength will be reduced to half.

[0061] When a certain wavelength optical signal is at the quiescent operating point of a non-equilateral MZM, the quadrature bias point Q is used. + If the phase difference between the two arms of a non-equilateral MZM decreases by π / 2, the power of the output optical signal of that wavelength will double. If the phase difference between the two arms of a non-equilateral MZM increases by π / 2, the power of the output optical signal of that wavelength will decrease to near zero.

[0062] In one embodiment, referring to the two edge criteria EF1 and EF2 mentioned above, when the first level signal output by the first detection unit in one of the first time window and the second time window is high, the pixel to be detected is located at the edge; When the second level signal output by the second detection unit in both the first and second time windows is high, the pixel to be detected is located at the edge.

[0063] In a more specific embodiment, taking lithium tantalate thin film as the material of the non-volatile modulation section of the waveguide in a non-equilateral MZM as an example, this embodiment discloses a high-speed image edge detection optical chip and its implementation method based on a lithium tantalate on insulator (LTOI) non-volatile dual-wavelength cascaded modulator array. The optical chip includes a dual-wavelength optical input unit, a 2×2 asymmetric Mach-Zehnder interferometer input distribution unit, two cascaded 1×4 non-equilateral LTOI Mach-Zehnder modulator arrays, two photodetector units, and a non-volatile refractive index modulation region for setting the modulator operating point. The arm length difference of the 2×2 asymmetric Mach-Zehnder interferometer determines the free spectral range. The interval between the first wavelength λ1 and the second wavelength λ2 is half of the free spectral range of the 2×2 asymmetric Mach-Zehnder interferometer, ensuring that the two wavelengths form a predetermined equal power distribution at both output ends.

[0064] It should be noted that the first detection unit and the second detection unit mentioned above can be a first photodetector and a second photodetector.

[0065] The two outputs of the aforementioned 2×2 asymmetric Mach-Zehnder interferometer are connected to the first and second optical computing modules, respectively. The output of the first optical computing module is read by the first photodetector to obtain the first edge criterion, and the output of the second optical computing module is read by the second photodetector to obtain the second edge criterion. Each 1×4 cascaded modulator is used to represent one edge criterion. The four non-equilateral modulators in each channel store two sets of 1×4 judgment arrays of the criterion at wavelengths λ1 and λ2, respectively, through LTOI ferroelectric domain flipping. During operation, the upper and lower four pixel groups of the pixel to be detected are converted into four high-speed radio frequency electrical signals in the first time window and loaded onto the traveling wave electrodes of the modulator. The left and right four pixel groups are converted into four high-speed radio frequency electrical signals in the second time window and loaded onto the same set of traveling wave electrodes of the modulator.

[0066] The first photodetector outputs a first output level and a second output level in sequence, representing the matching results of the upper and lower four-pixel groups and the left and right four-pixel groups for the first edge criterion, respectively; when at least one of the two output levels of the first photodetector is high, the center pixel passes the first edge criterion.

[0067] The second photodetector sequentially outputs a first output level and a second output level, representing the matching results of the top and bottom four-pixel groups and the left and right four-pixel groups to the second edge criterion, respectively. When both output levels of the second photodetector are high, the center pixel passes the second edge criterion. When the center pixel passes either the first or second edge criterion, the back-end logic classifies it as an edge pixel. The non-volatile refractive index control region induces at least some ferroelectric domains in the LTOI to flip through a programmed electric field or programmed electrical pulse, and maintains the modulator operating point after the write power is removed by utilizing defect dipoles, built-in electric fields, and electro-optic effects.

[0068] Understandably, this chip does not advocate disclosing the novelty of the edge detection algorithm itself, but rather provides an on-chip hardware implementation that maps the edge detection criteria to the LTOI high-speed optical modulator array and achieves zero static hold power consumption.

[0069] Since each modulator is a non-equilateral structure, the same modulator has different static phases at λ1 and λ2. Therefore, the elements at corresponding positions in two sets of judgment arrays can be stored at two wavelengths using the same physical modulator.

[0070] Each LTOI Mach-Zehnder modulator has a high-speed RF modulation electrode (i.e., the electrical signal input) and a non-volatile refractive index control region (i.e., the static operating point control region). During operation, the high-speed RF modulation electrode receives electrical signals derived from pixel grayscale relationships, such as whether the grayscale difference between the center pixel and neighboring pixels exceeds a threshold. Before operation, the non-volatile refractive index control region is written with a polarization electric field or polarization pulse to cause at least partial flipping of local LTOI ferroelectric domains, and after the writing power is removed, it maintains the modulator at two wavelengths of light transmission, off-state, or arbitrary operating point.

[0071] In one implementation, the neighborhood of the pixel to be detected is divided into a top-bottom four-pixel group and a left-right four-pixel group. The system converts the top-bottom four-pixel group or the left-right four-pixel group into four high-speed pixel comparison electrical signals in different time-division windows and loads them onto a two-channel cascaded modulator array. If all four cascaded modulators in a certain channel satisfy the pre-written judgment array, that channel outputs high light intensity at the corresponding wavelength; if at least one modulator does not satisfy the judgment array, the light intensity of that channel is significantly attenuated.

[0072] The output of the first optical computing module is read by the first photodetector, and the output of the second optical computing module is read by the second photodetector. The two outputs are not combined before detection. If the first photodetector provides a high-level matching result when inputting to either the upper / lower four-pixel group or the left / right four-pixel group, the center pixel is determined to be an edge. If the second photodetector provides a high-level matching result when inputting to both the upper / lower four-pixel group and the left / right four-pixel group, the center pixel is determined to be an edge. This determination logic can be implemented by an external control circuit, an on-chip control circuit, or a back-end processor.

[0073] Figure 7 In the left-middle figure, the blue curve represents the upper arm output optical power spectrum of the asymmetric MZI, and the red curve represents the lower arm output optical power spectrum. The 2×2 asymmetric Mach-Zehnder interferometer generates a periodic wavelength response through the arm length difference, with λ1 and λ2 separated by half a free spectral range, corresponding to different static operating points of each non-equilateral modulator at two wavelengths. Each 1×4 cascaded modulator stores an edge criterion, with each modulator corresponding to the same element in two sets of 1×4 decision arrays for the same edge criterion at λ1 and λ2.

[0074] Figure 8This is a schematic diagram illustrating the runtime sequence and decision logic of image edge detection in one embodiment of this application. The upper and lower four pixel groups and the left and right four pixel groups of the center pixel to be detected are respectively converted into four-way pixel comparison electrical signals, and loaded onto the high-speed radio frequency electrodes of two 1×4 series modulators in a first time window and a second time window. The first output level obtained in the first time window corresponds to the judgment result of the upper and lower four pixel groups, and the second output level obtained in the second time window corresponds to the judgment result of the left and right four pixel groups. At least one of the two output levels of the first photodetector is high, indicating that the center pixel passes the first edge criterion; both output levels of the second photodetector are high, indicating that the center pixel passes the second edge criterion; when the center pixel passes the first edge criterion or the second edge criterion, the center pixel is determined to be an edge pixel.

[0075] The optical chip provided in the above embodiments uses an LTOI substrate, which includes a supporting substrate, an insulating layer, and a lithium tantalate thin film. In a representative embodiment, the supporting substrate is silicon, the insulating layer is silicon dioxide, the lithium tantalate thin film is an x-cut lithium tantalate thin film, and the optical waveguide is a TE-mode ridge waveguide. The waveguide can be formed by processes such as electron beam lithography, ultraviolet lithography, hard mask etching, ICP dry etching, and cladding deposition.

[0076] The chip provided in this embodiment operates by receiving two continuous optical signals, λ1 and λ2, from a dual-wavelength light source. These two wavelengths can be located in the C-band, O-band, or other bands suitable for integrated photonic systems. Figure 1 As shown, λ1 and λ2 directly enter the 2×2 asymmetric MZI input distribution unit. The difference in arm length of the 2×2 asymmetric MZI determines the free spectral range, and the wavelength interval between λ1 and λ2 is set to half of the free spectral range. Therefore, both wavelengths are in an approximately 3 dB equal power output state at both output terminals of the 2×2 asymmetric MZI. The first output terminal connects to the first optical computing module, and the second output terminal connects to the second optical computing module, eliminating the need for a separate power distribution structure in the subsequent stage.

[0077] The first optical computing module consists of four non-equilateral LTOI MZI modulators connected in series; the second optical computing module also consists of four non-equilateral LTOI MZI modulators connected in series. Each modulator includes an input coupler, two interferometer arms, an output coupler, a high-speed RF modulation electrode, and a non-volatile write electrode. The high-speed RF modulation electrode is preferably a traveling-wave electrode, and high-frequency transmission in the range of 50 GHz to 120 GHz can be achieved through impedance matching, electrode speed matching, and low-loss metal design. In an illustrative experiment or design example, the 3 dB electro-optic bandwidth can exceed 67 GHz; in an optimized design, the signal transmission bandwidth can exceed 100 GHz.

[0078] A non-volatile write electrode is positioned near at least one interference arm. When a programming electric field or programming pulse is applied to the non-volatile write electrode, the ferroelectric domains in the local lithium tantalate film undergo partial or complete flipping. After the programming electric field or programming pulse is removed, due to the change in the relative states between the defect dipole, the built-in electric field, and the spontaneous polarization, this region maintains different effective refractive indices through an electro-optic effect, thereby keeping the static phase or operating point of the MZI in the state after writing.

[0079] Specifically, as shown in Figure 8, this embodiment represents the local similarity criterion in the disclosed edge detection algorithm as two edge criteria, each including two sets of 1×4 judgment arrays. A 2×2 asymmetric MZI input allocation unit provides two equal-power output terminals; the first optical computing module is used for the first edge criterion, and the second optical computing module is used for the second edge criterion. Since each modulator is a non-equilateral MZI, its static phase varies with wavelength. Therefore, the same modulator can represent the j-th element of a judgment array under λ1 and the j-th element of another judgment array of the same edge criterion under λ2.

[0080] The elements in the array can be considered as matches, mismatches, or arbitrary. Taking the grayscale difference between the center pixel and its neighboring pixels as an example, a grayscale difference less than the threshold can be considered a match, and a grayscale difference greater than or equal to the threshold can be considered a mismatch. For "arbitrary" elements, the modulator's static operating point, RF modulation depth, or bypass transmission threshold can be set to ensure that the serial optical path of that stage is not blocked when the input is a match or a mismatch.

[0081] Four modulators cascaded together effectively perform a logical AND matching of a 1×4 judgment array: the output light intensity of this path is higher than the threshold only when all four input pixel comparison signals satisfy the corresponding elements; if any position does not satisfy the threshold, at least one modulator is in a low-transmission state, and the output of the cascaded optical path is attenuated. λ1 and λ2 can pass through the same cascaded modulator simultaneously or in a set timing sequence, and each corresponds to two sets of judgment arrays for the same edge criterion, thereby expanding the criterion representation capability without increasing the number of modulators.

[0082] In one specific embodiment, for the image to be detected, the back-end processing unit first selects a center pixel and obtains its upper and lower four-pixel groups and left and right four-pixel groups. These four-pixel groups can come from a modified SUSAN crosshair template or from the local neighborhood required by other publicly available edge detection criteria. The processing unit converts the grayscale relationship between each neighboring pixel and the center pixel into a pixel comparison bit; these four pixel comparison bits constitute a 1×4 runtime input.

[0083] In the first time window, the system loads four high-speed electrical signals corresponding to the upper and lower four-pixel groups onto the high-speed RF modulation electrodes of the two optical computing modules, and reads the first output level of each of the first and second photodetectors. The first output level represents the judgment result of the upper and lower four-pixel groups. In the second time window, the system loads four high-speed electrical signals corresponding to the left and right four-pixel groups onto the same set of high-speed RF modulation electrodes, and again reads the second output level of each of the first and second photodetectors. The second output level represents the judgment result of the left and right four-pixel groups. The first photodetector corresponds to the first edge criterion, and the second photodetector corresponds to the second edge criterion; the two judgment results are not combined before detection.

[0084] In a preferred determination method, if at least one of the first and second output levels of the first photodetector is high, the center pixel passes the first edge criterion and is determined to be an edge pixel; if both the first and second output levels of the second photodetector are high, the center pixel passes the second edge criterion and is determined to be an edge pixel. If none of the above conditions are met, the center pixel is determined to be a non-edge pixel or a weak response pixel.

[0085] The above process can be executed continuously on multiple pixels in an image. Since the pixel comparison electrical signal is input through high-speed traveling-wave electrodes, the modulator array can perform optical domain matching on the continuous pixel stream at high frequency. The light intensity output from the two photodetectors is converted by an ADC to form a digital edge response, and the back-end processing unit can generate an edge image based on thresholds, criterion combinations, morphological rules, or display requirements.

[0086] In a more specific embodiment, before chip operation, the target transmission state of each optical computing module at two wavelengths is first determined based on the target edge criterion. Then, target modulators are selected one by one, and programming electrical pulses are applied through non-volatile write electrodes. The amplitude, width, number, polarity, duty cycle, and spacing of the programming electrical pulses can be determined based on the LTOI film thickness, electrode spacing, target phase change, and calibrated write curves.

[0087] After writing, the transmission intensity of the target modulator or the entire cascaded optical computing module is measured at the first and second wavelengths. If the transmission intensity does not meet the requirements of the target judgment array, one or more programming pulses are applied and the measurement is repeated. This measurement, writing, and retesting process can be executed iteratively until the target modulator corresponds to the predetermined on, off, or arbitrary state at the two wavelengths, respectively.

[0088] When it is necessary to change the edge criteria or recalibrate the chip, programming electrical pulses of opposite polarity or different parameters can be applied to the non-volatile refractive index control region to change the proportion of flipped domains, thereby restoring or changing the effective refractive index. Thus, the chip can support post-factory calibration, post-packaging calibration, algorithm criterion switching, or long-term drift compensation.

[0089] In a more specific embodiment, the edge detection system of this embodiment includes a dual-wavelength laser source, a polarization controller or on-chip polarization management structure, an LTOI edge detection optical chip, a high-speed arbitrary waveform generator or pixel driving circuit, an RF amplifier, a timing control circuit, a first photodetector, a second photodetector, an analog-to-digital converter, and an electronic processing unit. The electronic processing unit can be implemented by a computer, FPGA, ASIC, microcontroller, or on-chip control circuit.

[0090] Before system operation, the electronic processing unit controls the non-volatile write pulses according to the target criteria to complete the configuration of the operating points of each modulator. During system operation, the non-volatile write electrodes can be disconnected or maintain zero output without consuming static hold power; the high-speed RF modulation electrode receives the pixel comparison electrical signal, and the two photodetectors output the matching results of the first edge criterion and the second edge criterion respectively, realizing high-speed edge detection.

[0091] In summary, the LTOI substrate described in the above embodiments is not limited to x-cut, and can also be selected as y-cut, z-cut or other tangential lithium tantalate films depending on the target polarization, electro-optic tensor and process conditions.

[0092] The aforementioned optical waveguides are not limited to ridge waveguides; they can also be strip waveguides, shallow etched waveguides, deep etched waveguides, grooved waveguides, multimode waveguides, or hybrid waveguides.

[0093] The aforementioned non-volatile write electrodes are not limited to lateral electrodes, but can also be top electrodes, bottom electrodes, coplanar electrodes, interdigitated electrodes, local tip electrodes, multi-segment addressing electrodes, or combinations thereof.

[0094] The aforementioned edge criteria are not limited to the modified SUSAN cross template, but can also be derived from other publicly disclosed local similarity, grayscale comparison, directional edge or binary template criteria; as long as they can be represented as one or more 1×4 judgment arrays, they can be mapped to the cascaded modulator array of this application.

[0095] More specifically, the optical computing module described above is not limited to two channels, but can be expanded to three or more channels to represent more edge criteria; the number of modulators per channel is not limited to four, but can also be expanded to N cascaded modulators according to the template size.

[0096] The above readout methods are not limited to parallel readout of two independent PDs. Balanced detection, wave decomposition and multiplexing detection, on-chip monitoring port detection, external detector array or more detection channels can also be used for readout.

[0097] The aforementioned dual-wavelength input can be provided by two independent lasers, a multi-wavelength laser, a frequency comb, a modulation sideband, or an on-chip wavelength selector.

[0098] The LTOI non-volatile dual-wavelength cascaded modulator edge detection optical chip provided in this application can be realized using existing thin-film lithium tantalate photolithography, etching, cladding deposition, metal electrode fabrication, and optoelectronic packaging processes, providing a clear manufacturing path. This chip can store publicly disclosed edge detection criteria in a non-volatile manner within a high-speed optical modulator array and perform optical domain matching of high-speed pixel electrical signals without static power consumption. It is suitable for applications such as machine vision, edge computing, optoelectronic fusion sensing, data center visual preprocessing, unmanned systems, and low-power intelligent terminals.

[0099] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An optical computing chip for image edge detection, characterized in that, include: The first optical computing module includes: a first non-equilateral Mach-Zehnder modulator MZM, a second non-equilateral MZM, a third non-equilateral MZM, and a fourth non-equilateral MZM connected in series with an optical signal output terminal and an optical signal input terminal. The second optical computing module includes: a fifth non-equidistant arm unit MZM, a sixth non-equidistant arm unit MZM, a seventh non-equidistant arm unit MZM, and an eighth non-equidistant arm unit MZM connected in series with the optical signal output terminal and the optical signal input terminal. The electrical signal input terminals of the four non-equilateral MZMs in the first optical computing module and the electrical signal input terminals of the four non-equilateral MZMs in the second optical computing module are used to synchronously receive four sets of electrical signals corresponding to the similarity of four adjacent pixels of the pixel to be detected in the image, so as to perform edge detection on the pixel to be detected; the adjacent pixels are either horizontally adjacent or vertically adjacent. In this design, at least one waveguide arm of each non-equilateral MZM has a waveguide non-volatile modulation segment, and each waveguide non-volatile modulation segment is provided with a polarization electrode. The material of the waveguide non-volatile modulation segment is lithium tantalate crystal, and a built-in electric field exists within the lithium tantalate crystal. The polarization electrode is used to be subjected to a non-volatile polarization electric field exceeding a preset electric field strength threshold, causing at least some ferroelectric domains within the lithium tantalate crystal of the corresponding waveguide non-volatile modulation segment to flip, thereby causing a non-volatile change in the electro-optic coefficient of the lithium tantalate crystal. Combined with the built-in electric field, the refractive index of the corresponding waveguide non-volatile modulation segment is non-volatilely modulated to achieve control of the static operating point of the corresponding non-equilateral MZM. When the optical computing chip is working, the polarization electric field is removed.

2. The optical computing chip as described in claim 1, characterized in that, When the optical computing chip is working: The optical signal input terminals of the first non-equidistant MZM and the fifth non-equidistant MZM are used to synchronously receive a first wavelength optical signal and a second wavelength optical signal with a preset power ratio. The electrical signal input terminals of the first non-equilateral MZM to the fourth non-equilateral MZM are used to receive four sets of electrical signals corresponding to four horizontally adjacent pixels of the pixel to be detected in the first time window, and to receive four sets of electrical signals corresponding to four vertically adjacent pixels of the pixel to be detected in the second time window. The electrical signal input terminals of the fifth non-equilateral MZM to the eighth non-equilateral MZM are used to receive four sets of electrical signals corresponding to four horizontally adjacent pixels of the pixel to be detected in the first time window, and to receive four sets of electrical signals corresponding to four vertically adjacent pixels of the pixel to be detected in the second time window. The optical signal output terminal of the fourth non-equidistant MZM outputs a first optical signal in the first time window and a second optical signal in the second time window. The optical signal output terminal of the eighth non-equidistant MZM outputs a third optical signal in the first time window and a fourth optical signal in the second time window. The power of the first wavelength light signal and the second wavelength light signal in the first to fourth light signals reflects whether the pixel to be detected is located at the edge.

3. The optical computing chip as described in claim 1 or 2, characterized in that, Also includes: The non-equilateral Mach-Zehnder interferometer (MZI) has two input terminals for receiving optical signals of the first wavelength and the second wavelength, respectively, and two output terminals for connecting to the optical signal input terminals of the first non-equilateral MZM and the fifth non-equilateral MZM, respectively. The interval between the first wavelength and the second wavelength is half of the free spectral range of the non-equilateral MZI, and they operate at the positive and negative quadrature bias points of the non-equilateral MZI, respectively.

4. The optical computing chip as described in claim 1 or 2, characterized in that, The difference in waveguide arm length between the first non-equilateral MZM to the eighth non-equilateral MZM determines the free spectral range of the corresponding non-equilateral MZM. The free spectral ranges of the first to eighth non-equilateral MZMs are designed in combination with the first and second wavelengths, so that after the static operating points of the first to eighth non-equilateral MZMs are adjusted, the optical signal of the first wavelength operates at the negative quadrature bias point, the negative quadrature bias point, the power peak point, the power peak point, the positive quadrature bias point, the positive quadrature bias point, the positive quadrature bias point, and the negative quadrature bias point in the first to eighth non-equilateral MZMs, respectively. The optical signal of the second wavelength operates at the power peak point, the power peak point, the negative quadrature bias point, the negative quadrature bias point, the negative quadrature bias point, the negative quadrature bias point, the positive quadrature bias point, the positive quadrature bias point, and the positive quadrature bias point in the first to eighth non-equilateral MZMs, respectively.

5. The optical computing chip as described in claim 1 or 2, characterized in that, If the absolute value of the difference between the pixel value of the pixel to be detected and the pixel value of its neighboring pixel is not greater than a preset value, the electrical signal input to the corresponding non-equilateral MZM is the first electrical signal; otherwise, the electrical signal input to the corresponding non-equilateral MZM is the second electrical signal. When the first electrical signal is input to the electrical signal input terminal of any non-equal-arm MZM, the phase difference between the two arms of the corresponding non-equal-arm MZM increases by π / 2. When the second electrical signal is input to the electrical signal input terminal of any non-equilateral MZM, the phase difference between the two arms of the corresponding non-equilateral MZM decreases by π / 2.

6. The optical computing chip as described in claim 2, characterized in that, Also includes: The first detection unit includes a first input terminal and a first output terminal; The first input terminal is connected to the optical signal output terminal of the fourth non-equilateral MZM, and the first output terminal outputs a first level signal; The second detection unit includes a second input terminal and a second output terminal; the second input terminal is connected to the optical signal output terminal of the eighth non-equidistant MZM, and the second output terminal outputs a second level signal. When the first wavelength optical signal or the second wavelength optical signal satisfies the coherent constructive condition in the first to fourth non-equidistant MZM, the output power of the first wavelength optical signal or the second wavelength optical signal in the corresponding first optical signal or the second optical signal is close to the corresponding input power, and the corresponding first level signal is high level; otherwise, it is low level. When the first wavelength optical signal or the second wavelength optical signal satisfies the coherent constructive condition in both the fifth to the eighth non-equilateral MZM, the output power of the first wavelength optical signal or the second wavelength optical signal in the corresponding third or fourth optical signal is close to the corresponding input power, and the corresponding second level signal is high; otherwise, it is low.

7. The optical computing chip as described in claim 6, characterized in that, When the first level signal output by the first detection unit in one of the first time window and the second time window is high, the pixel to be detected is located at the edge; When the second level signal output by the second detection unit in both the first and second time windows is high, the pixel to be detected is located at the edge.

8. The optical computing chip as described in claim 1, characterized in that, The four horizontally adjacent pixels of the pixel to be detected include the two pixels adjacent to the left and the two pixels adjacent to the right; The four vertically adjacent pixels of the pixel to be detected include the two pixels adjacent to the top and the two pixels adjacent to the bottom.

9. The optical computing chip as described in claim 1, characterized in that, Before the polarization electric field is applied, the waveguide non-volatile modulation sections of the first to eighth non-equilateral MZMs have a first refractive index, and after the polarization electric field is removed, they have a second refractive index. When the polarization electric field is in the first direction, the second refractive index is greater than the first refractive index; the first direction is opposite to the direction of the ferroelectric domains in the lithium tantalate crystal in the initial state; When the polarization electric field is in the second direction, the second refractive index is less than the first refractive index; the second direction is the same as the direction of the ferroelectric domains in the lithium tantalate crystal in the initial state.

10. The optical computing chip as described in claim 1, characterized in that, When the polarization electric field is in the first direction, the degree of increase in the refractive index of the corresponding non-volatile control section of the waveguide is positively correlated with the degree of increase in the polarization electric field intensity; when the polarization electric field is in the second direction, the degree of decrease in the refractive index of the corresponding non-volatile control section of the waveguide is positively correlated with the degree of increase in the polarization electric field intensity. The first positive correlation has a first positive correlation coefficient, and the second positive correlation has a second positive correlation coefficient; the first positive correlation coefficient is less than the second positive correlation coefficient.

Images