Optical computing unit and method of manufacturing the same, optical computing array

Abstract

This invention provides an optical computing unit and its fabrication method, as well as an optical computing array, belonging to the field of optoelectronic computing technology. The optical computing unit includes: an electro-optic material waveguide, comprising a modulation region and a computational detection region arranged along the light propagation direction; an electro-optic modulator composed of the electro-optic material waveguide of the modulation region, used to receive the input optical signal and modulate the input optical signal according to the applied electrical signal, and output a modulated optical signal; a two-dimensional material layer in contact with the electro-optic material waveguide of the computational detection region, used to absorb the modulated optical signal and generate photogenerated carriers; an electrode structure in electrical contact with the two-dimensional material layer, used to collect photogenerated carriers and output photocurrent; preset structural parameters of the two-dimensional material layer are used to characterize the weighting coefficients of the optical computing unit, and the photocurrent is the result of weighted calculation of the modulated optical signal by the weighting coefficients. This invention achieves a compact, high-speed, low-power optical computing scheme that combines optical modulation and photodetection functions, and whose weights are flexibly programmable.

Description

Technical Field

[0001] This invention relates to the field of optoelectronic computing technology, and in particular to an optical computing unit and its fabrication method, and an optical computing array. Background Technology

[0002] Driven by computationally intensive tasks such as artificial intelligence and combinatorial optimization, unprecedented demands are being placed on the energy efficiency and speed of computing systems. Optical analog computing, with its inherent parallel processing capabilities and high-speed, low-power physical characteristics, demonstrates enormous potential to overcome the bottlenecks of traditional electronic computing. Realizing such optical computing systems typically requires three core functional modules: information modulation, weighted computation, and photodetection. In terms of modulation, the electro-optic modulator is the core component. Regarding weighted computation, existing technologies mainly rely on two approaches: one is to dynamically adjust the light intensity to achieve weight allocation using a modulator array composed of Mach-Zehnder interferometers; the other is to utilize the thermo-optic or electro-optic effects of microring resonators to tune their resonant state, thereby altering the intensity of the transmitted light signal. In terms of photodetection, existing solutions are mainly divided into off-chip detection schemes and on-chip integrated schemes.

[0003] However, all of the above-mentioned technical approaches have inherent drawbacks that limit further improvements in system performance and integration. In schemes that use a modulator array with a Mach-Zehnder interferometer to change light intensity to achieve weighting, the weighting units typically require millimeter-scale dimensions, limiting chip integration and computational scale. Furthermore, they are extremely sensitive to phase noise, requiring continuous application of driving voltage, resulting in high static power consumption. Thermo-optic or electro-optic tuning schemes using micro-ring resonators have limited tuning ranges, and their response speed is limited by thermal effects, making them unsuitable for high-speed computing. Off-chip detection schemes have low integration, require complex optical alignment and packaging, introducing additional coupling losses and signal delays. On-chip integration of III-V group detectors requires heterogeneous integration processes, resulting in high manufacturing costs and process complexity. The detectors themselves are also large, further limiting integration. In summary, existing technologies generally suffer from problems such as discrete functional modules, low integration, difficulty in balancing power consumption and speed, or complex processes. Summary of the Invention

[0004] This invention provides an optical computing unit and its fabrication method, as well as an optical computing array, achieving a compact, high-speed, low-power optical computing solution that combines optical modulation and photoelectric detection functions with flexibly programmable weights.

[0005] In a first aspect, the present invention provides an optical computing unit, comprising: An electro-optic waveguide, comprising a modulation region and a computational detection region arranged sequentially along the light propagation direction; An electro-optic modulator, comprising the electro-optic material waveguide of the modulation region, is used to receive an input optical signal and modulate the input optical signal according to an applied electrical signal, and output a modulated optical signal; A two-dimensional material layer is disposed in contact with the electro-optic material waveguide of the computational detection region, and is used to absorb the modulated optical signal and generate photogenerated carriers; An electrode structure is electrically contacted with the two-dimensional material layer to collect the photogenerated carriers and output photocurrent; wherein, the preset structural parameters of the two-dimensional material layer are used to characterize the weighting coefficients of the optical computing unit, and the photocurrent is the result of the modulated optical signal after weighted calculation by the weighting coefficients.

[0006] In some embodiments, the preset structural parameters include at least one of the following: the geometric length of the two-dimensional material layer along the propagation direction of the modulated optical signal, the geometric shape of the two-dimensional material layer, the material type of the two-dimensional material layer, and the number of sub-material layers of the two-dimensional material layer.

[0007] In some embodiments, the electro-optic modulator includes a traveling-wave electrode, and the electrode structure is located in the same metal layer as the traveling-wave electrode.

[0008] In some embodiments, the electrode structure is located on the side of the two-dimensional material layer away from the electro-optic material waveguide.

[0009] In some embodiments, the two-dimensional material layer is a heterojunction structure formed by stacking multiple two-dimensional materials along a vertical direction.

[0010] In some embodiments, the optical computing unit further includes: An adjustable medium microfluidic channel is disposed in direct or indirect contact with the two-dimensional material layer to change the dielectric environment and / or absorption spectrum of the two-dimensional material layer, thereby adjusting the weighting coefficient of the optical computing unit.

[0011] In some embodiments, the light absorption characteristics of the two-dimensional material layer are adjusted by applying a bias voltage to the electrode structure, so that the two-dimensional material layer operates in the linear region or the nonlinear region; wherein, in the linear region, the photocurrent is linearly related to the light intensity of the modulated light signal to realize the weighted calculation; in the nonlinear region, the photocurrent is nonlinearly related to the light intensity of the modulated light signal to realize the nonlinear activation function.

[0012] In a second aspect, the present invention also provides an optical computing array, comprising a plurality of optical computing units as described in the first aspect, wherein the plurality of optical computing units are arranged side by side, and the electrode structures of each optical computing unit have a preset electrical connection relationship, and the optical computing array is used to sum the photocurrent output by each optical computing unit.

[0013] In some embodiments, the electrode structure includes a first electrode and a second electrode. The first electrode of each optical computing unit is electrically connected to the same common node, and the second electrode of each optical computing unit is independently led out. By controlling the bias state of each second electrode, the optical computing array is controlled to operate in standard summation mode, differential weighted mode, or sparse gating mode.

[0014] In some embodiments, at least two of the plurality of optical computing units have two-dimensional material layers with different preset structural parameters to form a weight matrix.

[0015] Thirdly, the present invention also provides a method for fabricating an optical computing unit, used to fabricate the optical computing unit as described in the first aspect, the fabrication method comprising: The electro-optic material waveguide is formed on the substrate; The electro-optic modulator is fabricated using the electro-optic material waveguide in the modulation region. The two-dimensional material layer is formed on the electro-optical waveguide in the computational detection region; The electrode structure is formed to make electrical contact with the two-dimensional material layer.

[0016] In some embodiments, prior to forming the two-dimensional material layer on the electro-optic waveguide in the computational detection region, the method further includes: The waveguide cladding of the computational detection region is etched away to expose the electro-optical waveguide of the computational detection region.

[0017] This invention monolithically integrates an electro-optic modulator and a two-dimensional material layer onto the same electro-optic waveguide, enabling the modulation, weighting calculation, and photodetection of optical signals to be completed within a compact unit. This avoids the increased area, coupling loss, and signal delay caused by cascading multiple discrete components in traditional solutions. Furthermore, since the weighting coefficients are directly characterized by the preset structural parameters of the two-dimensional material layer, there is no need to dynamically maintain the weights by continuously applying a driving voltage, as in traditional solutions, thus significantly reducing static power consumption. In addition, this structure eliminates the need for complex dynamic tuning circuits, simplifies the control logic, and facilitates the realization of a high-speed, low-power, and highly integrated optical computing unit. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a top view structural diagram of an optical computing unit provided in an embodiment of the present invention; Figure 2 This is a top view schematic diagram of an optical computing array provided in an embodiment of the present invention; Figure 3 This is a bandwidth simulation diagram of a thin-film lithium niobate electro-optic modulator provided in an embodiment of the present invention; Figure 4 This is a simulation diagram of light absorption in an electro-optic waveguide by a two-dimensional material layer, provided by an embodiment of the present invention. Figure 5 This is a schematic flowchart of a method for fabricating an optical computing unit according to an embodiment of the present invention; Figures 6 to 12 This is a schematic diagram of the cross-sectional structure corresponding to different steps in the fabrication process of an electro-optic modulator provided in an embodiment of the present invention; Figures 13 to 16 This is a top view structural diagram of different steps in the preparation process of a two-dimensional material layer provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0021] Driven by computationally intensive tasks such as artificial intelligence and combinatorial optimization, unprecedented demands are being placed on the energy efficiency and speed of computing systems. Light-based analog computing, such as optical neural networks and optical Ising machines, demonstrates enormous potential to break through the bottlenecks of traditional electronic computing due to its inherent parallel processing capabilities and high-speed, low-power physical characteristics. Realizing such optical computing systems typically requires three core functional modules: information modulation, weighted computation, and photoelectric detection.

[0022] In modulation, electro-optic modulators are the core components. In recent years, thin-film lithium niobate, with its excellent electro-optic coefficient, low optical loss, and compatibility with complementary metal-oxide-semiconductor (CMOS) processes, has become an ideal platform for realizing high-bandwidth, low-half-wave voltage, and high-performance modulators, and its related manufacturing processes are becoming increasingly mature. In weighted calculation, the key lies in achieving precise and programmable control of the optical signal weights. Existing technologies mainly rely on two paths: one is to dynamically adjust the light intensity to achieve weight allocation through a modulator array composed of Mach-Zehnder interferometers; the other is to utilize the thermo-optic or electro-optic effects of microring resonators to tune their resonant state, thereby changing the intensity of the transmitted optical signal. In photoelectric detection, existing solutions are mainly divided into two categories: one is off-chip detection, where the modulated optical signal is coupled out of the chip via optical fiber and photoelectric conversion and signal integration are completed in a discrete detector; the other is on-chip integration, which typically combines III-V group semiconductor detectors, such as improved single-row carrier detectors, with silicon-based or silicon nitride photonic circuits through heterogeneous integration technology to complete photoelectric conversion on-chip.

[0023] While the aforementioned technical approaches each have their own characteristics, they all suffer from inherent drawbacks that limit further improvements in system performance and integration: First, in schemes that use a modulator array based on a Mach-Zehnder interferometer to change light intensity to achieve weighting, the weighting units typically require millimeter-scale dimensions, limiting chip integration and computational scale. Furthermore, this scheme is extremely sensitive to phase noise, and requires continuous application of a driving voltage to achieve dynamic tuning, resulting in high static power consumption. Second, thermo-optical or electro-optical tuning schemes using micro-ring resonators have limited tuning ranges, and their response speed is limited by thermal effects, making it difficult to meet high-speed computing requirements. Third, off-chip detection schemes have significantly low integration, requiring complex optical alignment and packaging, introducing additional coupling losses and signal delays. Traditional units can only perform a single function of modulation or detection; completing a complete optoelectronic computing link requires cascading multiple discrete devices, resulting in significant signal loss. Fourth, schemes that integrate III-V group detectors on-chip require heterogeneous integration processes, leading to high manufacturing costs and process complexity. Simultaneously, the large size of III-V group detectors themselves limits integration, thus restricting the overall computational scale.

[0024] In summary, existing technologies generally suffer from problems such as separate functional modules, low integration, difficulty in balancing power consumption and speed, or complex processes. There is a lack of an on-chip solution that can efficiently and compactly integrate high-performance modulation, programmable weighting, and efficient detection.

[0025] This invention aims to address the problems of large area, difficulty in balancing power consumption and speed, limited integration, and limited functionality in existing optical computing weighting units. It proposes a compact, high-speed, low-power, and programmable weighting monolithic optoelectronic computing unit and array based on thin-film lithium niobate and two-dimensional materials. This invention belongs to the field of integrated photonics and optoelectronic computing, specifically relating to optoelectronic integrated units and arrays for simulating optical computing, particularly for realizing programmable weighted calculation. By integrating a high-speed electro-optic modulator and an on-chip detector based on two-dimensional materials on the same thin-film lithium niobate chip, it integrates three major functions: optical signal modulation, weight calculation, and photoelectric conversion. Its core lies in directly defining the calculation weights using the material properties and geometric dimensions such as length of different two-dimensional materials, eliminating the need for complex dynamic tuning circuits. Ultimately, it achieves a highly compact, high-speed, low-power, programmable weighting, and naturally scalable optical computing hardware foundation unit and system solution.

[0026] Figure 1 This is a top view structural diagram of an optical computing unit provided in an embodiment of the present invention. Figure 1 As shown, the optical computing unit includes an electro-optic material waveguide 1, which includes a modulation region 2 and a computing detection region 3 arranged sequentially along the light propagation direction. The optical computing unit also includes an electro-optic modulator 4, a two-dimensional material layer 5, and an electrode structure. The electro-optic modulator 4 is composed of the electro-optic material waveguide 1 in the modulation region 2 and is used to receive the input optical signal and modulate the input optical signal according to the applied electrical signal, and output the modulated optical signal. The two-dimensional material layer 5 is in contact with the electro-optic material waveguide 1 in the computing detection region 3 and is used to absorb the modulated optical signal and generate photogenerated carriers. The electrode structure is in electrical contact with the two-dimensional material layer 5 and is used to collect the photogenerated carriers and output photocurrent. The preset structural parameters of the two-dimensional material layer 5 are used to characterize the weighting coefficients of the optical computing unit, and the photocurrent is the result of the modulated optical signal after weighted calculation by the weighting coefficients.

[0027] Specifically, the electro-optic waveguide 1 refers to a waveguide structure made of materials with electro-optic effects, such as thin-film lithium niobate, used to transmit optical signals and capable of changing its refractive index under the action of an applied electric field. The electro-optic waveguide 1 is divided into a modulation region 2 and a computational detection region 3 arranged sequentially along the light propagation direction. The modulation region 2 refers to the part of the electro-optic waveguide 1 used to form the electro-optic modulator 4. The electro-optic modulator 4 refers to a device that uses the electro-optic effect to load an electrical signal onto an optical signal. The electro-optic waveguide 1 in this region performs the function of optical signal modulation. The computational detection region 3 refers to the part of the electro-optic waveguide 1 located after the modulation region 2. The electro-optic waveguide 1 in this region is used to cooperate with the two-dimensional material layer 5 to realize weighted calculation and photoelectric detection functions.

[0028] The two-dimensional material layer 5 refers to a layered structure composed of atomically thin materials, such as graphene and transition metal chalcogenides, exhibiting excellent photoelectric response characteristics. The atomic-level thickness of the two-dimensional material layer 5 enables highly efficient coupling with the evanescent field of the electro-optic material waveguide 1 with extremely low optical mode mismatch. This allows for in-situ conversion of optical signals into electrical signals with almost no additional insertion loss, significantly improving system integration and reliability. Simultaneously, these materials possess ultra-high carrier mobility and tunable band structures, providing a physical basis for achieving ultra-high speeds exceeding 100 GHz and on-chip detection with a broad spectral response from visible light to mid-infrared. Furthermore, they have good process compatibility with thin-film lithium niobate platforms and are easy to integrate. The electrode structure refers to the conductive structure in electrical contact with the two-dimensional material layer 5, used to collect photogenerated carriers and output photocurrent, serving together with the two-dimensional material layer 5 as an on-chip two-dimensional material detector. Preset structural parameters refer to quantifiable characteristics related to the two-dimensional material layer 5, pre-defined during manufacturing, including geometric dimensions, shape, and material composition. The weighting coefficient refers to the scaling factor used to scale the input optical signal in optical computing, and is used to characterize the weight contributed by the optical computing unit in the computation.

[0029] An input optical signal enters the electro-optic waveguide 1. When the input optical signal propagates to the modulation region 2, the electro-optic modulator 4 modulates the input optical signal according to the externally applied electrical signal, loading electrical information onto the optical signal and outputting a modulated optical signal. The modulated optical signal continues to propagate along the electro-optic waveguide 1 and enters the computational detection region 3. In the computational detection region 3, a two-dimensional material layer 5, which is in contact with the electro-optic waveguide 1, absorbs the modulated optical signal. The energy of the photons is absorbed by the two-dimensional material layer 5, exciting photogenerated carriers within it. Electrode structures electrically in contact with the two-dimensional material layer 5 collect these photogenerated carriers and output a photocurrent. In this process, the preset structural parameters of the two-dimensional material layer 5, such as its geometric length along the direction of light propagation, determine the absorption capacity of the two-dimensional material layer 5 for the modulated optical signal. For a fixed input optical power and modulation state, the stronger the absorption capacity of the two-dimensional material layer 5, the more photogenerated carriers are generated, and the larger the output photocurrent. Therefore, there is a definite correspondence between the magnitude of the output photocurrent and the preset structural parameters of the two-dimensional material layer 5. This correspondence allows the preset structural parameters of the two-dimensional material layer 5 to characterize the weighting coefficients of the optical computing unit, and the output photocurrent is the result of the modulated optical signal after weighted calculation by these weighting coefficients. In other words, the optical computing unit simultaneously performs three functions—optical signal modulation, weighted calculation, and photoelectric detection—within a single structure, and the result of the weighted calculation is directly reflected in the magnitude of the photocurrent.

[0030] Therefore, this embodiment of the invention monolithically integrates the electro-optic modulator 4 and the two-dimensional material layer 5 onto the same electro-optic material waveguide 1, enabling the three functions of optical signal modulation, weighted calculation, and photoelectric detection to be completed within a compact unit. This avoids the increased area, coupling loss, and signal delay caused by cascading multiple discrete devices in traditional solutions. Furthermore, since the weighting coefficients are directly characterized by the preset structural parameters of the two-dimensional material layer 5, there is no need to dynamically maintain the weights by continuously applying a driving voltage, as in traditional solutions, thus significantly reducing static power consumption. In addition, this structure eliminates the need for complex dynamic tuning circuits, simplifies control logic, and facilitates the realization of a high-speed, low-power, and highly integrated optical computing unit.

[0031] Figure 2 This is a top view schematic diagram of an optical computing array provided in an embodiment of the present invention. Combined with... Figure 1 and Figure 2 Multiple optical computing units can form an optical computing array. Different electro-optic modulators 4 load different input information, and the type or size of the two-dimensional material layer 5 integrated in different channels can be adjusted. For example, by changing the geometric length of the two-dimensional material layer 5 along the propagation direction of the modulated optical signal, its light absorption capability can be adjusted, thereby regulating the output photocurrent and realizing coded weight control. For a fixed input optical power and modulation voltage, the photocurrent intensity Iph output by each optical computing unit is proportional to the geometric length L of the two-dimensional material layer 5 along the propagation direction of the modulated optical signal, i.e., Iph∝R*L, where R is the intrinsic responsivity of the two-dimensional material layer 5. Therefore, by changing R or designing different L, a fixed weight coefficient can be assigned to each optical computing unit. In the optical computing array, the photocurrent of different optical computing units naturally carries weight information. Each optical computing unit generates a weighted photocurrent in parallel based on its own weight. Summing these photocurrents at the circuit end can realize a single vector-matrix multiplication or weighted summation analog optical computing operation.

[0032] Figure 3 This is a bandwidth simulation diagram of a thin-film lithium niobate electro-optic modulator provided in an embodiment of the present invention. Figure 3 The horizontal axis represents frequency, and the vertical axis represents microwave scattering parameters, including return loss S. 11 With transmission coefficient S 21 .Depend on Figure 3 It can be seen that, under the condition of matching the speed of light waves and microwaves, the electro-optic -3dB bandwidth of electro-optic modulator 4 corresponds to the electrical -6.4dB bandwidth. Within the 0-300GHz range, S... 21 All values ​​are above -6.4dB, indicating that the electro-optic modulator 4 can achieve a bandwidth greater than 300GHz. Figure 4 This is a simulation diagram of light absorption in an electro-optic waveguide by a two-dimensional material layer, provided by an embodiment of the present invention. Figure 4The horizontal and vertical axes represent the two vertical position coordinates of waveguide 1, an electro-optic material. The values ​​on the right describe the light absorption effect. Taking a 10-layer graphene material layer 5 with a thickness of approximately 5 nm as an example, the following values ​​are used: Figure 4 It can be seen that the two-dimensional material layer 5 can effectively absorb the optical signal in the electro-optic material waveguide 1, and the absorption tends to saturate at a length of about 60 μm.

[0033] In some embodiments, the preset structural parameters include at least one of the following: the geometric length of the two-dimensional material layer 5 along the direction of propagation of the modulated optical signal, the geometric shape of the two-dimensional material layer 5, the material type of the two-dimensional material layer 5, and the number of sub-material layers of the two-dimensional material layer 5.

[0034] Specifically, geometric length refers to the extension dimension of the two-dimensional material layer 5 in the direction of modulation of the optical signal propagation. Geometric shape refers to the planar graphic outline of the two-dimensional material layer 5 on the surface of the electro-optic waveguide 1, such as a rectangle, trapezoid, or a patterned shape with periodic holes. Material type refers to the specific material type constituting the two-dimensional material layer 5, such as graphene, molybdenum disulfide, tungsten disulfide, black phosphorus, etc. Sub-material layer number refers to the number of layers contained when the two-dimensional material layer 5 is composed of multiple stacks of a single material.

[0035] The absorption capability of the two-dimensional material layer 5 for modulated optical signals is determined by multiple structural parameters. The longer the geometric length along the propagation direction of the modulated optical signal, the farther the signal travels within the two-dimensional material layer 5, resulting in more absorbed photons and more photogenerated carriers. The geometry affects the spatial overlap distribution between the optical field and the two-dimensional material layer 5; different shapes produce different absorption efficiency distributions. The type of material determines its photoelectric responsivity, i.e., the photocurrent generated per unit of optical power. Different materials have different band structures and carrier mobilities, resulting in different photoelectric response characteristics. The number of sub-material layers affects the total thickness and band structure of the two-dimensional material layer 5; more layers mean stronger light absorption. By selecting or designing one or more of the above parameters, the absorption capability of the two-dimensional material layer 5 for modulated optical signals can be precisely controlled, thereby defining the weighting coefficients of this optical computing unit.

[0036] Therefore, by providing a variety of adjustable preset structural parameters, this invention offers rich design freedom for defining the weighting coefficients of optical computing units. The weighting coefficients can be flexibly set by selecting one or more combinations of geometric length, geometric shape, material type, and number of sub-material layers according to the needs of specific computing tasks. This makes the setting of weighting coefficients more precise and flexible, compatible with existing semiconductor processes, and easy to implement.

[0037] In some embodiments, the electro-optic modulator 4 includes a traveling wave electrode 6, and the electrode structure is located on the same metal layer as the traveling wave electrode 6.

[0038] Specifically, the traveling-wave electrode 6 refers to the electrode structure used in the electro-optic modulator 4, which enables electrical and optical signals to propagate synchronously in the electro-optic waveguide 1. The same metal layer refers to the electrode structure and the traveling-wave electrode 6 being formed through the same metal deposition and patterning process during chip manufacturing, and located at the same vertical height layer. For example... Figure 1 As shown, the left side of the optical computing unit is the light source input port. During propagation, the light is split into two identical beams by the beam splitting unit 11 and input into the electro-optic modulator 4, which is composed of the electro-optic material waveguide 1 of the modulation region 2. The electro-optic modulator 4 adopts a Mach-Zehnder interferometer structure and uses a coplanar waveguide structure of signal-ground-signal as the traveling wave electrode 6 to achieve push-pull modulation. During the modulation stage, an electrical signal is applied to the traveling wave electrode 6, and the refractive index of the electro-optic material waveguide 1 is rapidly changed through the strong electro-optic effect, thereby modulating the phase or intensity of the optical signal transmitted therein. After the modulated optical signal is combined by the beam combining unit 12, its evanescent field interacts with the two-dimensional material layer 5 disposed on the electro-optic material waveguide 1 in the computing detection region 3. The two-dimensional material layer 5 absorbs photons to generate photogenerated carriers, which output photocurrent through the electrode structure.

[0039] On an electro-optic material platform such as thin-film lithium niobate, the traveling-wave electrode 6 of the electro-optic modulator 4 is formed through metal deposition and patterning processes. The electrode structure and the traveling-wave electrode 6 are fabricated simultaneously in the same metal layer. When the patterning of the traveling-wave electrode 6 is completed, the pattern of the electrode structure is also formed simultaneously, without the need for additional metal deposition and patterning steps. The electrode structure and the traveling-wave electrode 6 are located at the same vertical height layer of the chip, and they can be made of the same metal material, such as gold, aluminum, copper, etc.

[0040] Therefore, by placing the electrode structure and the traveling wave electrode 6 on the same metal layer, this embodiment of the invention simplifies the manufacturing process of the optical computing unit, reduces the number of photolithography and metal deposition steps required, and lowers manufacturing costs and process complexity. Simultaneously, same-layer fabrication avoids parasitic capacitance and resistance that may be introduced by cross-layer connections, which is beneficial for improving the high-speed performance of the device. Furthermore, this design allows for a more compact layout of the electrode structure and the traveling wave electrode 6, which is beneficial for increasing the chip's integration density.

[0041] In some embodiments, the electrode structure is located on the side of the two-dimensional material layer 5 away from the electro-optic material waveguide 1.

[0042] Specifically, the electrode structure is positioned above the two-dimensional material layer 5, i.e., the two-dimensional material layer 5 is located between the electro-optic waveguide 1 and the electrode structure. When the modulated optical signal propagates in the electro-optic waveguide 1, its evanescent field interacts with the two-dimensional material layer 5, causing the two-dimensional material layer 5 to absorb photons and generate photogenerated carriers. Since the electrode structure is located on the side of the two-dimensional material layer 5 away from the electro-optic waveguide 1, the photogenerated carriers need to pass through the thickness direction of the two-dimensional material layer 5 to be collected by the electrode structure. This configuration allows the electrode structure to directly form good electrical contact with the upper surface of the two-dimensional material layer 5, avoiding the introduction of an additional electrode layer between the two-dimensional material layer 5 and the electro-optic waveguide 1, thereby simplifying the fabrication process and reducing the potential interference of the electrode material on the propagation of the optical field.

[0043] Therefore, by placing the electrode structure on the side of the two-dimensional material layer 5 away from the electro-optic material waveguide 1, the fabrication process is simplified, the interference of the electrode material on the propagation of the light field is avoided, and the optoelectronic performance and integration density of the optical computing unit are improved.

[0044] In some embodiments, the two-dimensional material layer 5 is a heterojunction structure formed by stacking multiple two-dimensional materials in a vertical direction.

[0045] Specifically, a heterojunction structure refers to a composite structure formed by stacking two or more different materials in a vertical direction, with interfaces formed between the layers. The vertical direction refers to the direction perpendicular to the surface of the substrate where the electro-optic waveguide 1 is located. The two-dimensional material layer 5 is not limited to a single-layer or multi-layer structure of a single material; it can also use various different two-dimensional materials, such as graphene and molybdenum disulfide stacked in a vertical direction to form a heterojunction structure. Different two-dimensional materials have different band structures and photoelectric response characteristics. When the modulated optical signal propagates in the electro-optic waveguide 1, its evanescent field interacts with the two-dimensional material layers in the heterojunction structure. Due to the differences in the band structures of the materials in each layer, light of different wavelengths may be preferentially absorbed in different layers; due to the differences in carrier mobility of the materials in each layer, the transport and collection efficiency of photogenerated carriers in the vertical direction will also be affected. Therefore, the interlayer combination method, stacking order, and thickness of each layer of the heterojunction structure can all be used as part of the preset structural parameters to characterize the weighting coefficients of the optical computing unit.

[0046] Therefore, this invention, through the use of a heterojunction structure formed by stacking multiple two-dimensional materials along a vertical direction, further expands the programming dimension of weighting coefficients. By selecting different material combinations and stacking orders, richer and more refined weighting coefficient definitions can be achieved. The heterojunction structure can also enable wavelength-selective weighted calculations, that is, assigning different weighting coefficients to modulated optical signals of different wavelengths, thus providing possibilities for wavelength division multiplexing (WDM) optical computation.

[0047] In some embodiments, the optical computing unit further includes an adjustable dielectric microfluidic channel, which is disposed in direct or indirect contact with the two-dimensional material layer 5, for changing the dielectric environment and / or absorption spectrum of the two-dimensional material layer 5 to adjust the weighting coefficient of the optical computing unit.

[0048] Specifically, a tunable medium microfluidic channel refers to a micrometer-scale channel structure integrated on a chip for containing and transporting tunable media. Direct or indirect contact: Direct contact refers to direct contact between the tunable medium and the surface of the two-dimensional material layer 5; indirect contact refers to a very thin insulating layer between the two-dimensional material layer 5 and the tunable medium, with the tunable medium influencing the two-dimensional material layer 5 through capacitive coupling. Dielectric environment refers to the dielectric constant distribution in the region where the two-dimensional material layer 5 is located, affecting its carrier transport characteristics and optical response. Absorption spectrum refers to the distribution of absorption efficiency of the two-dimensional material layer 5 for different wavelengths of light signals.

[0049] A tunable medium microfluidic channel can be set in the empty area above or around the two-dimensional material layer 5. Different types of tunable media can be introduced into the channel, such as ionic liquids, solutions containing nanomaterials, and liquids with specific dielectric constants. When the tunable medium flows or remains in the channel, the dielectric environment around the two-dimensional material layer 5 can be changed through direct contact (e.g., ion implantation, charge transfer) or indirect contact (e.g., capacitive coupling, electric field modulation), or by chemical doping to change the carrier concentration of the two-dimensional material layer 5, thereby altering its band structure and absorption spectrum. Changes in the dielectric environment and absorption spectrum directly affect the absorption efficiency of the two-dimensional material layer 5 for modulated optical signals, as well as the generation and collection efficiency of photogenerated carriers, thus changing the magnitude of the output photocurrent. Since the photocurrent magnitude directly corresponds to the weighting coefficients, the weighting coefficients of the optical computing unit can be dynamically adjusted by controlling the tunable medium in the tunable medium microfluidic channel.

[0050] Therefore, by introducing an adjustable medium microfluidic channel, this embodiment of the invention provides the optical computing unit with the ability to adjust weights in the later stage, making up for the deficiency that the weights defined by preset structural parameters cannot be changed after manufacturing. This fluid programming method does not consume continuous electrical power, does not introduce electrical noise, and can realize global weight reconfiguration. It is particularly suitable for computing scenarios that require one-time programming and long-term use, or integrated sensing and computing systems that need to be adaptively adjusted according to the environment.

[0051] In some embodiments, the light absorption characteristics of the two-dimensional material layer 5 are adjusted by a bias voltage applied to the electrode structure, so that the two-dimensional material layer 5 operates in the linear region or the nonlinear region; wherein, in the linear region, the photocurrent is linearly related to the light intensity of the modulated light signal to achieve weighted calculation; in the nonlinear region, the photocurrent is nonlinearly related to the light intensity of the modulated light signal to achieve a nonlinear activation function.

[0052] Specifically, by changing the bias voltage applied to the electrode structure, the carrier concentration or band structure of the two-dimensional material layer 5 can be adjusted, thereby altering its absorption characteristics for optical signals. When the bias voltage causes the two-dimensional material layer 5 to operate in the linear region, its absorption coefficient is constant, and the photocurrent is linearly related to the incident light intensity. This is suitable for performing weighted summation operations in matrix multiplication, enabling controllable weighted calculation of optical signals. When the bias voltage causes the two-dimensional material layer 5 to operate in the nonlinear region, its nonlinear optical absorption characteristics are excited, such as saturated absorption and anti-saturated absorption, manifested as an absorption coefficient that changes with the incident light intensity. Taking saturated absorption as an example, the absorption coefficient is high when the incident light intensity is low, and when the incident light intensity exceeds a threshold, absorption tends to saturate, the absorption coefficient decreases significantly, and transparency occurs. Therefore, when the intensity of the modulated optical signal changes, the absorption efficiency of the two-dimensional material layer 5 changes nonlinearly, resulting in a nonlinear relationship between the number of photogenerated carriers and the incident light intensity, which in turn makes the output photocurrent nonlinearly related to the intensity of the modulated optical signal. This nonlinear input-output relationship precisely simulates the nonlinear activation functions in artificial neural networks, such as the Sigmoid function and the ReLU function. By further adjusting the bias voltage applied to the electrode structure, the threshold and slope of the nonlinear response can be continuously controlled, achieving dynamic adjustment of the activation function shape.

[0053] The material used to construct the two-dimensional material layer 5 can be selected from two-dimensional materials with tunable light absorption properties, such as multilayer graphene, transition metal chalcogenides, and black phosphorus. These materials not only possess excellent carrier mobility and photoelectric response characteristics, but their light absorption properties can also be tuned over a wide range through the electric field effect, thus enabling flexible switching between linear weighting and nonlinear activation functions on the same device. It should be noted that in optical computing, matrix operations constitute the majority of the computation; that is, the two-dimensional material layer 5 primarily operates in the linear region.

[0054] Therefore, this embodiment of the invention adjusts the working region of the two-dimensional material layer 5 by biasing the voltage, enabling the same optical computing unit to operate in both the linear region for weighted calculations and the nonlinear region for nonlinear activation functions. This eliminates the need for additional electronic circuitry to perform nonlinear activation after photoelectric conversion, as is required in traditional schemes, and also eliminates the need for separate devices for weighted calculations and nonlinear activation. This configurable design significantly reduces system latency and power consumption, increases computational density, and allows the optical computing unit to function as a complete programmable neuron node, providing a flexible hardware foundation for constructing fully optoelectronic hybrid deep neural networks.

[0055] This invention also provides an optical computing array, such as... Figure 2As shown, the optical computing array includes multiple optical computing units as described in the above embodiments. The multiple optical computing units are arranged side by side, and the electrode structures of each optical computing unit have a preset electrical connection relationship. The optical computing array is used to sum the photocurrent output by each optical computing unit.

[0056] Specifically, an optical computing array refers to an array structure formed by arranging multiple optical computing units according to a certain pattern, used to perform optical computing tasks such as vector-matrix multiplication in parallel. Side-by-side arrangement means that multiple optical computing units are arranged in parallel along a direction perpendicular to the direction of light propagation. Preset electrical connection relationship refers to the electrical connection method between the electrode structures of each optical computing unit, pre-defined according to computing requirements. Summation refers to the superposition of the photocurrents output by multiple optical computing units at the circuit terminals to obtain the total current.

[0057] Multiple optical computing units are arranged side-by-side on the same chip to form an optical computing array. Each optical computing unit independently receives the input optical signal, which can be multiple optical signals split from the same light source. It independently performs optical signal modulation, weighted calculation, and photodetection, and each outputs a weighted photocurrent. The electrode structures of each optical computing unit have preset electrical connections, for example... Figure 2 The diagram shows all the first electrodes of the optical computing units connected to the same common node. When each optical computing unit outputs photocurrent, these currents naturally superimpose at the common node, forming a total current. The magnitude of this total current is equal to the algebraic sum of the output photocurrents of each unit, thus achieving the summation operation of the output photocurrents of each unit. In optical computing, this summation operation corresponds to the accumulation step in vector-matrix multiplication. The input of each optical computing unit represents an element in a vector, the weight of the unit represents the corresponding element in the matrix, the output photocurrent represents the product result, and the summation yields the final vector-matrix multiplication result.

[0058] Therefore, this embodiment of the invention achieves parallel computing capabilities of an optical computing array by arranging multiple optical computing units side-by-side and establishing a preset electrical connection relationship. Each unit works independently, performing weighted calculations simultaneously and completing the summation at the circuit level. The entire process requires no complex timing control or data transfer, supporting large-scale expansion by adding more optical computing units to increase the number of parallel computing channels, thereby expanding the computing scale. Compared with traditional solutions, the array structure is more compact, consumes less power, and is easier to integrate with subsequent electronic processing circuits.

[0059] In some embodiments, the electrode structure includes a first electrode 7 and a second electrode 8. The first electrode 7 of each optical computing unit is electrically connected to the same common node A, and the second electrode 8 of each optical computing unit is independently led out. By controlling the bias state of each second electrode 8, the optical computing array can be controlled to work in standard summation mode, differential weighted mode, or sparse gating mode.

[0060] Specifically, the first electrode 7 and the second electrode 8 refer to the two electrodes in each optical computing unit that are electrically in contact with the two-dimensional material layer 5, respectively used to collect photogenerated carriers and form current loops. The common node A refers to the electrical node that connects the first electrodes 7 of all optical computing units. Independent lead-out means that the second electrode 8 of each optical computing unit is independently led out to an external control circuit, not directly connected to the second electrodes 8 of other optical computing units. Bias state refers to the voltage state of the second electrode 8 relative to a reference potential, such as ground potential, which can be set to ground, positive bias, negative bias, or high impedance. Standard summation mode refers to the mode where the second electrodes 8 of all optical computing units are grounded; in this case, the total current output by the common node is the algebraic sum of the photocurrents output by each optical computing unit. Differential weighting mode refers to the mode where the second electrodes 8 of some optical computing units are connected to a negative bias; in this case, the polarity of the photocurrents output by these optical computing units is reversed, achieving negative weighting calculation. In this case, the first electrode 7 is always connected to the common node, corresponding to the ground potential; the common node is connected to the input terminal of a current reading device such as an electricity meter. The sparse gating mode refers to a mode in which the second electrode 8 of some optical computing units is set to a high-impedance state. In this mode, these optical computing units do not participate in the summation, thus achieving sparse computation. At this time, the first electrode 7 is always connected to a common node, which is connected to the input terminal of a current reading device such as an electricity meter.

[0061] Each optical computing unit's electrode structure includes a first electrode 7 and a second electrode 8. The first electrodes 7 of all optical computing units are electrically connected to the same common node, serving as the summation output. The second electrode 8 of each optical computing unit is independently led out to an external control circuit. By controlling the bias state of each second electrode 8 through the external circuit, the output characteristics of the corresponding optical computing unit can be changed. In the standard summation mode, all second electrodes 8 are grounded, and photogenerated carriers flow from the first electrodes 7 to the common node, with the positive currents output by each optical computing unit superimposed at the common node. In the differential weighting mode, the second electrodes 8 of some optical computing units are connected to a negative bias voltage. The two-dimensional material layer 5 of these optical computing units is subjected to a reverse electric field, reversing the flow direction of photogenerated carriers. This causes the current flowing from the first electrode 7 to flow in the opposite direction to the standard mode, resulting in a negative current at the common node. Thus, the contribution of these optical computing units to the total current is negative, achieving negative weighting calculation. In the sparse gating mode, the second electrode 8 of some optical computing units is set to a high-resistivity state. The photogenerated carriers generated by the two-dimensional material layer 5 of these optical computing units cannot form an effective external circuit current. This is equivalent to these optical computing units being turned off and not participating in the summation, thus achieving the sparsification of the computing task and activating only the necessary optical computing units to reduce power consumption.

[0062] By controlling the bias state of each second electrode 8, the carrier concentration and photoelectric response characteristics of each two-dimensional material layer 5 can be changed. When the bias voltage is zero or a preset reference voltage, the two-dimensional material layer 5 operates in its inherent responsivity state, corresponding to the standard summation mode. When the bias voltage is positive or negative, the responsivity of the two-dimensional material layer 5 increases or decreases, which is equivalent to changing the weighting coefficient of the optical computing unit. When the bias voltage makes the responsivity negative, a differential weighted mode can be realized. When the bias voltage makes the two-dimensional material layer 5 in a carrier depletion state, its responsivity approaches zero, and the optical computing unit does not participate in the summation, realizing a sparse gating mode. In high-speed optical computing applications, a field-programmable gate array can be used to simultaneously realize bias voltage control and current readout.

[0063] Therefore, this embodiment of the invention achieves switching between multiple computing modes without the need for additional circuitry by sharing the first electrode 7 of each optical computing unit and independently leading out and controlling the bias state of the second electrode 8. The standard summation mode performs positive weight accumulation; the differential weighting mode enables the optical computing array to handle computing tasks with negative weights; and the sparse gating mode allows for the dynamic shutdown of some computing units based on the sparsity of the input data, further reducing power consumption. This multi-mode capability significantly enhances the flexibility and applicability of the optical computing array.

[0064] In some embodiments, among the plurality of optical computing units, at least two optical computing units have two-dimensional material layers 5 with different preset structural parameters to form a weight matrix.

[0065] Specifically, the weight matrix refers to a matrix formed by arranging the weight coefficients of multiple optical computing units in an array, used to represent the weight parameters in the computational task performed by the optical computing array. In the optical computing array, the weight coefficient of each optical computing unit is determined by the preset structural parameters of the two-dimensional material layer 5 of that unit. By designing different optical computing units with different preset structural parameters, such as different geometric lengths, different material types, and different numbers of sub-material layers, different optical computing units can have different weight coefficients. Organizing these optical computing units with different weight coefficients according to the array arrangement constitutes a weight matrix. When the input optical signal is split and input to each optical computing unit, each optical computing unit performs weighted calculations on the input signal according to its own weight coefficients, and outputs a weighted photocurrent. The result of summing the photocurrents is the product of the input vector and the weight matrix.

[0066] Therefore, by setting at least two optical computing units with different preset structural parameters, the optical computing array can achieve any specified weight matrix. According to the requirements of the specific computing task, the required weight coefficients can be set for each optical computing unit during the manufacturing stage through the design of a photolithographic mask, thereby directly writing the trained neural network weights onto the chip. This static weight matrix implementation method does not require dynamic tuning circuits, has extremely low power consumption and extremely high computing speed, and is particularly suitable for inference computing scenarios with fixed weights.

[0067] In summary, this invention integrates a two-dimensional material layer 5 in a specific region on the surface of the electro-optic waveguide 1 after the modulator, thereby enabling simultaneous on-chip electro-optic modulation and photoelectric detection functions. A novel weight definition method is proposed, which changes the weighting coefficients of the optical computing unit by controlling the geometric length of the two-dimensional material layer 5 in the light propagation direction. A novel programmable weight extension scheme is proposed, where the weighting coefficients can be defined and programmed by selecting two-dimensional material layers 5 with different photoelectric response characteristics. A novel optical computing array is proposed, consisting of multiple optical computing units, each of which integrates a fixed programmable weight matrix through two-dimensional material layers 5 with, for example, different geometric lengths and / or different material types, for performing parallel simulation calculations. Thus, by directly defining the inherent weights through, for example, the geometric dimensions of the two-dimensional material layer 5, the problems of low integration, large power consumption area, limited speed, and complex control inherent in traditional discrete schemes are fundamentally solved. The proposed structure combines high performance and high scalability, providing a new hardware foundation for building next-generation high-speed, low-power, compact analog optical computing systems. Meanwhile, corresponding principle verification and simulation analysis were carried out, and feasible micro-nano fabrication process steps were planned, including etching of electro-optic material waveguide 1, fabrication of electro-optic modulator 4, precise transfer and patterning of two-dimensional material layer 5, and metal electrode processing, laying the foundation for the actual fabrication and testing of this innovative structure.

[0068] This invention also provides a method for fabricating an optical computing unit. Figure 5 This is a schematic flowchart illustrating a method for fabricating an optical computing unit according to an embodiment of the present invention. The method for fabricating the optical computing unit can be used to fabricate the optical computing unit as described in the above embodiment. Figure 5 As shown, the fabrication method of the optical computing unit includes the following steps: S101. An electro-optic waveguide is formed on the substrate.

[0069] S102. An electro-optic modulator is fabricated using an electro-optic material waveguide in the modulation region.

[0070] S103. A two-dimensional material layer is formed on the electro-optical waveguide in the computational detection region.

[0071] S104. Form an electrode structure that is in electrical contact with the two-dimensional material layer.

[0072] In some embodiments, before forming the two-dimensional material layer on the electro-optic material waveguide of the computational detection region, the method further includes: etching away the waveguide cladding of the computational detection region to expose the electro-optic material waveguide of the computational detection region.

[0073] Specifically, the substrate refers to the base material used to support and manufacture optical computing units, such as a silicon dioxide substrate. The waveguide cladding refers to the dielectric layer covering the electro-optic waveguide, which can be made of materials such as silicon dioxide, and is used to protect the electro-optic waveguide and provide optical confinement. Etching removal refers to the selective removal of the waveguide cladding in specific areas using dry or wet etching processes.

[0074] First, an electro-optic waveguide, comprising a modulation region and a computational detection region, is formed on a substrate. Then, an electro-optic modulator is fabricated using the electro-optic waveguide in the modulation region. This step includes standard processes such as ridge waveguide etching, silicon oxide deposition, electrode trench windowing, and metal evaporation to complete the fabrication of the electro-optic modulator. Next, the cladding material on the waveguide of the computational detection region is etched away. This step removes the cladding material originally covering the electro-optic waveguide in the computational detection region, exposing the surface of the electro-optic waveguide in this region. Subsequently, a two-dimensional material layer is formed on the exposed electro-optic waveguide of the computational detection region, ensuring direct contact between the two-dimensional material layer and the surface of the electro-optic waveguide. Finally, an electrode structure electrically in contact with the two-dimensional material layer is formed.

[0075] Therefore, this embodiment of the invention fully utilizes the mature technology of the thin-film lithium niobate platform, adding only the steps of etching the cladding, transferring the two-dimensional material, and fabricating electrodes after the standard modulator fabrication process, resulting in good process compatibility. The sequence of fabricating the electro-optic modulator first and then integrating the two-dimensional material avoids damage to the two-dimensional material caused by high-temperature and plasma processes during modulator fabrication, ensuring the excellent optoelectronic performance of the two-dimensional material. The etching cladding step ensures direct contact between the two-dimensional material layer and the electro-optic waveguide, maximizing the efficiency of light absorption and photogenerated carrier generation.

[0076] The fabrication process of the optical computing unit will be described in detail below with reference to the accompanying drawings.

[0077] Figures 6 to 12 This is a schematic cross-sectional view of different steps in the fabrication process of an electro-optic modulator provided in an embodiment of the present invention. The thin-film lithium niobate wafer used is bonded to a silicon dioxide substrate, for example, 2 μm thick. The thickness of the thin-film lithium niobate layer is, for example, 400 nm. The silicon dioxide substrate is grown on a high-resistivity silicon substrate, for example, 575 μm thick. Regarding the fabrication process of the electro-optic modulator 4, as follows... Figure 6As shown, corresponding to modulation region 2, a full-layer electro-optic material waveguide layer 13, such as a lithium niobate waveguide layer, is formed on substrate 9, for example, a silicon dioxide substrate 9. This is achieved through electron beam lithography and ridge waveguide etching processes. Figure 7 The structure shown is formed using electron beam lithography and planar waveguide etching processes. Figure 8 The structure is shown. Then, a cladding 10, such as silicon dioxide, is deposited on the waveguide. Figure 9 The structure shown has a slot cut at the location of the traveling wave electrode 6 to remove the waveguide cladding 10 in the corresponding region, forming... Figure 10 The structure shown is formed by photolithography and evaporation in the corresponding region, creating a traveling wave electrode 6. Figure 11 The structure is shown. An end-face coupling layer 14, such as silicon oxynitride, is deposited at the locations corresponding to the beam splitting and combining units, and etched to form the desired shape. Then, it is diced and polished to form… Figure 12 The structure shown.

[0078] Figures 13 to 16 This is a top view structural diagram of different steps in the fabrication process of a two-dimensional material layer provided in an embodiment of the present invention. Regarding the fabrication process of the two-dimensional material layer 5, after the electro-optic modulator 4 is fabricated, the corresponding detection region 3 is calculated. At this time, the electro-optic material waveguide 1 is covered with a waveguide cladding 10, forming... Figure 13 The structure shown is formed by processing the waveguide cladding 10 in the region where the two-dimensional material layer 5 is located using photolithography and etching processes, exposing the electro-optic material waveguide 1. Figure 14 The structure shown is formed by transferring the two-dimensional material layer 5 to the region where the electro-optic waveguide 1 is exposed using a transfer process. Figure 15 The structure shown is formed by photolithography and evaporation processes in the corresponding regions, creating a first electrode 7 and a second electrode 8. Figure 16 As shown in the diagram, it should be noted that the electrode structure can be located on the side of the two-dimensional material layer 5 away from the substrate 9, or on the side of the two-dimensional material layer 5 adjacent to the substrate 9.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An optical computing unit, characterized in that, include: An electro-optic waveguide, comprising a modulation region and a computational detection region arranged sequentially along the light propagation direction; An electro-optic modulator, comprising the electro-optic material waveguide of the modulation region, is used to receive an input optical signal and modulate the input optical signal according to an applied electrical signal, and output a modulated optical signal; A two-dimensional material layer is disposed in contact with the electro-optical waveguide of the computational detection region to absorb the modulated optical signal and generate photogenerated carriers. An electrode structure is electrically contacted with the two-dimensional material layer to collect the photogenerated carriers and output photocurrent; wherein, the preset structural parameters of the two-dimensional material layer are used to characterize the weighting coefficients of the optical computing unit, and the photocurrent is the result of the modulated optical signal after weighted calculation by the weighting coefficients.

2. The optical computing unit according to claim 1, characterized in that, The preset structural parameters include at least one of the following: the geometric length of the two-dimensional material layer along the propagation direction of the modulated optical signal, the geometric shape of the two-dimensional material layer, the material type of the two-dimensional material layer, and the number of sub-material layers of the two-dimensional material layer.

3. The optical computing unit according to claim 1, characterized in that, The electro-optic modulator includes a traveling wave electrode, and the electrode structure is located in the same metal layer as the traveling wave electrode.

4. The optical computing unit according to any one of claims 1-3, characterized in that, The two-dimensional material layer is a heterojunction structure formed by stacking multiple two-dimensional materials along a vertical direction.

5. The optical computing unit according to any one of claims 1-3, characterized in that, Also includes: An adjustable medium microfluidic channel is disposed in direct or indirect contact with the two-dimensional material layer to change the dielectric environment and / or absorption spectrum of the two-dimensional material layer, thereby adjusting the weighting coefficient of the optical computing unit.

6. The optical computing unit according to any one of claims 1-3, characterized in that, The light absorption characteristics of the two-dimensional material layer are adjusted by applying a bias voltage to the electrode structure, so that the two-dimensional material layer operates in the linear region or the nonlinear region. In the linear region, the photocurrent is linearly related to the light intensity of the modulated light signal to realize the weighted calculation. In the nonlinear region, the photocurrent is nonlinearly related to the light intensity of the modulated light signal to realize the nonlinear activation function.

7. An optical computing array, characterized in that, The system includes multiple optical computing units as described in any one of claims 1-6, wherein the multiple optical computing units are arranged side by side, and the electrode structures of each optical computing unit have a preset electrical connection relationship, and the optical computing array is used to sum the photocurrents output by each optical computing unit.

8. The optical computing array according to claim 7, characterized in that, The electrode structure includes a first electrode and a second electrode. The first electrode of each optical computing unit is electrically connected to the same common node, and the second electrode of each optical computing unit is independently led out. By controlling the bias state of each second electrode, the optical computing array is controlled to work in standard summation mode, differential weighted mode, or sparse gating mode.

9. The optical computing array according to claim 7, characterized in that, In the plurality of optical computing units, at least two of the optical computing units have different preset structural parameters in their two-dimensional material layers to form a weight matrix.

10. A method for fabricating an optical computing unit, characterized in that, The method for fabricating the optical computing unit as described in any one of claims 1-6 comprises: The electro-optic material waveguide is formed on the substrate; The electro-optic modulator is fabricated using the electro-optic material waveguide in the modulation region. The two-dimensional material layer is formed on the electro-optical waveguide in the computational detection region; The electrode structure is formed to make electrical contact with the two-dimensional material layer.

Images