Managing coupling of optical processing stages in system

Abstract

In one aspect, an apparatus includes an integrated circuit device including a first layer including a metal, a second layer including a first semiconductor material, a third layer including an active region of a second semiconductor material, and a fourth layer including a third semiconductor material, where the second layer is between the first layer and the third layer, and the third layer is between the second layer and the fourth layer; an optical interface configured to provide optical waves into different respective portions of the active region along propagation axes substantially parallel to each other, the propagation axes including a first propagation axis and a second propagation axis; a plurality of metal contacts in electrical communication with the fourth layer, wherein a first subset and a second subset of the metal contacts are arranged along the first propagation axis and the second propagation axis; and a power source configured to apply a respective electric field between the first layer and each of the metal contacts.

Description

[0001] Cross-references to related applications

[0002] This application claims priority and benefit to U.S. Provisional Application Serial No. 63 / 629,260 entitled “Active Diffractive Optical Neural Networks”, filed October 3, 2023, the entire disclosure of which is incorporated herein by reference. Technical Field

[0003] This disclosure relates to the optical processing stage in a management system. Background Technology

[0004] In recent years, the field of artificial intelligence (AI) and neural network computing has experienced exponential growth. This field aims to revolutionize various industries, including machine learning, data analytics, and autonomous systems. Processors tailored for AI applications are gaining popularity because they can reduce the time, cost, and power consumption associated with training and operating the large-scale artificial neural networks (ANNs) that form the backend of popular AI services.

[0005] To date, the most popular processor for AI applications has been the graphics processing unit (GPU), which has been adapted from its original application (computer rendering of 3D images) to neural network processing. In some examples, GPUs are useful for AI due to their widespread commercial availability and the similarity in processing requirements between ANNs and image rendering (since both applications can perform massively parallel floating-point computations).

[0006] As engineers build increasingly larger ANNs, the demand for processing power is growing at a rate faster than transistors can keep up. To date, the solution to this problem has been to distribute computation across many chips, thus utilizing large server racks for computing. In some examples, combining chips in this way can be limited by practical aspects such as space, cost, and power.

[0007] To reduce the power, size, and cost of AI processing, some ANN processors can utilize fundamentally different architectures, such as photonic platforms or optical architectures capable of processing and / or manipulating light waves or light. Light can operate at frequencies approximately 10 times higher than typical computer clock frequencies. 5 Operating at frequencies many times higher than those of electrical signals allows light to carry signals with a bandwidth far exceeding that of electrical signals. Some photonic AI processors can use structures such as Mach-Zehnder interferometers (MZIs) or ring resonators to modulate and interfere with light as an alternative to the multiply-accumulate operations associated with ANNs. In some examples, the output of floating-point operations can be encoded in the electric field of light, enabling floating-point operations to be performed at sub-picosecond rates.

[0008] In some examples, the speed advantage associated with photonic processors may be hampered by the size of the photonic elements. The size of some photonic structures may be limited by the wavelength of light passing through them. For example, a 200μm × 20μm photonic structure can be equivalently fitted to approximately 1.2 million transistors fabricated using a 3nm process. Modern GPUs can achieve approximately 400 FLOPS / transistor, so photonic elements must achieve well greater than 480 million FLOPS to provide a chip area advantage. In some implementations, optimizing the design of photonic ANNs may involve increasing the speed of photonic elements and increasing the processing parallelism of the photonic ANN through methods such as wavelength division multiplexing. In some examples, increasing these factors may be associated with increased power consumption and complexity.

[0009] Some photonic processing devices may include semiconductor materials such as silicon or III / V compounds. Examples of III / V compounds include elements from Group III of the periodic table, such as boron, aluminum, gallium, or indium. Examples of III / V compounds include elements from Group V of the periodic table, such as nitrogen, phosphorus, arsenic, or antimony. In some implementations, the semiconductor material may be doped with p-type or n-type dopants. In some implementations, p-type dopants may include elements such as tin, germanium, silicon, tellurium, and sulfur. In some implementations, n-type dopants may include elements such as zinc, cadmium, beryllium, and magnesium.

[0010] Some photonic processing devices may include optical waveguide structures or optical paths configured to guide optical waves in optical wavelength regions of the electromagnetic spectrum. Some electromagnetic waves have a spectrum with peak wavelengths falling within a specific optical wavelength range (e.g., between about 100 nm and about 1 mm, or a subrange thereof), and are also referred to as optical waves, light waves, or simply light. In some implementations, an optical wave may be associated with one or more optical modes or spatial modes. In some implementations, an optical mode may be associated with a structure configured to guide the optical wave. Summary of the Invention

[0011] In one aspect, typically, a device includes: a plurality of optical processing stages configured to process two or more optical waves having a spectral peak wavelength λ, wherein each of the plurality of optical processing stages includes: two or more configurable optical structures substantially coplanar with a plane, wherein each configurable optical structure is configured to receive an optical wave propagating along a first axis substantially parallel to the plane, and each configurable optical structure includes an active region having: a width along a second axis substantially parallel to the plane and perpendicular to the first axis, wherein the width is less than or equal to 2. λ, a height along a third axis substantially perpendicular to the plane and perpendicular to the first axis, wherein the height is greater than λ / 10, and a length along the first axis less than or equal to 100λ; and an interface region configured to receive optical waves from each of the two or more configurable optical structures; wherein each interface region associated with a corresponding optical processing stage in at least two of the plurality of optical processing stages is configured to couple at least a portion of the optical waves received from at least one configurable optical structure to at least two configurable optical structures in a subsequent optical processing stage.

[0012] Each aspect may include one or more of the following features.

[0013] Each active region is configured to guide up to four spatial modes associated with optical waves.

[0014] Each active region is configured to contain a corresponding percentage of electromagnetic power associated with optical waves propagating through the respective active region, which is greater than 50% relative to the total electromagnetic power associated with optical waves propagating through a configurable optical structure including the respective active region.

[0015] Each active region is configured to contain a percentage of electromagnetic power associated with optical waves propagating through the respective active region, which is greater than 70% relative to the total electromagnetic power associated with optical waves propagating through the configurable optical structure including the respective active region.

[0016] At least a portion of each active region of the corresponding optical processing stage is separated from at least a portion of one or more other active regions of the corresponding optical processing stage by a portion of a region including insulating material.

[0017] Each region, including the insulating material, extends below the respective surface of each adjacent active region along a corresponding axis that is substantially perpendicular to the plane and parallel to the third axis.

[0018] At least a portion of each active region of the corresponding optical processing stage is separated from at least a portion of one or more other active regions of the corresponding optical processing stage through a corresponding air insulation gap.

[0019] Each interface region includes a planar waveguide structure formed within a substrate, wherein the planar waveguide structure is coupled to a plurality of configurable optical structures at a first end and at a second end opposite to the first end.

[0020] Each of two or more configurable optical structures is configured to provide an intensity variation of an optical wave propagating through a respective respective active region, wherein one or more of the intensity variations provides optical gain to the optical wave.

[0021] Each configurable optical structure is configured such that the optical gain provided to the optical wave is nonlinear with respect to the intensity of the optical wave.

[0022] At least one configurable optical structure can be configured to transmit at least a portion of one or more optical waves in a first operating mode, and can be configured to detect the intensity of the optical waves in a second operating mode.

[0023] Each active region of the configurable optical structure includes a first semiconductor material.

[0024] Each configurable optical structure further includes: a first layer comprising a first semiconductor material, a second layer comprising the active region, wherein the active region comprises a second semiconductor material, and a third layer comprising a third semiconductor material, wherein the second layer is located between the first layer and the third layer.

[0025] The second layer also includes a fourth layer and a fifth layer. The fourth layer includes a fourth semiconductor material, which is located between the first layer and the active region. The fifth layer includes a fourth semiconductor material, which is located between the third layer and the active region.

[0026] The fourth semiconductor material includes a composition of indium gallium arsenide phosphide.

[0027] A portion of the active region includes a quantum well.

[0028] A portion of the active region includes bulk semiconductor material.

[0029] The first layer includes a first semiconductor material internally mixed with dopants, and the third layer includes a third semiconductor material internally mixed with dopants.

[0030] (1) The dopant of the first layer includes a p-type dopant and the dopant of the third layer includes an n-type dopant, or (2) The dopant of the first layer includes an n-type dopant and the dopant of the third layer includes a p-type dopant.

[0031] The first semiconductor material and the third semiconductor material each comprise a composition of indium gallium arsenide phosphide.

[0032] The second semiconductor material includes a composition of indium gallium arsenide phosphide.

[0033] The corresponding optical waves are provided to each configurable optical structure of the optical processing stage in the plurality of optical processing stages.

[0034] The corresponding optical waves are provided to each configurable optical structure by the corresponding modulator.

[0035] At least the first active region is configured to control the optical properties associated with optical waves propagating through the first active region.

[0036] The optical property that the first active region is configured to control is the optical power associated with the optical wave propagating through the first active region.

[0037] The first active region is configured to increase the optical power associated with the optical waves propagating through the first active region.

[0038] The first active region comprises a semiconductor material, wherein the band gap energy of the semiconductor material is lower than the wavelength of the spectral peaks of the two or more optical waves. Related energy.

[0039] Semiconductor materials are direct bandgap semiconductor materials.

[0040] The first active region is configured to control optical properties associated with optical waves propagating through the first active region, at least in part, based on electro-optic or thermo-optic effects.

[0041] Each active region includes a material configured to control optical properties associated with optical waves traveling through the respective active region, at least in part, based on a nonlinear magnetic susceptibility associated with the material.

[0042] Each active region is configured to control the optical properties associated with the optical wave through one or more of the following electro-optic effects: (1) Franz-Keldish effect, (2) quantum confinement Stark effect, (3) Pockels effect, (4) plasma dispersion effect and (5) Kerr effect.

[0043] Each interface area is basically composed of passive materials.

[0044] On the other hand, typically, a device includes: an integrated circuit device comprising: a first layer comprising metal, a second layer comprising a first semiconductor material, a third layer comprising an active region comprising a second semiconductor material, and a fourth layer comprising a third semiconductor material, wherein the second layer is between the first layer and the third layer, and the third layer is between the second layer and the fourth layer; an optical interface configured to provide two or more optical waves to different corresponding portions of the active region along different corresponding propagation axes that are substantially parallel to each other, the different corresponding propagation axes including at least a first propagation axis and a second propagation axis; a plurality of metal contacts electrically in communication with the fourth layer, wherein a first subset of the metal contacts is arranged along the first propagation axis, and a second subset of the metal contacts is arranged along the second propagation axis; and a power supply configured to apply a corresponding electric field between the first layer and each of the plurality of metal contacts.

[0045] Each aspect may include one or more of the following features.

[0046] The first layer and the second layer are electrically connected.

[0047] The second layer further includes a dopant mixed in the first semiconductor material, and the fourth layer further includes a dopant mixed in the third semiconductor material.

[0048] (1) The dopant of the second layer includes an n-type dopant and the dopant of the fourth layer includes a p-type dopant, or (2) The dopant of the second layer includes a p-type dopant and the dopant of the fourth layer includes an n-type dopant.

[0049] The first semiconductor material and the third semiconductor material include indium phosphide.

[0050] The second semiconductor material includes a composition of indium gallium arsenide phosphide.

[0051] The fourth layer also includes multiple regions of a third semiconductor material internally mixed with dopants, wherein each of the multiple metal contacts is electrically connected to at least a portion of a different corresponding region within the multiple regions.

[0052] At least a portion of each of the multiple regions is separated from at least a portion of each of the other regions of the multiple regions by means of a portion of the third semiconductor material that is not doped or by means of a region that is not a third semiconductor material.

[0053] One or more regions without a third material include electrically insulating or optically transparent materials.

[0054] Each portion of the active region between the metal contacts of the plurality of metal contacts and the first layer is configured to provide an intensity variation of an optical wave propagating through a corresponding portion of the active region, at least in part based on a corresponding electric field applied between the metal contacts of the first layer and the plurality of metal contacts, wherein one or more of the intensity variations provide optical gain to the optical wave.

[0055] Each portion of the active region is configured such that the optical gain provided to the optical wave is nonlinear with respect to the intensity of the optical wave.

[0056] The third layer further includes: a fifth layer comprising a fourth semiconductor material, wherein the fifth layer is located between the second layer and the active region; and a sixth layer comprising the fourth semiconductor material, wherein the sixth layer is located between the fourth layer and the active region.

[0057] The fourth semiconductor material includes a composition of indium gallium arsenide phosphide.

[0058] A portion of the active region includes a quantum well.

[0059] A portion of the active region includes bulk semiconductor material.

[0060] On the other hand, typically, a device includes: a configurable optical structure configured to receive an input optical wave propagating along a first axis, wherein the input optical wave has a spatial profile distributed along a second axis perpendicular to the first axis and a third axis perpendicular to both the first and second axes; a first optical pump beam interface configured to provide a first set of two or more optical pump beams, wherein each optical pump beam in the first set of two or more optical pump beams is guided into the configurable optical structure along a corresponding axis perpendicular to the first axis and parallel to the second axis; and a second optical pump beam interface configured to provide a second set of two or more optical pump beams, wherein each optical pump beam in the second set of two or more optical pump beams is guided into the configurable optical structure along a corresponding axis perpendicular to the first axis and parallel to the third axis.

[0061] Each aspect may include one or more of the following features.

[0062] The configurable optical structure includes the region at the intersection of an optical pump beam from a first set of two or more optical pump beams and an optical pump beam from a second set of two or more optical pump beams.

[0063] The configurable optical structure is configured to control the intensity of the optical wave propagating through the region by providing optical gain to the input optical wave propagating through the region.

[0064] The region is configured such that the optical gain provided to the optical wave is non-linear with respect to the intensity of the optical wave.

[0065] The configurable optical structure includes multiple regions, each of which is located at the intersection of optical pump beams from a first set of two or more optical pump beams and optical pump beams from a second set of two or more optical pump beams.

[0066] Configurable optical structures include laser crystals or glasses doped with rare earth elements.

[0067] The input optical wave is a collimated light source.

[0068] Each optical pump beam in a first set of two or more optical pump beams includes an optical wave having a first wavelength, and each optical pump beam in a second set of two or more optical pump beams includes an optical wave having a second wavelength, wherein the first wavelength and the second wavelength are determined at least in part based on the material of the configurable optical structure.

[0069] Each aspect may possess one or more of the following advantages.

[0070] Some of the implementations disclosed in this paper can be used to develop artificial neural networks that can be configured electronically or optically.

[0071] Other features and advantages will become apparent from the following description, the accompanying drawings, and the claims. Attached Figure Description

[0072] This disclosure can be best understood from the following detailed description when read in conjunction with the accompanying drawings. It should be emphasized that, by convention, the various features in the drawings are not to scale. On the other hand, for clarity, the dimensions of the various features have been arbitrarily enlarged or reduced.

[0073] Figure 1A This is a schematic diagram of the example device.

[0074] Figure 1B This is a schematic diagram of an example optical processor.

[0075] Figures 2A-2B This is a schematic diagram of an example optical processor.

[0076] Figure 3 This is a schematic diagram of an example optical processor.

[0077] Figures 4A-4BIt is a prediction curve from a numerical simulation associated with the gain medium.

[0078] Figures 5A-5G This is a schematic diagram of an example optical processor.

[0079] Figures 6A-6B This is a schematic diagram of an example optical processor.

[0080] Figures 7A-7B This is a schematic diagram of an example optical processor.

[0081] Figure 7C This is a schematic diagram of an example energy level diagram.

[0082] Figures 8A-8F This is a schematic diagram of the example device.

[0083] Figure 9 This is a schematic diagram of the example device.

[0084] Figures 10A-10B This is a schematic diagram of an example optical processor. Detailed Implementation

[0085] Some ANNs are computational models that include layers of interconnected nodes or neurons. Some ANNs include multiple layers of nodes, such as input layers, hidden layers, and output layers. Some input layers can be configured to receive data and distribute data to neurons. Some hidden layers can perform computations on the input data. In some examples, this computation may include receiving input from neurons in the previous layer and applying some mathematical operations. Some mathematical operations may include applying weights to determine the effect of the input on the output. Some output layers can produce a final prediction or result. In some examples, this prediction may be associated with a probability distribution of multiple outcomes, where each outcome is associated with a corresponding weight. In some examples, the data associated with the various layers may be represented by vectors.

[0086] Some implementations of artificial neural networks can include optical components configured to simulate neural networks. Optical neural networks can be fabricated on photonic integrated circuits that allow light to propagate within planar waveguides. Some planar waveguides can include diffraction elements that can perturb the propagation of light, thus acting as substitutes for neurons within the artificial neural network. In some implementations, the elements can be physically separated, such that physical separation between elements phase-coupled by the diffraction elements can be used as a substitute for weights within a conventional neural network. Light can be detected at the end of the planar waveguide to determine the output of the computation of the photonic neural network. Such photonic neural networks can be used to address the challenges associated with photonic computing, namely, eliminating discrete photonic elements, thereby reducing the physical size of the computational units.

[0087] Without using the methods disclosed herein, some diffractive optical neural networks can have (1) lithographically defined neurons that can solve computations; and (2) linear diffractive propagation that can limit the effective depth of the neural network to a single layer. In some implementations, reprogrammable diffractive optical neural networks can simulate the nonlinear activation functions associated with conventional neural networks.

[0088] Some diffractive optical neural networks (DONs) include photonic elements designed to simulate input, hidden, and output layers of artificial neural networks. In some implementations, DONs can be constructed from materials exhibiting optical gain to enable reconfigurable photonic artificial intelligence (AI) processors. In some examples, DONs can include multimode media, such as planar mode waveguides or free space. In some examples, the multimode medium can be two-dimensional (2D) or three-dimensional (3D). Allowing light to propagate in a multimode medium overcomes a challenge associated with photonic computing: the size of photonic elements compared to the size of a transistor. In some implementations, incorporating optical gain into the propagation medium allows the weights of the DON to be reprogrammed after fabrication for training or operation. Incorporating gain also allows for photonic simulation of nonlinear neuron activation functions within the propagation medium. In some implementations, this simulation can allow for representations of artificial neural networks with more than one hidden layer. Some optical architectures can include reconfigurable or active elements and can be configured as DONs. Some DONs can be referred to as active diffractive optical neural networks (ADONNs). Optical architectures can be fabricated in 2D on photonic integrated circuits. In some implementations, the optical architecture can also be extended to 3D structures. In such implementations, the use of 3D structures can be associated with increased input vector size, output vector size, and computational power compared to other examples.

[0089] Figure 1AAn example system 100A, which can be implemented as an artificial neural network, is depicted. System 100A includes one or more light sources 102 capable of generating optical waves or photons, and one or more modulators 104 receiving optical photons from one or more light sources 102. The photons can be grouped into an input array to processor 106 and can represent numerical vector inputs to the neural network. In other words, in some examples, one or more light sources 102 and one or more modulators 104 can be associated with the input layer of the artificial neural network. One or more modulators 104 can encode signals at each vector input as properties of the electric field of light, such as amplitude. The optical vector inputs are injected into the optical processor 106. A pumping unit 108 spatially controls the pumping within the optical processor 106. This spatial pumping can be associated with one or more hidden layers of the neural network. Multiple detectors 110 detect light at the output of the optical processor 106, and signals from each detector can be decoded into vector outputs of the neural network. Detectors 110 can be associated with the output layer of the neural network. In some implementations, electronic circuit 112 can be configured as a component of control system 100A.

[0090] Figure 1B An example implementation 100B of an artificial neural network is depicted. Example implementation 100B includes multiple input beams 150 and multiple output beams 152. In some examples, the multiple input beams 150 may represent vectors of input to the neural network, where the inputs may be encoded in light intensity. In some examples, the multiple input beams 152 may represent vectors of output from the neural network, where the outputs may be encoded in light intensity. A gain medium 154 is configured to spatially vary the attenuation or gain associated with light propagating through the gain medium 154. By way of example, the gain medium 154 contains a numerically simulated prediction plot of the light propagating through the gain medium 154.

[0091] In some examples, the propagation time of light from the input to the output of the neural network can be set as a delay for the processor. In some implementations, one or more optical architectures configured as ADONN can be cascaded to increase the number of layers, inputs, and / or outputs of the neural network.

[0092] Figure 2A A top view depicts an example optical architecture 200 that can be used as an optical processor. The optical architecture 200 includes components configured to process wavelengths with spectral peaks. The system comprises two optical processing stages 202A and 202B for two or more optical waves. The first optical processing stage 202A includes two configurable optical structures 204A and 204B that are substantially coplanar with a plane (in this example, the xy-plane). Each configurable optical structure 204A and 204B is configured to receive an optical wave propagating along a first axis (in this example, the x-axis) substantially parallel to the plane. In this example, each configurable optical structure 204A and 204B receives an optical wave from an optical interface 206. The configurable optical structures 204A and 204B are separated by insulating regions 208A-208C. The first optical processing stage 202A also includes an interface region 210 configured to receive optical waves from each configurable optical structure 204A and 204B. The second optical processing stage 202B includes two configurable optical structures 212A and 212B separated by insulating regions 214A-214C. The second optical processing stage 202B also includes an interface region 216 configured to receive optical waves from each of the configurable optical structures 212A, 212B. The interface region 210 associated with the first optical processing stage 202A is configured to couple at least a portion of the optical waves received from at least one configurable optical structure 204A, 204B to at least two of the configurable optical structures 212A, 212B in the subsequent optical processing stage 202B. By way of example, portions of optical waves 213A, 213B are each coupled from configurable optical structure 204A to configurable optical structures 212A, 212B. In some implementations, the insulating region adjacent to each active region may include an optically transparent material and / or an electrically insulating material. In some implementations, the active regions may be separated by an insulating region including an air-insulating gap. This air-insulating gap may be formed, for example, by etching into the substrate forming the active regions.

[0093] Each configurable optical structure 204A, 204B and 212A, 212B may include an active region. Figure 2BA two-dimensional perspective view of the optical architecture 200 along plane 218 is depicted. The optical architecture 200 includes a substrate material 220 forming a layer and a shared bottom-side contact 222 forming another layer. The optical architecture 200 also includes a top-side contact 224 associated with a configurable optical structure 204B. In some implementations, the respective configurable optical structures 204A, 204B, 212A, 212B may be associated with different corresponding top-side contacts. The configurable optical structure 204B includes layers 226, 228, and 230. In some examples, layer 226 may include an active region. In some implementations, layer 228 may include a p-type dopant and layer 230 may include an n-type dopant. In some implementations, layer 228 may include an n-type dopant and layer 230 may include a p-type dopant. In some implementations, the respective insulating regions 208B, 208C may extend below layer 226 containing the active region, such that the active region is “deeply etched” as... Figure 2B As shown, in some implementations, insulating regions 208B and 208C may extend below the corresponding surfaces of adjacent active regions of layer 226.

[0094] In some implementations, the active region of layer 226 can be configured to guide one or more modes associated with optical waves. In some implementations, the active region of layer 226 can have a second axis that is substantially parallel to the plane and perpendicular to the first axis, having a shape less than or equal to... The width. In this example, the second axis is the y-axis. The active region of layer 226 can have a width greater than the first axis along a third axis that is substantially perpendicular to the plane and perpendicular to the first axis. The height. In this example, the third axis is the z-axis. The active region of layer 226 can have a height less than or equal to the first axis, which is substantially parallel to the plane. The length is parallel to the first axis. In some implementations, customization of the size of the active region of layer 226 can allow up to four modes associated with the optical waves guided by the active region 226.

[0095] In some implementations, tuning the geometry of the active region can increase the optical confinement factor associated with guiding optical waves through the active region. The optical confinement factor can describe the amount or percentage of electromagnetic power contained within the active region. Some active regions can be configured such that the percentage of electromagnetic power associated with optical waves or optical modes propagating through the active region relative to the total electromagnetic power associated with optical waves propagating through configurable optical structures including the respective active regions is greater than 50% or greater than 70%. In some examples, tuning the geometry of the active region can provide an increased gain associated with optical waves propagating through the active region.

[0096] In some implementations, layer 226 may include a fourth and a fifth layer, the fourth layer comprising a fourth semiconductor material located between the first layer and the active region, and the fifth layer comprising a fourth semiconductor material located between the third layer and the active region. In some examples, a portion of the active region of layer 226 may include a quantum well or a bulk semiconductor material.

[0097] Figure 3 An example optical architecture 300 that can be used as an optical processor is depicted. The optical architecture 300 includes multiple optical processing stages 302A-302N. Each optical processing stage 302A-302N is configured to process wavelengths with spectral peaks. Two or more optical waves. Optical processing stages 302A-302N each include two or more configurable optical structures 304Ai-304Ni, where i is the number of configurable optical structures within each optical processing stage 302A-302N. Each configurable optical structure 304Ai-304Ni is separated by insulating regions 306Aj-306Nj, where j is the number of insulating regions within each optical processing stage 302A-302N. Each optical processing stage 302A-302N also includes interface regions 308A-308N, configured to couple at least a portion of the optical wave received from at least one configurable optical structure 304Ai-304Ni to at least two configurable optical structures 304Ai-304Ni in subsequent optical processing stages 302A-302N. A first optical processing stage 310 is configured to receive optical waves from coupler 310.

[0098] In some implementations, one or more of the interface regions 308A-308N may be without an active region, making the interface regions 308A-308N passive, and the intensity of optical waves propagating through the respective interface regions 308A-308N is not controlled according to any control signal. Instead, the intensity (or other characteristics, such as phase) may experience uncontrolled effects from simple propagation through the material, such as slight intensity reduction due to inherent material losses.

[0099] In some implementations, the configurable optical structure can be configured to provide intensity variations of optical waves propagating through respective respective active regions, with one or more intensity variations providing optical gain to the optical waves. In some implementations, the optical gain provided by the respective active regions to the optical waves can be nonlinear in terms of the intensity of the optical waves.

[0100] In some implementations, active regions can be configured to control optical properties associated with optical waves propagating through the respective active region. For example, active regions can control optical properties such as optical power or intensity through processes such as stimulated emission or absorption. Some active regions can be configured to increase or decrease the intensity associated with optical waves propagating through the active region. Other optical properties that can be controlled by active regions include optical phase, optical wavelength, and polarization. Some active regions can also be configured to convert power from one optical wavelength or mode to another via nonlinear conversion.

[0101] Some active regions can have their optical properties controlled, at least in part, based on control signals. Examples of control signals that can be associated with an active region include optical or electrical pumping, mechanical deformation, and temperature. Some examples of optical pumping can be associated with optical power applied to the active region. In some examples, electrical pumping can be associated with electrical power, current, or voltage applied to the active region.

[0102] Some active regions can have their optical properties controlled, at least in part, based on an applied electric field characterized as an electro-optic effect. Examples of electro-optic effects include: (1) the Franz-Keldysh effect, (2) the quantum-confined Stark effect, (3) the Pockels effect, (4) the plasmon dispersion effect, or (5) the Kerr effect. In some examples, the electro-optic effect can be associated with nonlinear properties of the material, such as nonlinear optical susceptibility. For example, the Pockels effect (sometimes called the linear electro-optic effect) can be associated with the material's χ² value. (2) Electromagnetic susceptibility is related. The Kerr effect (sometimes called the secondary electro-optic effect) can be related to the material's χ² value. (3) Electromagnetic susceptibility is relevant. In some examples, nonlinear materials can control optical properties associated with optical modes, such as optical phase, at least in part based on an external electric field via electro-optic effects (such as the Kerr effect). Some active regions can control optical properties of optical modes at least in part based on applied temperature modulation via thermo-optic effects or at least in part based on applied sound waves via acousto-optic effects.

[0103] Some active regions can be configured to control the optical power or intensity associated with optical waves using properties associated with the semiconductor material. Some active regions may comprise semiconductor materials with bandgap energies lower than the photon energy of optical waves propagating through the active region. Some semiconductor materials have band structures associated with electrons in the semiconductor. In some examples, the band structure can be expressed as a function of the crystal momentum associated with electrons in the lattice. In some examples, the semiconductor material may have a direct bandgap, where electrons in the valence band can be excited to the conduction band without changing the crystal momentum.

[0104] In some implementations, the configurable optical structure can be configured to transmit at least a portion of one or more optical waves in a first operating mode and to detect the intensity of the optical waves in a second operating mode. In some implementations, configuring the configurable optical structure to detect the intensity of the optical waves simplifies the optical architecture design because a photodetector can be omitted. In some implementations, the configurable optical structure configured to detect intensity can provide feedback during neural network training, calibration, operation, or processing.

[0105] In some implementations, the gain or absorption at various locations within an optical architecture configured as an optical processor can represent weights associated with hidden layers of a neural network, and the propagation of light can serve as both an interconnection between neurons and a mechanism for computation. In some implementations, the optical processor may include a gain medium. Some gain media can be strong absorbers at a designed gain wavelength when not pumped. As the pumping force increases, more and more carriers can be excited from the ground state to higher energy levels. This excitation can simultaneously reduce the number of carriers available for stimulated recombination and increase the number of carriers available for stimulated emission. In some examples, the excitation can ultimately make the material transparent and then act as an amplifier at a sufficiently strong pumping level. Some gain media can be electrically pumped, and some can be optically pumped. In some electrically pumpable materials, the current density can be varied across the propagation medium, and in optically pumpable materials, the force of the pump light can be varied across the material to change the absorption / gain coefficient of the propagation medium.

[0106] In some implementations, the intensity of the electric or optical pump can be controlled quickly and digitally. This control allows ADONN to electronically reconfigure the weights of the neural network. Network reconfiguration is useful for both training the neural network and changing its computational function.

[0107] In some implementations, gain materials can be used as nonlinear activation functions. At high throughput, some gain materials can exhibit saturation effects in both the absorption and gain modes. In the absorption mode, this effect can be called saturable absorption, while in the gain mode, it can be called gain saturation. Figure 4A Figure 400A depicts the predicted curves of the transmitted light as a function of the electric field in the gain mode, based on numerical simulations. Figure 4B Figure 400B depicts the predicted curves of the transmitted light as a function of the electric field in the absorption mode, based on numerical simulations. From the perspective of the region serving as the gain medium for neurons in ADONN, gain saturation can lead to activation roll-off for both strong positive and negative values, which can be analogous to the hyperbolic tangent activation function.

[0108] In some implementations, the detector can also be used as a nonlinear activation function. Although propagation in the medium can manipulate the electric field of light, photodiodes and other optical detectors can generate a signal proportional to the power of the light, which can be no less than zero. This signal may result in a nonlinearity similar to the rectified linear unit (ReLU) activation function.

[0109] In some implementations, the input modulator can be eliminated, and instead, the intensity of the input signal can be encoded using a pump force applied at the beginning of the gain medium. This encoding eliminates the need for discrete light sources or modulators. Alternatively, individual light sources can be distributed across an array.

[0110] In some implementations, attenuation-only materials can be used as an alternative to the gain medium. For example, a planar mode PN junction can be used. The current density can be varied across the planar waveguide to spatially modulate the absorption coefficient of the propagation medium. This implementation offers the advantage that the propagation medium can be defined locally within the silicon photonics process. In some examples, sufficient optical power can be provided at the network input to overcome losses that can accumulate in the planar mode waveguide.

[0111] In some implementations, the optical architecture can be configured to provide local gain control of the optical gain medium, including planar waveguides. Figure 5A A top view of an optical architecture 500A, which can be used as an optical processor, is depicted. The optical architecture 500A includes a plurality of metal contacts 502Aa-502Nn arranged above a surface 504. The optical architecture 500A also includes an optical interface 508 configured to provide two or more optical waves 509A, 509B to different corresponding portions of an active region along different, substantially parallel, respective propagation axes (including at least a first propagation axis and a second propagation axis). The plurality of metal contacts 502Aa-502Nn includes a first subset 502Aa-502Na of metal contacts arranged along the first propagation axis and a second subset 502An-502Nn of metal contacts arranged along the second propagation axis.

[0112] In some examples, the optical architecture may include one or more layers configured to provide gain control.

[0113] Figure 5BAn example two-dimensional perspective view of an optical architecture 500A along plane 507 is depicted. In this example, the optical architecture 500A includes: a first layer 570 comprising metal, a second layer 572 comprising a first semiconductor material, a third layer 574 comprising an active region comprising a second semiconductor material, and a fourth layer 576 comprising regions 580A-580N in electrical contact with a plurality of metal contacts 502An-502Nn. Regions 580A-580N are separated from regions 582A-582N. The plurality of metal contacts 502An-502Nn are arranged on the fourth layer 576. An optical interface 508 is configured to receive an optical wave 509B propagating 578 through the third layer 574. In some examples, the optical interface 508 may include a waveguide junction or a doped glass.

[0114] Figure 5C An example two-dimensional perspective view along plane 506 depicts an optical architecture 500A. In this example, the optical architecture 500A includes: a first layer 510 comprising metal, a second layer 512 comprising a first semiconductor material, a third layer 514 comprising an active region comprising a second semiconductor material, and a fourth layer 515 comprising regions 516A-516C separated by regions 518A and 518B. In this example, the third layer 514 is the active region of the second semiconductor material. A plurality of metal contacts 502Aa-502Ac are arranged on the fourth layer 515. The fourth layer 515 comprises the third semiconductor material. Each metal contact 502Aa-502Ac is associated with a region 516A-516C. In some implementations, regions 516A-516C may include a third semiconductor material internally mixed with dopants, and regions 518A and 518B may include a third semiconductor material internally unmixed with dopants. In some examples, the dopant may be a p-type dopant. In some implementations, the second layer 512 may include a first semiconductor material internally mixed with dopants. In some examples, the dopant may be n-type. The second layer 512 is between the first layer 510 and the third layer 514, and the third layer 514 is between the second layer 512 and the fourth layer 515.

[0115] Figure 5DAn example two-dimensional perspective view along plane 506 depicts an optical architecture 500A. In this example, the optical architecture 500A includes: a first layer 520 comprising metal, a second layer 522 comprising a first semiconductor material, a third layer 524 comprising an active region comprising a second semiconductor material, and a fourth layer 525 comprising a third semiconductor material. The fourth layer 525 includes regions 526A-526C separated by regions 528A, 528B that do not contain the third semiconductor material. In this example, the third layer 524 is the active region of the second semiconductor material. A plurality of metal contacts 502Aa-502Ac are arranged on the fourth layer 525. Each metal contact 502Aa-502Ac is associated with regions 526A-526C. In some implementations, regions 526A-526C may include a third semiconductor material internally mixed with dopants. In some examples, the dopant may be a p-type dopant. In some implementations, the second layer 522 may include a first semiconductor material internally mixed with dopants. In some examples, the dopant may be n-type.

[0116] Figure 5E An example two-dimensional perspective view along plane 506 depicts an optical architecture 500A. In this example, the optical architecture 500A includes: a first layer 530 comprising metal, a second layer 532 comprising a first semiconductor material, a third layer 534 comprising active regions comprising a second semiconductor material, and a fourth layer 535 comprising a third semiconductor material. The fourth layer 535 includes regions 536A-536C separated by regions 538A and 538B comprising the third semiconductor material and regions 539A and 539B without the third semiconductor material. In this example, the third layer 534 is the active region of the second semiconductor material. A plurality of metal contacts 502Aa-502Ac are arranged on the fourth layer 535. Each metal contact 502Aa-502Ac is associated with regions 536A-536C. In some implementations, regions 536A-536C and regions 538A and 538B may include a third semiconductor material internally mixed with dopants. In some examples, the dopant may be a p-type dopant. In some implementations, the second layer 532 may include a first semiconductor material internally mixed with dopants. In some examples, the dopant may be n-type.

[0117] like Figures 5C-5E As shown, at least a portion of each of the multiple regions in the fourth layer is separated from at least a portion of each of the other regions by either a dopant-free portion of the third semiconductor material or a region without the third semiconductor material. In some implementations, the regions without the third semiconductor material may comprise electrically insulating or optically transparent materials. Some examples of electrically insulating or optically transparent materials include silicon dioxide (SiO2), silicon nitride (Si3N4), and benzocyclobutene (BCB).

[0118] In some implementations, the third layer may include a fifth and a sixth layer containing a fourth semiconductor material. Figure 5F An example two-dimensional perspective view along plane 506 depicts an optical architecture 500A. In this example, the optical architecture 500A includes: a first layer 540 comprising metal, a second layer 542 comprising a first semiconductor material, a third layer 544 comprising an active region 548 comprising a second semiconductor material, and a fourth layer 545 comprising regions 546A-546C separated by regions 543A and 542B without a third semiconductor material. The third layer 544 also includes a fifth layer 547 and a sixth layer 549 comprising a fourth semiconductor. A plurality of metal contacts 502Aa-502Ac are arranged on the fourth layer 545. The fourth layer 545 comprises the third semiconductor material. Each metal contact 502Aa-502Ac is associated with a region 546A-546C. In some implementations, regions 546A-546C may comprise a third semiconductor material internally mixed with dopants. In some examples, the dopant may be a p-type dopant. In some implementations, the second layer 542 may comprise a first semiconductor material internally mixed with dopants. In some examples, the dopant can be n-type.

[0119] Figure 5G An example two-dimensional perspective view along plane 506 depicts an optical architecture 500A. In this example, the optical architecture 500A includes: a first layer 550 comprising metal, a second layer 552 comprising a first semiconductor material, a third layer 554 comprising an active region 558 comprising a second semiconductor material, and a fourth layer 555 comprising regions 556A-556C separated by regions 553A and 553B. The third layer 554 also includes a fifth layer 557 and a sixth layer 559 comprising a fourth semiconductor material. A plurality of metal contacts 502Aa-502Ac are arranged on the fourth layer 555. The fourth layer 555 comprises the third semiconductor material. Each metal contact 502Aa-502Ac is associated with a region 546A-546C. In some implementations, regions 546A-546C may comprise a third semiconductor material internally mixed with dopants. In some examples, the dopant may be a p-type dopant. In some implementations, the second layer 552 may comprise a first semiconductor material internally mixed with dopants. In some examples, the dopant can be n-type.

[0120] In some implementations, the active region 558 may be thin, such that the interaction between the semiconductor material of the active region 558 and the light propagating through the active region 558 can be characterized by quantum behavior. In some implementations, these quantum behaviors may be referred to as quantum wells. Some optical architectures may include multiple thin layers of the active region, each layer being separated by a semiconductor material to form multiple quantum wells. In some implementations, tuning the composition of the individual layers, as described below, can tune the physical or material properties associated with the optical architecture. In some implementations, the active region 558 may be thick, such that the material or optical properties of the active region 558 are associated with bulk material properties. Some active regions may include quantum dots or semiconductor nanocrystals.

[0121] In some implementations, the first and third semiconductor materials may include indium phosphide, and the second semiconductor material may include a composition of indium gallium arsenide phosphide. In some implementations, the fourth semiconductor material may include a composition of indium gallium arsenide phosphide. In some examples, the second and fourth semiconductor materials may include different compositions of indium gallium arsenide phosphide. Some compositions of indium gallium arsenide phosphide may omit or omit elements. In some implementations, changing the ratio of elements in the composition can change the band gap or lattice constant associated with the semiconductor material and tune the optical properties associated with the semiconductor material.

[0122] In some implementations, Figures 5B-5G Each of the second layers can include a first semiconductor material internally mixed with dopants. In some implementations, the dopant can be p-type. In some implementations, Figures 5B-5G Each of the fourth layers can include a third semiconductor material internally mixed with dopants. In some implementations, the dopants can be n-type.

[0123] like Figures 5A-5G As shown, each of the plurality of metal contacts can be arranged on the surface of the fourth layer. In some examples, each of the plurality of metal contacts can be electrically connected to or in contact with the fourth layer or a portion thereof. In some implementations, the metal layer can be electrically connected to another layer comprising a semiconductor. In some implementations, the metal layer electrically connected to the semiconductor layer can be in direct contact with the semiconductor layer. In some implementations, the metal layer electrically connected to the semiconductor layer can have a separation layer between the metal layer and the semiconductor layer. For example, a thin ohmic contact layer can be placed between the fourth layer and the plurality of metal contacts. The thin ohmic contact layer can also be placed between a first layer comprising metal and a second layer comprising a semiconductor.

[0124] In some implementations, the power supply can be configured to apply an electric field to the optical processor. Some power supplies may include voltage sources, while others may include current sources. In some implementations, a corresponding electric field can be applied between each of the multiple metal contacts and the first layer comprising the metal. In some implementations, the electric field may include alternating current (AC) or direct current (DC).

[0125] In some implementations, the optical architecture may include an optical pump gain medium configured as an optical processor. Figure 6A An isometric view of an example optical architecture 600 that can be used as an optical processor is depicted. The optical architecture 600 includes a configurable optical structure 602. The configurable optical structure 602 receives an input optical wave 604 propagating along a first axis (x-axis in this example). In some examples, the input optical wave 604 may have a spatial profile distributed along a second axis perpendicular to the first axis and a third axis perpendicular to both the first and second axes. In this example, the second axis is the y-axis, and the third axis is the z-axis. A first optical pump beam interface 610 is configured to provide a first set of two or more optical pump beams 606A-606N guided along a respective axis perpendicular to the first axis and parallel to the second axis into the configurable optical structure 602. A second optical beam interface 612 is configured to provide a second set of two or more optical pump beams 608A-608N guided along a respective axis perpendicular to the first axis and parallel to the third axis into the configurable optical structure 602.

[0126] Figure 6B An example two-dimensional perspective view along the yz plane is depicted for the optical architecture 600. The configurable optical structure 602 includes a region 614 at the intersection of an optical pump beam 606A from a first optical pump beam interface 610 and an optical pump beam 608A from a second optical pump beam interface 612. In some implementations, the configurable optical structure 602 may include a bulk gain medium, such as a laser crystal or a rare-earth-doped glass. Some rare-earth-doped glasses may include dopants such as neodymium, ytterbium, erbium, thulium, cerium, titanium, or chromium. In some implementations, the configurable optical structure 602 can be configured to control the intensity of the optical wave propagating through the region 614 by providing optical gain to the input optical wave propagating through the region.

[0127] Figure 7AAn isometric view of an example optical architecture 700 that can be used as an optical processor is depicted. Collimated light 704 is guided into a laser gain medium 702. A first optical pump beam interface 710 is configured to provide a first set of two or more optical pump beams 706A-706N guided into the configurable optical structure 702. A second optical beam interface 712 is configured to provide a second set of two or more optical pump beams 708A-708N guided into the configurable optical structure 702. In some examples, the respective optical beam interfaces 710, 712 can be configured to provide optical beams in a 2D array. Light propagates through the gain medium 702, and an output 714 is detected by a two-dimensional (2D) array of photodetectors 716. In some implementations, electronic circuitry (not shown) can amplify the photocurrent from the photodetector 716 and digitize the data for further processing. In some examples, electronic circuitry (not shown) can control the intensity of the respective optical beams in the first optical beam interface 710 and the second optical beam interface 712.

[0128] Figure 7B An example two-dimensional perspective view along the yz plane is depicted for the optical architecture 700. The configurable optical structure 702 includes a region 718 at the intersection of an optical pump beam 706A from a first optical pump beam interface 710 and an optical pump beam 708A from a second optical pump beam interface 712. The configurable optical structure 702 also includes a region 720 at the intersection of an optical pump beam 706A from the first optical pump beam interface 710 and an optical pump beam 708B from the second optical pump beam interface 712. In some implementations, these regions 718, 720 may be referred to as 3D pixels or voxels.

[0129] In some implementations, two arrays of the light source can be used to selectively pump one or more voxels within the laser crystal. Some voxel elements can be pumped when light is simultaneously present at a first wavelength λ1 and a second wavelength λ2. In some optical architectures, pumping can be controlled by carefully selecting the wavelengths and gain materials. For example, a laser gain material with three energy levels—e0 [ground state], e1, and e2—can be used in this process, where e1 can have a short carrier lifetime and e2 can have a long carrier lifetime. Figure 7CAn example energy level diagram 700C associated with a gain medium is depicted. In some implementations, the input light guided to the laser gain medium can have a wavelength λ0. In some examples, λ0 can be chosen such that its photon energy is equal to e2-e0, λ1 can be chosen such that its photon energy is equal to e1-e0, and λ2 can be chosen such that its photon energy is equal to e2-e1. In some examples, if light of λ1 is guided only at voxels, electrons in the gain material can be raised to the e1 state, but once the pump light is removed, the short carrier lifetime can cause the electrons to decay back to the ground state. In some examples, if light of λ2 is guided only at voxels, the light can simply pass through the gain material. In some examples, when both wavelengths of light are present, electrons can be pumped to state e1 first with λ1 and then pumped back to state e2 with λ2. Similar effects can also be achieved using other combinations of energy levels.

[0130] In some implementations, a device comprising one or more photonic integrated circuits can be used as a two-dimensional electrically or optically pumped ADONN.

[0131] Figure 8A A side view of an example device 800A comprising a chip 802 of a first material is depicted. The chip 802 includes an optical source 804, a modulator 806, an optical processor 808, and a photodiode 810. The device 800A includes an electronic integrated circuit 812, which can be configured to control the optical source 804, modulator 806, optical processor 808, and photodiode 810. The electronic integrated circuit 812 is connected to components on the chip 802 via a plurality of conductive structures 814. In some implementations, the chip 802 may include a III / V material, and the plurality of conductive structures 814 may include metal bumps, such that the chip 802 and the electronic integrated circuit 812 are connected in a flip-chip configuration.

[0132] In some implementations, one or more photonic integrated circuits comprising III / V materials can be co-packaged with one or more silicon photonic integrated circuits (such as photodetectors and modulators) that include a portion of an ADONN. In some implementations, the optical source can be separated from other components.

[0133] Figure 8B A side view of an example device 800B comprising a chip 822 of a first material is depicted. The chip 822 includes a modulator 824, an optical processor 826, and a photodiode 828. The device 800B includes an optical source 830 coupled to the chip 822. In some implementations, the first material may include silicon. An electronic integrated circuit 832 is connected to various structures of the chip 822 via a plurality of conductive structures 834.

[0134] Figure 8CA side view of an example device 800C is depicted, showing a first chip 842 comprising a first material and a second chip 843 comprising a second material. The first chip 842 includes a modulator 844 and a photodiode 845. The second chip 843 includes an optical processor 846. The device 800C also includes an electronic integrated circuit 847 connected to the second chip 843 via a plurality of conductive structures 848. An optical source 849 is coupled to the first chip 842. In some implementations, the first material may include silicon, and the second material may include a III / V material.

[0135] Figure 8D A side view of an example device 800D is depicted, showing a first chip 852 comprising a first material and a second chip 853 comprising a second material. The first chip 852 includes a modulator 854. The second chip 853 includes an optical processor 855. The device 800D also includes a first electronic integrated circuit 856 connected to the modulator 854 of the first chip 852 via a plurality of conductive structures 857, and a second electronic integrated circuit 858 connected to the optical processor 855 of the second chip 853 via a plurality of conductive structures 859. An optical source 860 is coupled to the first chip 852. A detector (not shown) can detect light from the second chip 853. In some implementations, the first material may include silicon, and the second material may include a III / V material.

[0136] Figure 8E A side view of an example device 800E is depicted, comprising a first chip 862 and a second chip 863 of a first material, and a third chip 864 of a second material. The first chip 862 includes a modulator 865, the second chip 863 includes a detector 866, and the third chip 864 includes an optical processor 867. The device 800E also includes electronic integrated circuits 868, 869, and 870. Each electronic integrated circuit 868, 869, and 870 is connected to its respective first chip 862, second chip 863, and third chip 864 via corresponding plurality of conductive structures 871, 872, and 873. An optical source 874 is coupled to the first chip 862. In some implementations, the first material may include silicon, and the second material may include a III / V material.

[0137] Figure 8FA side view of an example device 800F is depicted, showing a first chip 882 comprising a first material and a second chip 883 comprising a second material. The first chip 882 includes a modulator 884. The second chip 883 includes an optical processor 885 and a detector 886. Device 800D also includes a first electronic integrated circuit 887 connected to the first chip 882 via a plurality of conductive structures 888 and a second electronic integrated circuit 889 connected to the second chip 883 via a plurality of conductive structures 890. An optical source 891 is coupled to the first chip 882. In some implementations, the first material may include silicon, and the second material may include a III / V material.

[0138] like Figures 8A-8F As shown, in some implementations, the III / V material can provide optical gain and may include some of the other functions of the ADONN, such as generating light, modulating light, or detecting light. In some implementations, the device can be constructed entirely using silicon photonics processes that include in-process gain and utilize integral gain for planar mode propagation. In some implementations, the ADONN can be constructed entirely using silicon photonics processes and may be a PN junction configured as a planar mode waveguide to attenuate intensity and / or modulate the phase of light within the planar surface.

[0139] Figure 9 A top view of an example device 900 including a photonic integrated circuit 902 is depicted. The photonic integrated circuit 902 includes a segmented region 904 configured to receive light from an optical source 905 and segment the light into N waveguides 906A-906N, where N is any number. In some implementations, each waveguide 906A-906N may include an intensity modulator. Each intensity modulator may be driven by electronic circuitry (not shown) that can convert a digital input signal into a voltage or current that changes the modulation intensity. The output 908 of the modulator is directed to an optical processor 910, which includes an optical gain medium forming a planar pattern waveguide with gain. The output 912 of the optical processor 910 is directed to a plurality of waveguides 914A-914N. In some implementations, each waveguide 914A-914N may be coupled to a corresponding photodetector or photodiode. Electronic circuitry (not shown) can amplify and digitize the photocurrent generated by the individual photodiodes.

[0140] In some implementations, an array of N waveguides can be coupled to a junction formed at the interface between the p-doped and n-doped regions in the optical gain medium, thereby forming a planar mode waveguide with gain. Figure 10AA side view of an example optical processor 1000A is depicted. The optical processor 1000A includes a cathode layer 1002 comprising metal, a semiconductor material layer 1004, an active region 1006 forming the layer, and a semiconductor material layer 1008. In some implementations, layer 1004 may be doped with an n-type dopant and layer 1008 may be doped with a p-type dopant. The active region 1006 may be coupled to one or more waveguides at an optical interface 1010. Multiple metal contacts 1012A-1012F each form a corresponding anode contact. In some implementations, the anode contacts may be collectively closely spaced at a spacing close to the wavelength of the light used in the system. Electronic circuitry (not shown) may be connected to the metal contacts 1012A-1012F, thereby allowing different currents 1014A-1014F to be driven by the individual metal contacts 1012A-1012F. In some implementations, as indicated by the size of the arrows, the currents 1014A-1014F can be different from each other. In some implementations, when the electronic circuitry changes the current entering the individual metal contacts 1012A-1012F, the absorption / gain of the planar waveguide can locally vary around the metal contacts 1012A-1012F due to the change in local current density. As indicated by the thickness of the solid black arrows, the intensity of the light 1016 coupled to the active region 1006 can vary due to the currents 1014A-1014F.

[0141] In some implementations, the optical processor can be optically pumped. Figure 10B A side view of an example optical processor 1000B is depicted. The optical processor 1000B includes a first material layer 1024, an active region 1026, and a second material layer 1028. In some implementations, the material may include glass, and the active region 1026 may include doped glass or a laser crystal. Multiple light sources 1030A-1030F emit corresponding light 1032A-1032F. Electronic circuitry (not shown) can control the intensity of each light source 1030A-1030F. The light 1032A-1032F drives corresponding currents 1034A-1034F in the active region 1026. In some implementations, as indicated by the size of the arrows, the currents 1034A-1034F may be different from each other. As indicated by the thickness of the solid black arrows, the intensity of the light 1036 coupled to the active region 1026 may vary due to the currents 1034A-1034F. An array of detectors or photodiodes (not shown) at the output of the flat panel can convert light into photocurrent, and electronic circuitry (not shown) can amplify and digitize the light from the individual photodiodes.

[0142] The architecture for photonic neural network processors allows for on-the-fly reconfiguration of weights and nonlinear activation functions in the propagation medium. Photonic neural networks can encode the weights of hidden layers using variable-pump optical gain and solve the neural network using the propagation of light. In some implementations, photonic neural networks can achieve low latency and high computational density. Some photonic neural networks can be constructed from silicon photonic integrated circuits, III / V photonic integrated circuits, heterogeneous integrated silicon and III / V integrated circuits, or from bulk optics that can be controlled and monitored by electronic and photonic integrated circuits.

[0143] While this disclosure has been described in conjunction with certain embodiments, it should be understood that this disclosure is not limited to the disclosed embodiments, but rather is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims, which will be given the broadest interpretation to cover all such modifications and equivalent structures permitted by law.

Claims

1. An apparatus comprising: Multiple optical processing stages are configured to process wavelengths with spectral peaks. Two or more optical waves, wherein each of the two or more optical processing stages in the plurality of optical processing stages includes: Two or more configurable optical structures substantially coplanar with a plane, wherein each configurable optical structure is configured to receive an optical wave propagating along a first axis substantially parallel to the plane, and each configurable optical structure includes an active region having: The width along a second axis that is substantially parallel to the plane and perpendicular to the first axis, wherein the width is less than or equal to , The height along a third axis that is substantially perpendicular to the plane and perpendicular to the first axis, wherein the height is greater than... ,as well as Less than or equal to along the first axis The length; and An interface region configured to receive optical waves from each of the two or more configurable optical structures. Each interface region associated with a corresponding optical processing stage in at least two of the plurality of optical processing stages is configured to couple at least a portion of an optical wave received from at least one configurable optical structure to at least two configurable optical structures in a subsequent optical processing stage.

2. The device according to claim 1, wherein, Each active region is configured to guide up to four spatial modes associated with optical waves.

3. The device according to claim 1, wherein, Each active region is configured to contain a corresponding percentage of electromagnetic power associated with optical waves propagating through the respective active region, which is greater than 50% relative to the total electromagnetic power associated with optical waves propagating through the configurable optical structure including the respective active region.

4. The device according to claim 3, wherein, Each active region is configured to contain a percentage of electromagnetic power associated with optical waves propagating through the respective active region, which is greater than 70% relative to the total electromagnetic power associated with optical waves propagating through a configurable optical structure including the respective active region.

5. The device according to claim 1, wherein, At least a portion of each active region of the corresponding optical processing stage is separated from at least a portion of one or more other active regions of the corresponding optical processing stage by a portion of a region including insulating material.

6. The device according to claim 5, wherein, Each region, including the insulating material, extends below the respective surface of each adjacent active region along a corresponding axis that is substantially perpendicular to the plane and parallel to the third axis.

7. The device according to claim 1, wherein, At least a portion of each active region of the corresponding optical processing stage is separated from at least a portion of one or more other active regions of the corresponding optical processing stage by a corresponding air insulation gap.

8. The device according to claim 1, wherein, Each interface region includes a planar waveguide structure formed within a substrate, wherein the planar waveguide structure is coupled to a plurality of configurable optical structures at a first end and at a second end opposite to the first end.

9. The device according to claim 1, wherein, Each of the two or more configurable optical structures is configured to provide an intensity variation of an optical wave propagating through a respective active region, wherein one or more of the intensity variations provides optical gain to the optical wave.

10. The device according to claim 9, wherein, Each of the configurable optical structures is configured such that the optical gain provided to the optical wave is nonlinear with respect to the intensity of the optical wave.

11. The device according to claim 1, wherein, At least one configurable optical structure can be configured to transmit at least a portion of one or more optical waves in a first operating mode, and can be configured to detect the intensity of the optical waves in a second operating mode.

12. The device according to claim 1, wherein, Each active region of the configurable optical structure includes a first semiconductor material.

13. The device according to claim 1, wherein, Each configurable optical structure also includes: Including the first layer of the first semiconductor material, The second layer includes the active region, wherein the active region includes a second semiconductor material, and Including the third layer of the third semiconductor material, The second layer is located between the first layer and the third layer.

14. The device according to claim 13, wherein, The second layer also includes: A fourth layer comprising a fourth semiconductor material, wherein the fourth layer is located between the first layer and the active region, and The fifth layer includes the fourth semiconductor material, wherein the fifth layer is located between the third layer and the active region.

15. The device according to claim 14, wherein, The fourth semiconductor material comprises a composition of indium gallium arsenide phosphide.

16. The device according to claim 13, wherein, A portion of the active region includes a quantum well.

17. The device according to claim 13, wherein, A portion of the active region includes bulk semiconductor material.

18. The device according to claim 13, wherein, The first layer includes the first semiconductor material in which dopants are mixed, and the third layer includes the third semiconductor material in which dopants are mixed.

19. The device according to claim 18, wherein, (1) The dopant of the first layer includes a p-type dopant and the dopant of the third layer includes an n-type dopant, or (2) The dopant of the first layer includes an n-type dopant and the dopant of the third layer includes a p-type dopant.

20. The device according to claim 13, wherein, The first semiconductor material and the third semiconductor material each comprise a composition of indium gallium arsenide phosphide.

21. The device according to claim 13, wherein, The second semiconductor material comprises a composition of indium gallium arsenide phosphide.

22. The device according to claim 1, wherein, The corresponding optical wave is provided to each configurable optical structure of the optical processing stage among the plurality of optical processing stages.

23. The device according to claim 22, wherein, The corresponding optical waves are provided to each configurable optical structure by the corresponding modulator.

24. The device according to claim 1, wherein, At least the first active region is configured to control the optical properties associated with optical waves propagating through the first active region.

25. The device according to claim 24, wherein, The optical property that the first active region is configured to control is the optical power associated with the optical wave propagating through the first active region.

26. The device according to claim 25, wherein, The first active region is configured to increase the optical power associated with the optical waves propagating through the first active region.

27. The device according to claim 24, wherein, The first active region comprises a semiconductor material, wherein the band gap energy of the semiconductor material is lower than the wavelength of the spectral peaks of the two or more optical waves. Related energy.

28. The device according to claim 27, wherein, The semiconductor material is a direct bandgap semiconductor material.

29. The device according to claim 24, wherein, The first active region is configured to control optical properties associated with optical waves propagating through the first active region, at least in part, based on electro-optic or thermo-optic effects.

30. The device according to claim 1, wherein, Each active region includes a material configured to control optical properties associated with optical waves traveling through the respective active region, at least in part, based on a nonlinear magnetic susceptibility associated with the material.

31. The device according to claim 30, wherein, Each active region is configured to control the optical properties associated with the optical wave through one or more of the following electro-optic effects: (1) Franz-Keldish effect, (2) quantum confinement Stark effect, (3) Pockels effect, (4) plasma dispersion effect and (5) Kerr effect.

32. The device according to claim 1, wherein, Each interface area is basically composed of passive materials.

33. An apparatus comprising: An integrated circuit device comprising: Including the first layer of metal, Including the second layer of the first semiconductor material, The third layer, including the active region of the second semiconductor material, and Including the fourth layer of the third semiconductor material, Wherein, the second layer is located between the first layer and the third layer, and the third layer is located between the second layer and the fourth layer; An optical interface configured to deliver two or more optical waves to different corresponding portions of the active region along different corresponding propagation axes that are substantially parallel to each other, the different corresponding propagation axes including at least a first propagation axis and a second propagation axis; A plurality of metal contacts electrically connected to the fourth layer, wherein a first subset of the metal contacts is arranged along the first propagation axis, and a second subset of the metal contacts is arranged along the second propagation axis; and A power source is configured to apply a corresponding electric field between the first layer and each of the metal contacts in the plurality of metal contacts.

34. The device according to claim 33, wherein, The first layer and the second layer are electrically connected.

35. The device according to claim 33, wherein, The second layer further includes a dopant mixed in the first semiconductor material, and the fourth layer further includes a dopant mixed in the third semiconductor material.

36. The device according to claim 35, wherein, (1) The dopant of the second layer includes an n-type dopant and the dopant of the fourth layer includes a p-type dopant, or (2) The dopant of the second layer includes a p-type dopant and the dopant of the fourth layer includes an n-type dopant.

37. The device according to claim 33, wherein, The first semiconductor material and the third semiconductor material include indium phosphide.

38. The device according to claim 33, wherein, The second semiconductor material comprises a composition of indium gallium arsenide phosphide.

39. The device according to claim 33, wherein, The fourth layer also includes multiple regions of the third semiconductor material internally mixed with dopants, wherein each of the multiple metal contacts is electrically connected to at least a portion of different corresponding regions in the multiple regions.

40. The device according to claim 39, wherein, At least a portion of each of the plurality of regions is separated from at least a portion of each of the other regions of the plurality of regions by either a portion of the third semiconductor material that is free of dopants or a region that is free of the third semiconductor material.

41. The device according to claim 40, wherein, One or more of the regions without the third material include electrically insulating or optically transparent materials.

42. The device according to claim 33, wherein, Each portion of the active region between the metal contacts of the plurality of metal contacts and the first layer is configured to provide an intensity variation of an optical wave propagating through a corresponding portion of the active region, at least in part, based on a corresponding electric field applied between the metal contacts of the plurality of metal contacts and the first layer, wherein one or more of the intensity variations provide optical gain to the optical wave.

43. The device according to claim 42, wherein, Each portion of the active region is configured such that the optical gain provided to the optical wave is nonlinear with respect to the intensity of the optical wave.

44. The device according to claim 33, wherein, The third layer also includes: A fifth layer comprising a fourth semiconductor material, wherein the fifth layer is located between the second layer and the active region, and A sixth layer comprising the fourth semiconductor material, wherein the sixth layer is located between the fourth layer and the active region.

45. The device according to claim 44, wherein, The fourth semiconductor material comprises a composition of indium gallium arsenide phosphide.

46. ​​The device according to claim 33, wherein, A portion of the active region includes a quantum well.

47. The device according to claim 33, wherein, A portion of the active region includes bulk semiconductor material.

48. An apparatus comprising: A configurable optical structure is configured to receive an input optical wave propagating along a first axis, wherein the input optical wave has a spatial profile distributed along a second axis perpendicular to the first axis and a third axis perpendicular to both the first and second axes. A first optical pump beam interface is configured to provide a first set of two or more optical pump beams, wherein each optical pump beam in the first set of two or more optical pump beams is guided into the configurable optical structure along a respective axis perpendicular to the first axis and parallel to the second axis. A second optical pump beam interface is configured to provide a second set of two or more optical pump beams, wherein each optical pump beam in the second set of two or more optical pump beams is guided into the configurable optical structure along a respective axis perpendicular to the first axis and parallel to the third axis.

49. The device according to claim 48, wherein, The configurable optical structure includes the region at the intersection of an optical pump beam from a first set of two or more optical pump beams and an optical pump beam from a second set of two or more optical pump beams.

50. The device according to claim 49, wherein, The configurable optical structure is configured to control the intensity of the optical wave propagating through the region by providing optical gain to the input optical wave propagating through the region.

51. The device according to claim 50, wherein, The region is configured such that the optical gain provided to the optical wave is non-linear with respect to the intensity of the optical wave.

52. The device according to claim 48, wherein, The configurable optical structure includes multiple regions, wherein each of the multiple regions is located at the intersection between optical pump beams from a first set of two or more optical pump beams and optical pump beams from a second set of two or more optical pump beams.

53. The device according to claim 48, wherein, The configurable optical structure includes laser crystals or glass doped with rare earth elements.

54. The device according to claim 48, wherein, The input optical wave is a collimated light source.

55. The device according to claim 48, wherein, Each optical pump beam in the first set of two or more optical pump beams includes an optical wave having a first wavelength, and each optical pump beam in the second set of two or more optical pump beams includes an optical wave having a second wavelength, wherein the first wavelength and the second wavelength are determined at least in part based on the material of the configurable optical structure.

Images