Ultra wide band optical processor

Abstract

To provide a photonic computing system for enabling optical coupling between light fields associated with various propagation conditions in one or more photonic guide units associated with one or more processing tasks of an artificial light neuron unit, and to provide a network rack for housing a plurality of hardware modules including the photonic computing system.SOLUTION: A photonic computing system comprises an arrangement of a plurality of photonic processing units BFBl having an input port and an output port 14 / 16, each of the photonic processing units comprising: an array of photonic guiding units PGUn configured to define propagation conditions for a plurality of light fields associated with one or more optical processing tasks; and a plurality of optical connectors BCk. Each of the optical connectors couples the light field to the light field each other between the input port and the output port of the photonic processing units, thereby providing a network of communicating processing units.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to photonic processors particularly useful for optical data communications, and to optical computing devices suitable for use in optically integrated artificial neuron networks. [Background technology]

[0002] Optical / photonic computing performs computational processes by manipulating visible or infrared light. Optical computing can perform calculations faster than electronic systems because it allows for faster manipulation of light pulses and broader bandwidth information transmission.

[0003] Optical computing can provide low-power, high-speed solutions to real-time machine learning and artificial intelligence (AI) needs, thereby enabling the implementation of artificial neural networks (ANNs). An ANN is essentially a computer system trained to automatically recognize data and optimize its performance to perform relevant actions in response to the input data. A typical ANN system is based on a set of connected units or nodes (artificial neurons), and the connections between nodes allow signals to be transmitted from one node to another. Various inputs to a particular artificial neuron are multiplied by linear weights, and the output of the artificial neuron is generated by applying a nonlinear function (such as sigmoid or RelU) to the sum of the weighted inputs, adding a bias coefficient. The artificial neuron receives signals, processes them, and transmits the resulting signals to one or more artificial neurons connected to it.

[0004] Generally, implementations of photonic computing are based on silicon photonics and photonic integrated circuit platforms, or free-space solutions. However, such implementations are limited by phase noise, thermal sensitivity, and inefficiencies dependent on scaling. In contrast, optical fibers offer a robust, stable, and well-developed technology that has been well-suited to low-power and high-bandwidth optical communication applications.

[0005] For example, WO2017 / 033197 by Zalevsky et al. teaches an integrated optical module. The optical module includes multiple optically coupled channels, enabling its use in artificial neural networks (ANNs). According to some embodiments, the integrated optical module includes a multicore optical fiber, where the cores are optically coupled.

[0006] E. Cohen et al., "Neural networks within multi-core optic fibers," Sci Rep. July 7, 2016; 6:29080, describes a conceptual design of optical neural networks within fibers. Neurons and synapses are realized as individual silica cores within multi-core fibers. Optical signals are transmitted across the cores by optical coupling. Pump-driven amplification of erbium-doped cores mimics synaptic interactions. Simulations show that the network can distinguish given inputs depending on the specific configuration of the amplification, and that multi-core fibers can function as components of large-scale, low-capacitance optical artificial neural networks. [Overview of the project]

[0007] As mentioned above, photonic computing systems offer significantly superior computing bandwidth and efficiency compared to electronic computing systems. At the system level, adding modules (such as communication modules, storage modules, and cooling modules) to an electronic computing system consumes more power without increasing computing bandwidth, thus reducing overall efficiency (measured in computed watts).

[0008] In AI applications, there is a clear positive correlation between the number of parameters (nodes, connections, layers, etc.), computational accuracy, and the system's hierarchical generalization ability. In some embodiments, this may relate to saving or losing human lives (i.e., autonomous vehicles, health-related AI applications, and safety automation). On the other hand, supercomputing systems require megawatts and are not energy-sustainable. Therefore, various applications, especially AI applications, require photonic supercomputing systems that offer extremely high performance and far greater efficiency (low power consumption).

[0009] Today, server-sized systems typically have 8 to 16 processors, with 3 to 6 of these servers forming a rack (or pod), while supercomputing systems consist of tens to over 100 such racks (a total of several thousand processors).

[0010] As mentioned above, the coupling of light propagating through photonic guide units (such as optical fibers and waveguides) can be used for a variety of processing tasks.

[0011] The present invention provides a novel solution that enables optical coupling between light fields related to various propagation conditions in one or more photonic guide units related to (involved in) one or more processing tasks of an optical / photonic processor, such as an artificial optical neuron unit.

[0012] In the following description, such photonic guide units will be referred to as “fibers” or “optical fibers,” but it should be understood that the principles of the present invention are not limited to the use of fiber-based photonic processors. Furthermore, the present invention can also be implemented using other types of waveguides or generally photonic guide units that can perform photonic computing / processing via light field interactions.

[0013] Furthermore, for the purposes of the present invention, it should be noted that "light fields related to various propagation conditions" may relate to various propagation modes in a multimode photonic guide unit (fiber), and / or various cores of one or more multicore fibers of any type of single-mode, minority-mode, or multimode fiber, and / or various fibers which are any of single-mode, minority-mode, or multimode fibers, and can support various frequencies.

[0014] The technology of the present invention enables a so-called rack configuration of basic functional blocks formed by arrays of the aforementioned photonic guide units, each capable of operating as a photonic processor (e.g., a fiber-based system). This rack configuration can be used, in conjunction with any known appropriate technology, to realize photonic-type communications or neural networks (e.g., multi-core-based neural networks), thereby enabling the compact realization of ultrawide databand processing machines. Such a novel array of photonic processors provides a very large number of processors (e.g., 1800 or more processors) that can be mounted in parallel in a single rack. The inventors demonstrate exemplary information flow, utilization, and programmability of such an array and demonstrate a method that potentially improves computational efficiency by more than six orders of magnitude while achieving supercomputing performance.

[0015] The technology of the present invention can be implemented using commercially available components to form the building blocks (processing units) of a photonic computing neuron, enabling efficient interfacing of electronic devices with the outside world via conventional optical communication components. By using the technology of the present invention, predictive performance exceeding 1 million tera-computations (TOPs) per second and efficiency exceeding 1000 TOPs / watt can be achieved.

[0016] In the following description, the terms “computation unit” and “processing unit” are used interchangeably to refer to an array of photonic guide units configured to perform data / signal processing tasks through coupling and interaction between light fields. Similarly, the terms “computation system” and “processing system” are used interchangeably to refer to a system formed by multiple / arrays of computing / processing units and their associated input / output modules.

[0017] Therefore, according to a broad aspect of the present invention, a photonic computing system is provided, which comprises an array of photonic processing units, each of which is an array of a predetermined number of photonic processing units having input ports and output ports, each of which includes an array of photonic guide units configured to define propagation conditions for a plurality of light fields related to one or more optical processing tasks; and a plurality of optical connectors, each of which performs coupling of light fields between the input ports and output ports of adjacent photonic processing units in the array of photonic processing units, thereby providing a network of communicating processing units.

[0018] Optical connectors are configured as bridge-type connectors.

[0019] A photonic processing unit having optical connectors is arranged in parallel with a gap therebetween to form a processor panel, whereby a photonic computing system is configured as a module that can be accommodated in a network rack.

[0020] In some embodiments, at least one or some of the optical connectors are configured to connect an output port of one processing unit to an input port of another processing unit.

[0021] ​​​​​​​​​​​​​​​​​The optical connector can be configured and operable to apply a pre-set amplification pattern to the light field, together with each input port and output port of the photonic processing unit. The optical connector can further include a local controller configured and operable to control the applied amplification pattern and control the readout of the output light.

[0026] The photonic processing unit includes a photonic body configured to define two or more input ports and two or more output ports. In some embodiments, the photonic body has a hexagonal cross-section, and its facets are associated with the input ports and output ports.

[0027] In some embodiments, at least one or some of the input ports include an input synchronization unit, and this input synchronization unit can include a clock lock utility configured to perform client-server clock locking and / or a clock phase alignment utility configured to perform optical clock phase delay synchronization.

[0028] In some embodiments, at least one or some of the input ports are configured and operable to perform wavelength conversion. <0000!05> In some embodiments, at least one or some of the output ports are configured and operable to perform reshaping of the output signal.

[0030] The photonic guide unit is configured to define one or more types of light fields, including a multi-mode light field, a multi-core light field defined by a combination of single-mode, multi-mode or both optical channels, a light field propagating through an erbium-doped core, a light field propagating through a core based on a photonic crystal fiber, and a light field confined in a 2D or 3D photonic crystal. [[ID=*1]]

[0031] The photonic processing unit can utilize a photonic lantern created by adiabatically coupling multiple single-mode cores into a single multi-mode core.

[0032] According to another aspect of the present invention, a network rack is provided for housing a plurality of hardware modules, including one or more photonic computing systems configured as described above. [Brief explanation of the drawing]

[0033] To better understand the subject matter disclosed herein and to illustrate how it may be put into practice, embodiments are described below, with reference to the accompanying drawings, as merely non-limiting examples. [Figure 1] Figure 1 is a schematic diagram of the photonic computing system of the present invention. [Figure 2] Figures 2A to 2C show the arrangement and operation of several basic functional blocks of the photonic computing system of Figure 1. Figures 2A and 2C show a portion of the photonic processing system formed by the multiple basic blocks, and Figure 2B shows an exemplary configuration of the photonic body of a processing unit that includes multiple photonic guide units. [Figure 3] Figure 3 shows a network rack configured to house one or more photonic computing systems of the present invention. [Figure 4] Figure 4 schematically shows an embodiment of the present invention in a multi-core rack configuration of the photonic computing system of the present invention. [Figure 5] Figures 5A and 5B show the configuration of the input port, including the input synchronization unit and wavelength converter. [Figure 6]Figures 6A to 6F illustrate the fabrication and operation of the photonic computing system of the present invention, where Figure 6A shows the setup of a single computing unit, Figure 6B shows the corresponding fabricated multimode image, Figures 6C and 6D show the mode propagation and crosstalk of a combination of two input signals and the mixed signal across the output plane, and Figures 6E and 6F show the nonlinear activation functions that can be provided by the fiber amplifier. [Figure 7] Figures 7A to 7D illustrate the configuration and operation of the EDMF-based computing unit / basic block of the present invention, with Figures 7A and 7B showing a side view and cross-section of an example of the photonic body of the basic block, respectively, Figure 7C showing the spectrum of the computing unit with and without coupling between different cores of the photonic body in Figures 6A and 6B, and Figure 7D showing another example of an EDMF-based photonic body with a network realized by core doping. [Modes for carrying out the invention]

[0034] Referring to Figure 1, a photonic processing / computing system 10 of the present invention is schematically shown, configured and operable to perform various data processing tasks / functions. This system 10 can be used as an optical communication system or an optical computing system.

[0035] The processing system 10 consists of a predetermined number L of basic functional blocks BFB1...BFB configured as a photonic processing unit / computing unit. l Includes array / sequence 12. Each i-th photonic processing unit BFB i It has its associated input / output ports 14, 16. Each i-th photonic processing unit BFB iThe (basic function block) is configured and operable to perform one or more computing / processing tasks / functions and consists of multiple N photonic guide units PGU1...PGU that together form multiple light field channels (generally at least two such channels) configured to define M light field propagation conditions. n This includes M ≥ N.

[0036] As described above, multiple optical channels can be associated with multiple fibers, for example, one or more single-mode fibers and / or one or more multi-mode fibers and / or single-core or multi-core fibers.

[0037] In the following description, the photonic processing unit and the array 12 of such units may also be referred to as a fiber array or fiber-based unit. However, as stated above, the principles of the present invention relating to a novel architecture of the photonic processing system 10 are not limited to a fiber-based configuration.

[0038] The photonic guide unit defines at least one optical propagation path (core). For the purposes of this invention, it should be understood that the meaning of the term “optical channel” should not be limited to “light of a specific spectrum” or to a physical optical propagation path, but should be interpreted broadly to include the light field of a specific propagation mode in a multimode fiber core.

[0039] Therefore, for example, the photonic guide unit may be configured as a fiber unit, and the light field channels may be associated with multiple cores of the same fiber and / or multiple modes of the same or different cores (i.e., a broad core supporting multiple spatial light modes). Consider an example where at least some of the multiple optical channels are associated with different optical paths (e.g., different fibers or different cores of the same or different fibers), each of such optical paths has a first end for receiving input light and a second end for providing output light. It should be understood that when multicore fibers are used, they should be appropriately configured to provide the desired coupling between the cores.

[0040] According to the present invention, input / output units 14, 16 are configured to collect light from at least one output of one photonic processing unit and couple the light to one or more relevant inputs of one or more adjacent photonic processing units in order to perform the associated photoprocessing task. As further shown in Figure 1, the system 10 consists of a plurality of K optical connectors BC1...BC configured as a bridge-like connector (optical path) for connecting the input / output ports 14, 16 of adjacent photonic processing units. k This includes the following. As will be further explained below, the photonic processing unit can also be associated with multiple input ports and / or multiple output ports.

[0041] It should be noted that a bridge-type optical connector can be configured to connect the output of one photonic processing unit to the input of another photonic processing unit (i.e., a 1x1 optical connection), or it can be configured to connect the output of one photonic processing unit to the inputs of G (G≧2) processing units (i.e., a 1xG optical connection). In the case of a 1xG optical connection, the relevant input / output units are appropriately equipped with one or more optical splitters and couplers. This is the optical connector BC. k-1 and BC k However, the same photonic processing unit BFB lThe input ports 14 / 16 are connected to the photonic processing unit BFB l-1 and BFB l-2 are schematically shown in FIG. 1 which shows that they are respectively connected to the outputs 14 / 16. For this purpose, the input / output ports 14 / 16 of the photonic processing unit BFB l are equipped with an optical splitter and coupler assembly 15.

[0042] Also, it should be noted that the optical connectors can be appropriately configured to maintain or not maintain the polarization of light as required for specific optical processing tasks. Polarization maintenance can be implemented by using reflection, diffusion and / or refraction structures or interfaces.

[0043] [[ID=]] In some embodiments, the optical connector / coupler can be configured to maintain a light field of one or more specific wavelengths (or wavelength ranges) and can be appropriately equipped with a wavelength selective splitter. Those features may be useful for the WDM function.

[0044] Referring to FIGS. 2A to FIG. 2C, it shows how the basic blocks described above are arranged in a photonic processing / computing system. For ease of understanding, the same reference numerals are used to identify similar components in all examples described herein.

[0045] Figure 2A schematically shows a portion of the photonic processing system 10 formed by multiple / arrays of the processing units (basic functional blocks) BFB described above. Figure 2B shows more specifically that each processing unit BFB contains an array of photonic guide units PGUs and that each processing unit BFB is associated with input / output units / ports 14, 16. As shown in Figure 2C, each processing unit BFB is surrounded by other processing units (unless restricted by the edges of the entire processing unit assembly) and can communicate with all surrounding processing units via bridge connectors BC to receive or transmit optical carrier information from them.

[0046] Preferably, the processing unit has a hexagonal cross-section photonic body 18 containing a plurality of photonic guide units PGU, with hexagonal external facets F1 to F6 used as input / output ports 14 and 16 on both sides of the body 18, and the photonic body 18 associated with 12 ports. Each port can implement either input or output functionality and can inject both signal wavelength patterns and amplification patterns into the processing unit.

[0047] Light from each output port enters a U-shaped bridge connector BC. The latter includes an electronic cable connected to a local controller in its rear section 20, which controls the injected amplified light and reads out the output light (which can be used for training and calibration). The bridge connector BC can include multimode fiber and a pixel-level weighting device (e.g., based on the use of DMD or LCD technology) that allows for flexible manual reconfiguration of the system architecture, or a fixed filter that can be replaced with a flexible one after training is complete.

[0048] Thus, the photonic processing system 10 forms a network of communicating processing units 15. The number of processing units 15 in the network and their connectivity are limited by the number of photonic cores defined by the photonic guide unit PGU. However, they are programmable because they depend on amplification patterns that are optimized during the system's learning phase. By using amplification for both computation and routing within the processing units, the flow of light from the input port to the output port is determined. At least one input port and at least one output port are configured by programmable amplification patterns. The divergence and convergence of data within the network elements are also controlled by the amplification patterns.

[0049] The approach of the present invention provides a basic functional block (BFB) and a "rack-type" configuration of the processing system 10, i.e., a high-density processing unit within a limited volume similar to a communications rack (19 inches wide, 2-2.5 m high). This enables the compact realization of an ultrawide databand processor. Furthermore, this technology allows for the flexible manual reconfiguration of the optical processing architecture and complexity.

[0050] This is schematically shown in Figure 3, which illustrates a network rack 30 configured to house network hardware. System 10 is configured as a processor panel formed by the above processing units, with associated bridge-like connectors spaced apart and arranged parallel to each other (in a matrix arrangement). Thus, system 10 can form one of the hardware modules of the network rack. This allows for the configuration of a network rack containing one or more photonic processing systems 10 and other electronic components EC.

[0051] Figure 4 schematically illustrates the implementation of the above approach of the present invention in a photonic processing system 10 configured as a so-called multicore rack utilizing multiple fiber-based basic functional blocks (processing units) BFB. The processing unit BFB is configured as a multicore segment having desired coupling between cores. A bridge-type optical connector BC is appropriately provided to couple the output core of one input fiber segment to the input core in the other fiber segment.

[0052] As described above, any known and suitable techniques for implementing fiber-based communications or neural networks (e.g., multicore-based neural networks) can be integrated into the aforementioned rack configuration, enabling the compact implementation of ultrawide databand processing machines. Bricks such as fiber segments through which light travels from an input core to an output core (e.g., multiple cores of a multicore fiber with appropriate coupling between cores) can be connected to additional bricks using U-shaped optical connectors BC that provide coupling between cores.

[0053] For example, an adjustable device containing liquid crystal-based pixels can be integrated in front of the right-hand core to perform amplitude / phase modulation on the spatial core and / or can also be constructed as a spectral filter to allow attenuation of a desired spectral band for each desired core in the brick.

[0054] Generally, the processing unit BFB that can be used in the photonic processing system 10 of the present invention may be one or more of the following types: multimode-based, multicore-based (single / multimode or a combination thereof), erbium-doped glass with or without femtosecond laser scribing (DLW), photonic crystal fiber-based, or 2D / 3D photonic crystal.

[0055] As schematically shown in Figure 5A, the basic block can be appropriately configured to define input synchronization units 14, 16. The input synchronization units 14 / 16 include a clock lock utility 40 configured to perform client-server clock locking. For this purpose, a round-trip protocol having a clock reference and counter is used, thus providing that all signal sources can be synchronized to the same clock. The input synchronization units also preferably include a clock phase alignment utility 42 configured to provide optical clock phase delay synchronization. As shown, an optically controlled variable delay mechanism can align the clock phases to some extent (a few percent).

[0056] In some embodiments, input units 14 and 16 are configured to perform wavelength conversion. For this purpose, signals from separate DWDM frequencies can be converted to a single wavelength to generate a desired interaction using the cross-phase modulation phenomenon. Similarly, a single wavelength can be converted to various DWDM frequencies to avoid coherence.

[0057] This is illustrated in Figure 5B. Wavelength conversion can be achieved when two input signals, λ1 (intensity-modulated input) and λ2 (a continuous wave (CW signal), or pulse train, or return-to-zero (RZ) signal), are injected together into an active device such as an erbium-doped fiber amplifier (EDFA) or a semiconductor optical amplifier (SOA). The input signals are compared for gain resources (cross-gain modulation) or through nonlinear processes such as the Kerr effect (cross-phase modulation) or four-wave mixing.

[0058] In some embodiments, output units 14, 16 are configured to perform output pulse reshaping. This may include amplification based on a standard erbium-doped fiber amplifier and WDM multiplexing using a cross-gain modulated optical ADC (which converts to a digital signal if necessary). The same nonlinear processes described above can be used to imprint a signal pulse shape onto a continuous wave light source or to reshape existing pulses. A photonic processing unit BFB suitable for use in the system of the present invention can utilize a photonic lantern made by adiabatically coupling multiple single-mode cores into a single multimode core. This provides a low-loss interface between a single-mode system and a multimode system, where precise optical mapping between the core and individual modes is not required. Specifically, this is a photonic lantern [e.g., TABirks et al., "The photonic lantern", Adv. Opt. Photonics 7, 107 (2015)], where fan-in and fan-out components are joined back-to-back, and seven or more input single-mode fibers are adiabatically tapered to form a multimode section, which is adiabatically connected to seven output single-mode fibers.

[0059] The data can be injected into a fiber-based photonic processing unit by a standard fiber-coupled transceiver (transmitter / receiver) that also functions as a receiving unit for setup. Weights can be supplied by a computer-controlled variable optical attenuator (VOA), and the output signal is amplified by an erbium-doped fiber amplifier (EDFA).

[0060] Refer to Figures 6A to 6F illustrating the manufacturing and operation of the computing system of the present invention. Figure 6A shows the setup / module for a single computing unit BFB. Figure 6B shows an image of the corresponding manufactured multimode device.

[0061] The optical connector BC can be configured to apply a preset modulation to the light input to the photonic processing unit BFB and the light output from the photonic processing unit BFB. Such modulation may include one or more linear and / or nonlinear amplification patterns of the optical signal.

[0062] For example, the optical connector BC provides excitation light of a specific wavelength to apply a nonlinear amplification pattern to the light field passing through each photonic guide unit PGU. The photonic guide unit may include one or more doping materials that can be excited by excitation light (e.g., centered around approximately 980 nm / 980 nm), and that excitation light causes spontaneous secondary emission (centered around approximately 1550 nm / 1550 nm) resulting from the relaxation of excited ions in one or more doping materials, which provides the desired amplification of the optical signal.

[0063] The optical connector BC may include a weighting unit configured to apply weights to the optical signal, thereby applying a linear amplification pattern to the light field.

[0064] All single-mode fibers can be coupled to a so-called multimode intermediate section (fan-in). VOA-weighted input data can be mixed in the multimode intermediate section to generate a multiply-accumulate (MAC) operation. Figures 6C and 6D illustrate mode propagation and crosstalk with a combination of two input signals and show the mixed signal across the output plane. Figure 6C shows a simulation of mode propagation along a multimode section with one input each at the central core and periphery, and Figure 6D shows a simulation of mode propagation along a multimode section with both inputs at the periphery.

[0065] The nonlinear activation function in the computing unit BFB is provided by the fiber amplifier. To characterize the activation function, a scan was performed for all possible combinations of binary input data and weights, and the probability of receiving a logic "1" at the unit's output was calculated. The activation function was then plotted as the sum of the output probability versus the weight input. The amplification level of the EDFA was used to fine-tune the type of activation function and its bias. Figure 6E shows a sigmoid-like activation function, and Figure 6F shows a RelU-like activation function.

[0066] The inventors used the results of the multimode computing unit BFB to emulate neurons in the ResNet Filter, and then in the neural networks ResNet18 and ResNet50, and simulated these networks on the established benchmarks ImageNet and KITTI. The performance of such multimode processors is shown in the table below. TIFF2022159140000002.tif41170

[0067] In the table, the notation 3×R means three repetitions, which were averaged across each test input. The system's accuracy was 89.9-92.7%, comparable to the state-of-the-art NVidia Titan RTX GPU card (97.3-103% relative accuracy), achieving a 1600x (tasks / second) speedup and a 78x efficiency improvement. The estimated performance of the simulated system was 6.67·10 5 The TOPS efficiency is 14.8 TOPS / watt.

[0068] As described above, a photonic guide unit (PGU) suitable for use in the processing unit of the present invention may be based on Er-doped fibers, particularly erbium-doped multicore fibers (EDMFs). Crosstalk between different cores can be controlled by pumping each core, and it has been previously shown that under appropriate arrangement and weighting, such fibers can function as a neural network [E. Cohen, D. Malka, A. Shemer, A. Shahmoon, Z. Zalevsky, and M. London, "Neural networks within multi-core optic fibers", Sci. Rep. 6, 29080 (2016)].

[0069] The inventors manufacture EDMF by bundling and tapering Er-doped fibers in a manner similar to that of a multimode computing unit. In this regard, Figures 7A to 7D show the configuration and operation of such an EDMF computing unit 15.

[0070] Figures 7A and 7B show side and cross-sectional views, respectively, of a photonic body containing an array of photonic guide units having Er-doped portions / regions. Figure 7C shows the spectra of the computing unit with and without coupling between different cores. For this purpose, the inventors inserted signals with wavelengths close to 1550 nm into two cores and excited one core with a wavelength of 980 nm in different combinations, and measured the coupled output spectra. The output shows a large coupling between the two input signals. Figure 7D shows an EDMF-based computing unit BFB with a network realized by core doping. According to this network architecture, the computing unit is connected to doped fiber F doped and undoped fiber F undopedIt includes many core / photonic guide units (PGUs) formed by [unclear]. Therefore, in some embodiments, fiber-based photonic computing units that can function as components of an incoherent photonic neural network can be formed by [unclear] EDMF, or by a combination of EDMF and undoped fiber. Such networks, though still large (1-3 standard 19' communication racks), provide supercomputing capabilities and achieve computing efficiencies exceeding 1,000 TOPS / watt with precision comparable to state-of-the-art systems.

Claims

1. 1. A photonic computing system, comprising: an array of a predetermined number of photonic processing units, each of the photonic processing units having an input port and an output port, configured and operable to perform one or more light processing tasks, each of the photonic processing units including an array of photonic guiding units configured to define propagation conditions for a plurality of light fields associated with the one or more light processing tasks; a plurality of optical connectors, each configured as an optical path for connecting input and output ports of adjacent photonic processing units in the array of photonic processing units to perform light field to light field coupling between the input and output ports; The photonic processing units, each equipped with an optical connector, are arranged in a spaced apart relationship to form a processor panel, each of the photonic processing units in the array is surrounded by other photonic processing units, except for a photonic processing unit at an end of the array of photonic processing units, and communication between the photonic processing unit and all surrounding photonic processing units is via the optical connector to receive optical carrier information from surrounding photonic processing units or transmit optical carrier information to surrounding photonic processing units, thereby providing a network of the photonic processing units and being configured as a hardware module that can be housed in a network rack.

2. 2. The system of claim 1, The system, wherein the optical connector is configured as a bridge-like connector.

3. The system of claim 2, wherein the optical connector is configured as a U-shaped bridge.

4. In the system according to any one of claims 1 to 3, The optical connector comprises: the optical connectors include at least one optical connector configured to connect an output port of one photonic processing unit to an input port of an adjacent photonic processing unit; the optical connectors include at least one optical connector configured to connect an output port of one photonic processing unit to an input port of two or more other photonic processing units; 23. A system comprising at least one of the following configurations:

5. 5. The system of claim 4, the optical connectors include at least one optical connector configured to connect an output port of one photonic processing unit to an input port of two or more other photonic processing units; A system, wherein the output and input ports of the processing units connected by the at least one optical connector include optical splitting and combining elements.

6. In the system according to any one of claims 1 to 3, A system, wherein the one or more optical connectors include at least one optical connector configured to preserve one or more characteristics of a respective light field.

7. 7. The system of claim 6, The one or more characteristics include at least one of a light polarization characteristic and a light spectral characteristic.

8. In the system according to any one of claims 1 to 3, the one or more processing tasks performed by the photonic processing unit are defined by an amplification pattern that is controllably applied to a light field; The preset number of photonic processing units is determined by the number of light-propagating cores defined by the photonic guiding units. A system characterized by at least one of the following:

9. 9. The system of claim 8, A system characterized in that the configuration of said photonic guiding units is such as to define a number of light propagating cores according to an optimized amplification pattern.

10. 9. The system of claim 8, The system, wherein the optical connector, together with respective input and output ports of the photonic processing unit, is configured and operable to apply a pre-set amplification pattern to a light field.

11. 11. The system of claim 10, The system, wherein the optical connector further comprises a local controller configured and operable to control an applied amplification pattern and to control a readout of the output light.

12. In the system according to any one of claims 1 to 3, The system, wherein the photonic processing unit comprises a photonic body configured to define two or more input ports and two or more output ports.

13. 13. The system of claim 12, The photonic body has a hexagonal cross-section, facets of which are associated with the input port and the output port.

14. In the system according to any one of claims 1 to 3, The system, wherein the input ports include at least one input port including an input synchronization unit, the input synchronization unit including at least one of a clock lock utility configured and operable to perform client-server clock locking, and a clock phase alignment utility configured and operable to perform optical clock phase delay synchronization.

15. In the system according to any one of claims 1 to 3, The system, wherein the input ports include at least one input port configured and operable to perform wavelength conversion.

16. In the system according to any one of claims 1 to 3, The system, wherein the output ports include at least one output port configured and operable to perform reshaping of an output signal.

17. In the system according to any one of claims 1 to 3, The system, characterized in that the photonic guiding unit is configured to define one or more types of light fields: a multimode light field, a multicore light field defined by single mode, multimode or a combination of both optical channels, a light field propagating through an erbium doped core, a light field propagating through a photonic crystal fiber based core, a light field confined in a 2D or 3D photonic crystal.

18. In the system according to any one of claims 1 to 3, 11. A system comprising: at least one photonic processing unit including a photonic lantern fabricated by adiabatically coupling multiple single-mode cores to a multimode core.

19. A network rack, A network rack housing a plurality of hardware modules including one or more photonic computing systems configured as the system according to any one of claims 1 to 3.

Images