System and method for high precision digital optical processing

Abstract

The present invention relates to a digital optical computing processing unit configured to perform analog-digital convolution by processing at least a first electrical digital signal and a second electrical digital signal, configured to perform analog-digital convolution by processing at least the first electrical digital signal, a digital optical computing processing unit configured to perform analog-digital continuous convolution of the first electrical digital signal, the second electrical digital signal and the third electrical digital signal, and a digital optical computing processing unit configured to perform digital-digital multiplication by processing at least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal. A method for hybrid photoelectric convolution between matrices and a method for hybrid photoelectric multiplication of a first digital signal with a second digital signal are also disclosed.

Description

Technical Field

[0001] The present disclosure relates to systems and methods for semi-digital and fully digital high-precision digital optical processing. Background Art

[0002] We stand at the dawn of the intelligent age. Driven by deep learning, machines are capable of learning certain specific tasks exceedingly well, performing them better than humans (e.g., Google's AlphaGo). Deep learning architectures are constantly expanding in complexity and require computational power to process vast amounts of data, primarily for matrix-vector multiplication (MVM). This has led to the development of domain-specific hardware accelerators (e.g., Google's Tensor Processing Unit (TPU)), which feature advanced parallelism for energy efficiency and low latency. However, microelectronics technology is encountering fundamental bottlenecks in speed, energy consumption, heat generation, and interconnect latency that can no longer be addressed through scaling.

[0003] A typical neural network (whether optical or analog) contains linear operations and nonlinear operations. Linear operations include convolutional layers and fully connected layers. Nonlinear operations include activation layers and softmax layers. Most of the computing time (up to about 90%) is spent on linear layers. Therefore, current artificial intelligence (AI) or neural network processors are mainly designed to optimize linear operation processes, such as matrix multiplication. Google's TPU is one of the most advanced AI processors. The core component of Google's TPU is the matrix multiplication unit, which also consumes most of the computing time and power consumption. In addition, the matrix multiplication unit is essentially a multiplier, adder and shifter, which limits the computing speed to the electronic clock frequency, resulting in a computing speed limit of several GHz. When the clock speed is accelerated, timing violations and power consumption soar.

[0004] Matrix multiplication refers to the multiplication between digital signals, and more generally refers to a series of arithmetic logic circuits. Preferably, digital multiplication can be performed by digital circuits that follow a digital multiplier architecture (such as the Baugh-Wooley algorithm, Wallace tree, or Dadda multiplier). However, digital multipliers implemented using digital circuits often require complex architectures, which not only consume high energy but also have low speed. These architectures impose limitations on the processing speed of digital multiplication.

[0005] In this context, photonic integrated circuits (PICs) overcome these obstacles. First, PICs offer higher data transmission efficiency, eliminating the need for charging and discharging metal wires. Second, PICs can utilize optical multiplexing, enabling high bandwidth densities exceeding hundreds of terabits per second on a single chip. Third, the energy consumption of light propagating through a PIC is significantly low relative to the number of operations. PICs are also compatible with mature CMOS technology, enabling large-scale integration and production.

[0006] PIC-based ONNs are physical implementations of artificial neural networks (ANNs) using ultrafast optical integrated components. PIC-based ONNs have demonstrated computational speeds (>10 3 ), energy consumption (>10 2 ) and calculated density (>10 2 Despite this great potential, ONNs also present significant scientific and technological challenges. First, these ONNs are all based on analog computing architectures (analog ONNs), where input data and weights are carried on light intensity. This analog computing architecture faces a major scientific challenge: insufficient signal-to-noise ratio due to accumulated noise and crosstalk in the system. Therefore, at the desired high speed, the intensity of the optical signal is typically limited to about 4 bits of precision. In order to achieve reasonable training convergence and many demanding convolutional neural network (CNN) applications (such as autonomous driving, high-definition image processing and 3D computer vision), 16 bits of computing precision is actually required.

[0007] Secondly, analog ONNs are incompatible with microelectronics and therefore require additional expensive, high-resolution, and energy-consuming data processing, such as digital-to-analog conversion (DAC) and / or analog-to-digital conversion (ADC). Therefore, breakthrough technologies are needed to address challenges in ONNs, including computational accuracy and compatibility with microelectronics. Summary of the Invention

[0008] Therefore, there is a need for an optical neural network that can use a digital hardware architecture that surpasses existing analog ONNs. As disclosed herein, in one aspect, this can be achieved by a digital optical computation processing unit configured to perform analog-to-digital convolution by processing at least a first electrical digital signal and a second electrical digital signal and a first optical signal, preferably comprising at least two wavelengths. The digital optical computation processing unit may include: a first waveguide for carrying a first optical signal; a light source configured to provide the first optical signal, comprising at least two wavelengths, to the first waveguide; at least a first electro-optical converter and a second electro-optical converter, preferably arranged in series and preferably included in the first waveguide, configured to be biased by at least a first weight and a second weight, wherein the at least first electrical digital signal and the second electrical digital signal modulate the first optical signal in the at least first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate the first wavelength and the second electro-optical converter is configured to modulate the second wavelength; and a first optical-to-electrical converter arranged at a first output end of the first waveguide, and wherein the first optical-to-electrical converter is configured to output a first electrical output signal based on the first optical signal.

[0009] The digital optical computation processing unit may include an optical computation unit, which may include: a first waveguide for carrying a first optical signal; at least a first electro-optical converter and a second electro-optical converter, preferably arranged in series and included in the first waveguide, configured to be biased by at least a first weight and a second weight, wherein the at least first electrical digital signal and the second electrical digital signal modulate the first optical signal in the at least first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate the first wavelength and the second electro-optical converter is configured to modulate the second wavelength. The optical computation unit may include at least one optical device and / or at least one electro-optical device to perform analog-to-digital convolution.

[0010] The first and second electro-optical converters are configured to modulate the first optical signal using at least a first electrical digital signal and a second electrical digital signal, thereby enabling the digital optical computing unit to process calculations using electrical digital signals rather than electrical analog signals. Advantageously, electrical digital signals are more robust to noise than analog signals. Therefore, compared to optical processing units that may use electrical analog signals, the signal-to-noise ratio robustness is improved. By using electrical digital signals, the first electrical output can be a first electrical digital output signal, which is easier to detect, process, and determine than analog signals. Furthermore, by using electrical digital signals as electrical digital signal inputs, the resolution of the analog-to-digital converter placed after the first photoelectric converter is freed up by several bits compared to the prior art. By eliminating the need for a high-speed digital-to-analog converter at the input, the digital optical processing unit reduces costs, saves energy, and simplifies system complexity.

[0011] At least the first and second electro-optical converters can be biased or loaded with first and second weights, which can multiply at least the first and second electrical digital signals by the at least first and second weights, respectively. The first optical signal is then modulated at each wavelength by an electrical digital signal, which is the result of a number of multiplications between the at least first and second electrical digital signals and the at least first and second weights in the at least first and second electro-optical converters disposed on the first waveguide. The first optical-to-electrical converter is then configured to output a first electrical output signal based on the first optical signal.

[0012] The first optical-to-electrical converter can advantageously be wavelength-independent. Thus, the first electrical output comprises the sum of a plurality of multiplications of at least the first and second electrical digital signals, respectively, with at least the first and second weights, which can be defined as a convolution operation. The digital optical processing unit is configured to process the convolution operation between the digital signal and the weights, wherein the weights can be analog signals or digital signals that can be converted into analog signals.

[0013] Figure 1A to Figure 1BA schematic diagram of an embodiment of a digital optical computation processing unit configured to perform analog-digital convolution is shown. A first light source (here a multi-wavelength light source) is configured to generate a first optical signal comprising at least two wavelengths and feed the first optical signal into a first waveguide. The first waveguide comprises 9 electro-optical converters connected in series, and more specifically in this embodiment comprises 9 micro-ring resonators. Each micro-ring resonator is biased by a weight, and each micro-ring resonator is configured to modulate the first optical signal using an electrical digital signal, each micro-ring resonator modulating the first optical signal at a different wavelength, in this embodiment, the electrical digital signal has 8-bit precision and represents data from an image. The electrical digital signal is a digital representation of the data input to be processed by the digital optical computation processing unit. In Figure 1A In the embodiment, the data input to be processed is data from an image, preferably information contained in at least one pixel. The electrical digital signal encodes the data input in the form of a binary word. For example, Figure 1A As shown, D1 is a first electrical digital signal, which is a digital representation of the decimal number "39" in the form of an 8-bit length binary word, where "39" can represent the intensity of the corresponding pixel from the image because the image is in grayscale format. Figure 1A As shown, the decimal number "39" is then encoded as the binary word "00100111" which is equal to "39". It will be understood by those skilled in the art that any other digital representation can be achieved by increasing or decreasing the number of bits of the binary word, depending on the resolution that the user of the digital optical computing processing unit wants to achieve. Then, all wavelengths (i.e., λ1-λ2-...-λ9) are added in the first photoelectric converter (specifically a photodetector in this embodiment), which is arranged at the first output end of the first waveguide. The first electrical signal based on the first optical signal is a multi-level signal generated by the addition of 9 wavelengths within each symbol period. As shown Figure 1B As shown, the multi-level signal is then converted into a binary signal, which includes 9 electrical digital signals (ie, D1-D x -...-D9) and the matrix including 9 weights (i.e., W 11 -W xy -…-W 33 ) between kernel matrices. Compared to electrical computations, digital optical processing units can compute convolutions much faster and consume less power. This is because convolution is performed in the optical domain rather than the electrical domain. By utilizing digital electrical signals for convolution operations, improved noise robustness is achieved compared to analog signals (such as those used in optical neural networks).

[0014] On the other hand, the present disclosure relates to a digital optical computation processing unit configured to perform analog-digital continuous convolution by processing at least a first electrical digital signal, a second electrical digital signal, and a third electrical digital signal and preferably a first optical signal comprising at least two wavelengths. The digital optical computation processing unit may include: a first waveguide; a light source configured to provide a first optical signal comprising at least two wavelengths to the first waveguide; a main electro-optical converter, preferably included in the first waveguide, wherein the first optical signal is modulated in the main electro-optical converter by at least the first electrical digital signal; at least a first electro-optical converter and a second electro-optical converter, preferably arranged in series and preferably included in the first waveguide and also arranged in series with the main electro-optical converter, wherein at least the first electro-optical converter and the second electro-optical converter are configured to be biased or modulated or loaded by at least a first weight and a second weight, and wherein at least the first electrical digital signal and the second electrical digital signal are modulated or loaded by at least a first weight and a second weight. The digital signal modulates a first optical signal in at least a first electro-optical converter and a second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate a first wavelength and the second electro-optical converter is configured to modulate a second wavelength; a first waveguide portion is configured to disperse the first optical signal or introduce a wavelength-dependent time delay into the first optical signal, wherein the first waveguide portion may be included in the first waveguide and is preferably arranged before at least the first electro-optical converter and the second electro-optical converter; a first photoelectric converter is preferably arranged at a first output end of the first waveguide, and wherein the first photoelectric converter is configured to output a first electrical output signal based on the first optical signal.

[0015] It will be apparent to those skilled in the art that a digital optical computation processing unit configured to perform analog-digital continuous convolution may include the same features as the digital optical computation processing unit configured to perform analog-digital convolution described herein. At least one difference may be that the first waveguide portion may be configured to disperse the first optical signal or introduce a wavelength-dependent time delay into the first optical signal, thereby enabling the digital optical computation processing unit to advantageously perform continuous convolution between the analog signal and the digital signal.

[0016] By having a first waveguide portion that can be configured to disperse a first optical signal or introduce a wavelength-dependent time delay into the first optical signal, wherein the first waveguide portion is included in the first waveguide and disposed before at least a first electro-optical converter and a second electro-optical converter, at least two wavelengths included in the first optical signal are dispersed or time-delayed by any other known method, thereby being time-delayed with respect to one another. Prior to being time-delayed with respect to one another, at least two wavelengths are modulated by at least a first electrical digital signal using a main electro-optical converter. The at least two wavelengths are then modulated by the same electrical digital signal (i.e., at least the first electrical digital signal) and can further be modulated at the first wavelength and the second wavelength, respectively, within the at least first electro-optical converter and the second electro-optical converter, using at least a second electrical digital signal and a third electrical digital signal. A digital optical computation processing unit configured to perform analog-to-digital continuous convolution can implement a time-delayed interleaving technique. Time-delayed interleaving can allow for continuous processing of data on at least two wavelengths. "Time-delayed interleaving" can be understood as interleaving in the time, wavelength, and / or spatial dimensions. This maximizes the processing speed of the digital optical processing unit, enabling more data to be processed in a given amount of time, or preferably, enabling more convolutions to be performed between analog and digital signals or between digital and digital signals.

[0017] Figure 2 A schematic diagram of an embodiment of a digital optical computation processing unit configured to perform analog-digital continuous convolution is shown. A light source (here a multi-wavelength light source) is configured to generate a first optical signal comprising at least two wavelengths (i.e., λ1-λ2-…-λ9) and feed the first optical signal into a first waveguide. The first waveguide comprises: a main electro-optical converter (i.e., a Mach-Zehnder modulator) connected in series with nine electro-optical converters (more specifically, nine microring resonators in this embodiment) connected in series. The Mach-Zehnder modulator is configured to utilize at least a first electrical digital signal (i.e., d j,k , where j and k are the dimensions of a matrix including j*k electrical digital signals) modulates the first optical signal. Then, at least two wavelengths pass through a first waveguide portion (i.e., a dispersive waveguide) configured to disperse the first optical signal, thereby generating a time delay between each wavelength included in the first optical signal. Then, each wavelength is further weighted by a weight matrix (i.e., w i,j, where i and j are the dimensions of a matrix including i*j weights) modulated. Then, all wavelengths are summed in a first photoelectric converter (specifically, a photodetector in this embodiment), which is arranged at the first output end of the first waveguide. The first electrical signal based on the first optical signal is a multi-level signal generated by summing 9 wavelengths in each symbol period. Then, the multi-level signal is converted into a binary signal, which represents the matrix d j,k With the matrix w i,j The digital optical computation processing unit can advantageously process more convolutions in a faster manner by using a main electro-optical converter (such as Mach-Zehnder) and a high-speed modulated input of a microring resonator, such as performing a different convolution between two sub-matrices in each symbol period of the system. By using a high-speed modulated input of a microring resonator, different weights can also be applied in each symbol period of the system, which makes the digital optical computation processing unit suitable for processing convolutions in the case where the kernel matrix containing the weights may be different when the kernel matrix is ​​convolved with data from the same source.

[0018] In a third aspect, the present disclosure relates to a digital optical computation processing unit configured to perform digital-digital multiplication by processing at least a multiplier and a multiplicand, wherein the multiplicand may include at least a first electrical digital signal and the multiplier may include at least a second electrical digital signal and a third electrical digital signal. The digital optical computing processing unit may include: a first waveguide; a light source configured to provide a first optical signal including at least two wavelengths to the first waveguide; a main electro-optical converter, preferably included in the first waveguide, wherein the first optical signal can be modulated by at least a first electrical digital signal in the main electro-optical converter; at least a first electro-optical converter and a second electro-optical converter, preferably arranged in series and preferably included in the first waveguide, and also arranged in series with the main electro-optical converter, and wherein at least the second electrical digital signal and the third electrical digital signal modulate the first optical signal in at least the first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate the first wavelength and the second electro-optical converter is configured to modulate the second wavelength; a first waveguide portion configured to disperse the first optical signal or introduce a wavelength-dependent time delay into the first optical signal, wherein the first waveguide portion may be included in the first waveguide and preferably arranged after at least the first electro-optical converter and the second electro-optical converter; a first optical-to-electrical converter, preferably arranged at a first output end of the first waveguide, and wherein the first optical-to-electrical converter can be configured to output a first electrical output signal based on the first optical signal.

[0019] It will be apparent to those skilled in the art that a digital optical computation processing unit configured to perform digital-digital multiplication may include the same features as the digital optical computation processing unit configured to perform analog-digital continuous convolution described herein. At least one difference is that the first waveguide portion configured to disperse the first optical signal or preferably introduce a wavelength-dependent time delay in the first optical signal is arranged after at least the first electro-optical converter and the second electro-optical converter. This enables digital multiplication between at least the first electrical digital signal and at least the second electrical digital signal and the third electrical digital signal. In a preferred use case, at least the second electrical digital signal and the third electrical digital signal are elements and / or symbols of a secondary electrical digital signal. The digital multiplication referred to herein is a digital multiplication between at least the first electrical digital signal and the secondary electrical digital signal.

[0020] As defined herein, "digital-digital multiplication" refers to the multiplication of two numbers (preferably, two binary numbers) or two multi-level digital signals / numbers. Advantageously, multi-level digital signals can allow each symbol to carry more data, thereby allowing the multiplication of two numbers to be performed while requiring the digital optical processing unit to process fewer symbols.

[0021] Digital multiplication performed by a digital optical computation processing unit configured to perform digital-to-digital multiplication is superior to digital multiplier architectures because it advantageously utilizes features of electro-optical converters and / or photonic circuits to perform digital-to-digital multiplication. By utilizing at least one wavelength in a first optical signal and a wavelength-independent electro-optical converter, in combination with a wavelength-dependent electro-optical converter and delay, and a wavelength-independent opto-electrical converter, the first optical signal can be independently modulated at each of the at least one wavelengths, with a time delay between the wavelengths generated by the first waveguide portion. This architecture can replicate well-known methods for multiplying two digital words, where the two digital words are preferably binary numbers. Figures 3A to 3B The multiplication between two binary words (i.e., "0110" and "1011") is shown, where the first binary word "0110" needs to be multiplied by each symbol or bit of the second binary word "1011", and each multiplication result needs to be shifted by one symbol or one bit from the previous multiplication result. Adding all the contents together can produce a binary word that represents the result of the multiplication between the two binary words. Advantageously, each multiplication between each symbol or bit of the first binary word and the second binary word can be performed for each wavelength, as shown in FIG. Figure 3B The summation of all these multiplications may be performed by a first optoelectronic converter, which may preferably be wavelength independent.

[0022] Figure 13A schematic diagram of an embodiment of a digital optical processing unit is shown. The digital optical processing unit includes an optical device, such as a light source and an optical computation unit. As described herein, the light source can be configured to generate a first optical signal comprising at least two wavelengths. The optical computation unit can be configured to convert at least a first electrical digital signal and a second electrical digital signal into at least a first optical signal and a second optical signal. Preferably, the optical computation unit can be configured to process a digital signal input (which can be at least a first electrical digital signal and a second electrical digital signal) using a weight control signal (which can be at least a first weight and a second weight). The optical computation unit can include at least two electro-optical conversion units. The at least two electro-optical conversion units can include at least a first electro-optical converter and a second electro-optical converter. The optical computation unit can include a first waveguide, at least a first electro-optical converter and a second electro-optical converter configured to be biased by at least a first weight and a second weight, and / or a main electro-optical converter. The digital optical processing unit also includes a signal post-processing unit, wherein the signal post-processing unit includes at least one analog-to-digital converter (ADC), at least one multiplier, and at least one full adder. The shift circuit can be a combination of any electrical delays to shift binary data left or right. Shift circuits can be of two types: serial-in / serial-out (SISO) and serial-in / parallel-out (SIPO). In a SISO circuit, binary data is shifted in serially (bit by bit) and output in the same fashion. In a SIPO circuit, binary data is shifted in serially but output in parallel (all bits at once). These circuits can use a shift register, which can include flip-flops connected in series. A clock signal can determine when the binary data is shifted, and the direction of the shift (left or right) can be controlled by at least one shift control input. It should be noted that Figure 13 The drawings may or may not be drawn to scale, and elements having similar structures or functions are represented by the same reference numerals throughout the drawings. It should also be noted that Figure 13 It is intended to facilitate the description of the embodiments. Figure 13 It is not intended to be exhaustive or to limit the scope of the invention.

[0023] In one aspect, the present disclosure also relates to a method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, the method comprising the following steps: providing an optical signal having ixj different optical carriers (such as ixj different wavelengths); modulating the optical signal using ixj digital data signals such that the nth optical carrier among the ixj digital data signals is modulated by the nth digital data signal among the ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix; modulating the optical signal using the kernel matrix such that the kernel matrix applies weights to the ixj optical carriers; and summing the ixj optical carriers at each symbol level in a wavelength-independent optoelectronic converter to generate an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

[0024] On the other hand, the present disclosure also relates to a method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, the method comprising the following steps: providing an optical signal having ixj different optical carriers (such as ixj different wavelengths); modulating the optical signal with ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix, and each of the ixj optical carriers is modulated with the ixj digital data signals, thereby obtaining ixj optical data signals; providing a time delay between the ixj optical data signals, such that each of the ixj optical data signals is delayed in time relative to each other; modulating the ixj optical data signals with the kernel matrix, such that the nth element of the kernel matrix applies an nth weight to the nth optical carrier of the ixj optical carriers; summing the ixj optical carriers at each symbol level in a wavelength-independent optoelectronic converter, thereby generating an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

[0025] In one aspect, the present disclosure also relates to a method for hybrid optoelectronic multiplication of a first digital signal comprising j symbols and a second digital signal comprising k symbols, the method comprising the following steps: providing an optical signal having i different optical carriers (such as i different wavelengths), wherein i is greater than or equal to j, and wherein j is greater than or equal to k; modulating the optical signal with a first digital signal such that each of the i optical carriers is modulated with the first digital signal, thereby defining i optical digital signals; modulating each of the i optical digital signals on an nth optical carrier among the i optical carriers with an nth symbol among the k symbols; providing a time delay between the i optical digital signals such that each of the i optical digital signals is separated in time; summing the i optical carriers at each symbol level in a wavelength-independent optoelectronic converter to generate an electrical output signal, wherein the electrical output signal represents a product between the first digital signal and the second digital signal.

[0026] As disclosed herein, convolution may refer to an operation between two matrices in which each element at the same position in the two matrices is multiplied, and all products are added.

[0027] The digital optical calculation processing unit disclosed herein can be configured to perform any convolution and multiplication method disclosed herein. Likewise, the convolution and multiplication method disclosed herein can include any features of the digital optical calculation processing unit disclosed herein.

[0028] In the present disclosure, electrical digital signals may refer to digital binary words. Those skilled in the art will appreciate that electrical digital signals may represent inputs to the digital optical computing processing unit and method disclosed herein. These inputs may be encoded as digital signals, and thus as digital binary words or digital vectors. As described herein, analog signals are inherently susceptible to noise and crosstalk. By using electrical digital signals, better noise immunity and higher computational accuracy may be achieved while being free from the constraints of architectures such as high-speed, advanced digital-to-analog converters used in analog optical computing systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The embodiments and examples will be described in more detail below with reference to the accompanying drawings:

[0030] Figure 1A to Figure 1B shows a schematic diagram of an embodiment of a digital optical computation processing unit configured to perform analog-to-digital convolution,

[0031] Figure 2 shows a schematic diagram of an embodiment of a digital optical computation processing unit configured to perform analog-to-digital continuous convolution,

[0032] Figures 3A to 3CA diagram showing an example of multiplication between two binary numbers and a schematic diagram of an embodiment of a digital optical computation processing unit configured to perform digital-digital multiplication,

[0033] Figures 4A to 4E A schematic diagram of an embodiment of a microring resonator is shown, and graphs are shown showing the transmission curve of the microring resonator with different biases, the transmission loss at a given wavelength, and the residual power after the microring resonator at a given wavelength for different bias voltages,

[0034] Figures 5A to 5H shows a schematic diagram of an embodiment of a microring resonator and shows different graphs showing the resulting optical amplitude of an optical signal modulated by a microring resonator biased with different weights by an electrical digital signal with a peak-to-peak amplitude of Vpp,

[0035] Figures 6A to 6C Figures and schematics showing embodiments of a digital optical processing unit with parallel waveguides,

[0036] Figure 7 shows a schematic diagram of an embodiment of a digital optical processing unit, wherein at least a first electrical digital signal, a second electrical digital signal and a third electrical digital signal are modulated by a carrier wave,

[0037] Figures 8A to 8B shows a graph representing a statistical summary of the correlation between the root mean square error (RMSE) or signal error rate (SER) and the signal-to-noise ratio (SNR) for the PAM-16 modulation scheme, the QAM-16 modulation scheme, and the analog scheme,

[0038] Figures 9A to 9L shows an example of the working principle of the LUT used in the disclosed digital optical processing unit,

[0039] Figure 10 shows a schematic diagram of an embodiment of a digital optical computation processing unit configured to perform analog-to-digital continuous convolution,

[0040] Figures 11A to 11B Figures and schematics showing embodiments of a digital optical processing unit with parallel waveguides,

[0041] Figures 12A to 12G The digital optical processing unit is used to verify the modified National Institute of Standards and Technology (MNIST) dataset.

[0042] Figure 13 A schematic diagram showing an embodiment of a digital light processing unit is shown. DETAILED DESCRIPTION

[0043] The first electrical output signal is a first electrical digital output signal. The first electrical digital output signal may be a multi-level digital output signal. The first electrical digital output may be generated by a first photoelectric converter based on the first optical signal. Since the first optical signal may be modulated by at least the first electrical digital signal and the second electrical digital signal, the first electrical output signal may advantageously be digital. By making the first electrical output signal digital, it is possible to more easily process the first electrical output signal using common reshaping, filtering, or equalization techniques known in the art. These techniques may be implemented using a digital signal processing system. The digital signal processing (DSP) system may include the following features:

[0044] • Equalizer: Compensates for channel impairments such as distortion and attenuation in high-speed communication systems.

[0045] Digital filters: Remove unwanted noise from a signal and extract specific frequency components.

[0046] Clock and data recovery systems: Extract clock signals from high-speed data streams, allowing the data to be accurately sampled and processed.

[0047] Signal generator: Generates high-frequency signals that can be used for testing and calibration of high-speed systems.

[0048] Time Domain Reflectometry: Locates faults in high-speed transmission lines by measuring signal reflections at discontinuities.

[0049] Modulation and demodulation systems: Modulate data onto a high-frequency carrier signal and demodulate the received signal back to baseband.

[0050] Error correction system: Corrects errors that occur during high-speed data transmission due to noise and interference.

[0051] The digital optical computing processing unit may further include at least one signal processing unit configured to convert the first electrical digital output signal into a binary signal. The at least one signal processing unit may include an electrical / electronic circuit capable of converting the digital signal into a binary signal. The digital signal may be a multi-level digital signal, and therefore a processing unit may be required to convert the multi-level digital signal into a binary digital signal. Binary digital signals are more easily processed by common electronic processing units (such as field programmable gate arrays (FPGAs)). The at least one signal processing unit may include an electronic signal processing circuit capable of decoding the multi-level electrical digital signal into a binary signal.

[0052] In one embodiment, the output of the first electro-optical converter is connected to the input of the second electro-optical converter. The first electro-optical converter and the second electro-optical converter may have an input and an output, wherein the input and the output are interconnected. The input and the output may be connected via a waveguide or directly. Preferably, the electro-optical converters are symmetrical, such that the input may be the output and vice versa. In a preferred embodiment, "series arrangement" may mean that the output of the first electro-optical converter is connected to the input of the second electro-optical converter.

[0053] In this disclosure, the term "series arrangement" should be understood in its broadest sense. It can be understood as a configuration in which components or items are connected end-to-end in a single line or sequence. A sequence can be understood as a chain-like structure, where components or items are connected in the same chain, but not necessarily directly connected to each other. For example, a first electro-optical converter can be arranged in series with a main electro-optical converter, but they are not necessarily connected to each other. Instead, they share the same optical signal, namely, the first optical signal on the first waveguide.

[0054] At least the first electro-optical converter and the second electro-optical converter can be arranged in parallel. The first waveguide comprising at least two wavelengths can be divided into at least two secondary waveguides, wherein each secondary waveguide comprises one of the at least two wavelengths. Since the at least first electro-optical converter and the second electro-optical converter are respectively arranged on one of the secondary waveguides, the at least first electro-optical converter and the second electro-optical converter can be wavelength-independent electro-optical converters. The secondary waveguides can then be recombined into the first output waveguide by combining the at least two wavelengths in the first output waveguide. The recombination of the at least two wavelengths can be handled by an arrayed waveguide grating, or preferably by a photonic wavelength demultiplexer. The separation of the at least two wavelengths can be handled by an arrayed waveguide grating, or preferably by a photonic wavelength demultiplexer.

[0055] In another embodiment, the primary electro-optical converter is wavelength-independent. The primary electro-optical converter can be configured to modulate all wavelengths included in the optical signal passing through the primary electro-optical converter with the same information / data. The optical signal can include at least two wavelengths, such as two, ten, fifty, or more wavelengths. The primary electro-optical converter can advantageously be a Mach-Zehnder modulator (MZM) or an electro-absorption modulator (EAM). An MZM is an interference structure made of materials with a strong electro-optic effect, such as LiNbO3, GaAs, or InP. Applying an electric field to an arm modulates the optical path length of the optical signal passing through the arm, thereby generating phase modulation. Combining two arms with different phase modulations can convert the phase modulation into intensity modulation. This intensity modulation can be wavelength-independent. An EAM can typically be composed of a semiconductor waveguide doped with a material exhibiting electro-absorption, such as indium gallium arsenide / indium phosphide (InGaAs / InP) or gallium arsenide / aluminum gallium arsenide (GaAs / AlGaAs). In an EAM, an optical signal is passed through the waveguide, and an electrical bias voltage is applied across the waveguide to modulate the material's absorption. This causes the material's refractive index to change in response to an applied voltage, which in turn modulates the intensity of the optical signal passing through the waveguide. Advantageously, the intensity modulation can be independent of wavelength. MZMs may be well-suited for high-speed modulation and low-power applications, while EAMs may be ideal for low-voltage modulation, simple biasing, and temperature control.

[0056] At least the first electro-optical converter and the second electro-optical converter may be related to wavelength. At least the first electro-optical converter and the second electro-optical converter may be configured to modulate a specific wavelength group included in the optical signal passing through each of the first electro-optical converter and the second electro-optical converter. Preferably, the specific wavelength group of each of at least the first electro-optical converter and the second electro-optical converter may not overlap. This will allow the digital optical computing processing unit to specifically modulate the specific wavelength group using different electrical digital signals (such as at least the first electrical digital signal, the second electrical digital signal and / or the third electrical digital signal described herein). At least the first electro-optical converter and the second electro-optical converter may be configured to modulate a certain range of wavelengths. The wavelengths of this certain range may be defined as a plurality of wavelengths close to each other. The wavelengths of this certain range may be defined as a wavelength band of a wavelength spectrum (such as the wavelength spectrum used in the optical waveguide defined in this application). Preferably, the wavelength may be included between 1.3 and 1.6 μm, more preferably included between 1.1 and 1.7 μm.

[0057] At least the first electro-optical converter and the second electro-optical converter can be configured to have a passband or E / O bandwidth included between 30 and 50 GHz, preferably between 35 and 45 GHz, more preferably between 37 and 43 GHz, and even more preferably between 39 and 41 GHz. The term "passband" can be described relative to the optical carrier, so the term "passband" is used in this patent application. The term "E / O bandwidth" can be described as the inherent bandwidth of at least the first electro-optical converter and the second electro-optical converter, which will measure how quickly the at least first electro-optical converter and the second electro-optical converter can respond to changes in the electrical input and generate corresponding optical output. The passband of at least the first electro-optical converter and the second electro-optical converter can be defined based on frequency or wavelength. The wavelength is inversely proportional to the frequency of the wave. Therefore, the passband wavelength of at least the first electro-optical converter and the second electro-optical converter can be calculated using the following equation:

[0058]

[0059] Where c is the speed of light in a vacuum, n(λ0) is the refractive index of the medium at wavelength λ0, and f is the frequency of the wave. The wavelength range defined herein may refer to the passband bandwidth as defined herein. The refractive index of silicon may be around 3.4 to 3.8. Microring resonators can be configured to slightly alter the refractive index of the medium, enabling better wavelength selectivity.

[0060] The at least first and second electro-optical converters may be at least first and second microring resonators or at least first and second cascaded Mach-Zehnders.

[0061] Cascaded Mach-Zehnder devices can be designed to function as wavelength filters by exploiting the wavelength-dependent phase shifts that occur within the device. When light of different wavelengths passes through a material or optical component, the light can experience different phase shifts depending on the wavelength. By selecting the length and refractive index of the waveguides in each MZI stage, a wavelength-dependent phase shift can be generated that causes the device to transmit or reflect light of a specific wavelength. For example, a cascaded Mach-Zehnder device can be designed as a bandpass filter, where only a specific range of wavelengths is transmitted while reflecting or attenuating other wavelengths. This can be achieved by selecting the length and refractive index of the waveguides in each stage so that the phase shift accumulates to a maximum for the desired wavelength while canceling out at other wavelengths. Alternatively, a cascaded MZI can be designed as a notch filter, which transmits most wavelengths but blocks a specific narrow range of wavelengths. This is achieved by adding a phase shift that is out of phase with the desired wavelength, thereby creating destructive interference that cancels out the transmission of that wavelength.

[0062] A microring resonator is a passive device consisting of a waveguide loop coupled to a waveguide, such as the first waveguide disclosed herein. A microring resonator uses the principle of resonant wavelength filtering to selectively enhance or suppress specific wavelengths of an optical signal. When light is coupled into the first waveguide, a portion of the light is coupled into the microring, where it circulates around the ring. The resonant conditions of the microring determine which wavelengths are transmitted or reflected from the device.

[0063] By applying an electrical signal to a microring resonator, the resonant conditions of the device can be changed, which allows for the selective modulation of specific wavelengths of light. This makes microring resonators useful in wavelength-selective switching, filtering, and modulation applications. Microring resonators can also be used to modulate groups of wavelengths, not just individual wavelengths.

[0064] At least the first microring resonator and the second microring resonator include at least a first heater modulator and a second heater modulator. In the case of microrings, the heater modulator can be included in the microring to adjust the wavelength selectivity. The heater modulator can be a metal heater layer placed on top of the waveguide with a dielectric layer in between. The heater modulator can be limited by the so-called RC effect. Therefore, the RC effect can limit the modulation of the heater modulator to a modulation rate in the kHz range. The heater modulator can be used to heat the metal layer, thereby affecting the refractive index n of the material. This phenomenon is called the thermo-optic effect. As the temperature increases, the refractive index of silicon increases, and silicon can be a material used for the waveguides disclosed herein (such as the first waveguide). The refractive index can switch the resonant condition of the heater, thereby switching the wavelength selectivity. This can allow certain wavelengths to pass through the microring resonator while some other wavelengths are blocked. Figure 4A A schematic diagram of an embodiment of a microring resonator is shown. The microring resonator includes a heater modulator and an RF modulator. The RF modulator can be based on the plasma diffusion effect of the PN junction included in the microring resonator. By applying a high-speed varying electric field to the doped silicon waveguide, high-speed modulation of carriers can be achieved, thereby changing the refractive index of the material at a high rate and ultimately obtaining a high-speed modulated optical signal based on the high-speed electrical signal. The PN junction of the microring resonator can be processed, for example, so that the capacitance between the p-junction and the n-junction is minimized, such that the bandwidth of the RF modulator can be maximized.

[0065] The heater modulators of at least the first electro-optical converter and the second electro-optical converter can be configured to be controlled by at least a first weight and a second weight. The weights can be named according to neural network terminology, where weights refer to a set of parameters that can be learned during a training process to adjust the behavior of the network. A neural network can be composed of multiple layers of interconnected nodes or neurons that process and transmit information. Between the input layer and the output layer, there can be one or more hidden layers that perform complex calculations on the input data. Each neuron in a layer receives input from a neuron in the previous layer, applies a mathematical function to the input, and generates an output signal that is passed to the next layer. The weights in a neural network represent the strength of the connection between neurons. Each neuron in a layer is connected to each neuron in the next layer, and each connection has an associated weight. During the training process, the network adjusts these weights to improve its performance on specific tasks (such as image recognition or language translation).

[0066] The weights are initialized with random values ​​at the beginning of the training process and then iteratively updated using an optimization algorithm such as gradient descent. The goal of the optimization algorithm is to minimize a loss function, which measures the difference between the network's predictions and the true values. By adjusting the weights, the network learns to make more accurate predictions on the training data and generalize to new, unseen data. Figures 4B to 4C Graphs showing microring resonator transmission curves with different biases are shown. Figure 4D shows the transmission loss at a given wavelength, Figure 4E The residual power after the microring resonator at a given wavelength for different bias voltages is shown. Figure 4B shows an example of a transmission curve for a microring resonator with a given ring length, Figure 4C The effect of different bias voltages on the microring resonator is shown. Different bias voltages shift the resonance of the microring with respect to wavelength. Resonance is defined as the region on the transmission curve where the intensity of the optical power transmission reaches a minimum. Figure 4D As shown, it may be advantageous to set the bias voltage of the microring resonator so that the wavelength at which the microring can modulate the first optical signal has a maximum intensity and a minimum intensity. Figure 4E Shown is the residual power at the output of the microring resonator at a given wavelength when the bias voltage of the heater modulator is switched between -0.4 V and 0.4 V.

[0067] In a preferred embodiment, at least the first weight and the second weight are normalized and comprised between -1 and +1. Figure 5AA schematic diagram of an embodiment of a microring resonator is shown. The heater modulator is controlled or biased by a command signal designated W, which may preferably be an analog signal. W may represent "weight," and this is what is referred to as "weight" in the present disclosure. X is a high-speed signal used to modulate the optical signal passing through the microring resonator. The optical signal is a CW signal, wherein the optical signal is a constant optical signal with a fixed amplitude at the input, such that the optical signal is modulated by the high-speed signal X at the output and is also biased by the weight W. Therefore, the optical signal at the output of the microring resonator is referred to as Y, where Y=WX. Figure 5B A signal X is shown in a graph representing the amplitude of the signal X over time, where the peak-to-peak amplitude of X is Vpp. 0 and 1 are logical binary states of the signal X. Figure 5C and Figure 5D An example is shown where W has a first positive value. The resulting modulated optical signal has Figure 5C The upper point is located on the transmission curve of the microring resonator when the high-speed signal X becomes more positive compared to the initial state. Figure 5E and Figure 5F An example is shown where W has a second positive value. The resulting modulated optical signal has Figure 5E It can be seen that the peak-to-peak amplitude of the modulated optical signal generated by the second positive value of W is lower than Figure 5D The peak-to-peak amplitude of the modulated optical signal in . Figure 5G and Figure 5H An example is shown where W has a negative value. Figure 5H As shown, the polarity of the modulated optical signal is then flipped, with the amplitude corresponding to Figure 5G The distance between the two points seen on the image. By adjusting the voltage applied to the heater modulator, the amplitude of the modulated light signal can be controlled. As will be understood by those skilled in the art, the command signal W, defined as a weight, has a maximum value and a minimum value. The maximum value can be normalized to +1, while the minimum value can be normalized to -1. Therefore, at least the first and second weights described herein can be normalized and included between -1 and +1.

[0068] At least the first weight and the second weight can be represented by at least a first electrical digital weight signal and a second electrical digital weight signal. The first electrical digital weight signal and the second electrical digital weight signal can carry weight values ​​in the digital weight. The information can be digitally encoded using a multi-level signal or a binary signal, and the information can be implemented in a weight processing unit in the system. The weight processing unit can be configured to decode at least the first electrical digital weight signal and the second electrical digital weight signal into at least a first signal and a second signal, which are suitable for controlling the heater modulator in a configuration required by the user or the system. The digital optical calculation processing unit may also include a digital-to-analog converter (DAC) configured to convert at least the first electrical digital weight signal and the second electrical digital weight signal into at least a first electrical analog weight signal and a second electrical analog weight signal. By converting at least the first digital weight signal and the second digital weight signal into at least a first electrical analog weight signal and a second electrical analog weight signal, control of the heater modulator can be achieved. Since the heater modulator is configured to heat the metal layer, the heater modulator can be controlled by an analog signal. Therefore, an analog signal may be more suitable than a digital signal.

[0069] In one embodiment, at least the first electrical analog weight signal and the second electrical analog weight signal are at least a first bias signal and a second bias signal. At least the first bias signal and the second bias signal can be configured to provide at least a first bias and a second bias to the heater modulator to better fine-tune the bandwidth of the microring resonator and define the operating point of the microring resonator. The operating point can define the peak-to-peak value of the optical signal modulated by the electrical digital signal or the electrical analog signal in the microring resonator. At least the first bias signal and the second bias signal can be configured to bias at least a first heater modulator and a second heater modulator of at least a first electro-optical converter and a second electro-optical converter, respectively. At least the first heater modulator and the second heater modulator are configured to modulate at least a first amplitude and a second amplitude of a first wavelength and a second wavelength of the first optical signal, respectively.

[0070] At least the first weight and the second weight can be weights of a weight matrix of a neural network. In the context of a neural network, a weight matrix can be a mathematical representation of the connections between neurons in two adjacent layers of the network. A weight matrix can be a rectangular array that can specify the strength and direction of the connection between each neuron in the current layer and each neuron in the next layer. For example, consider a neural network with an input layer of 10 neurons, a hidden layer of 5 neurons, and an output layer of 3 neurons. The weight matrix between the input layer and the hidden layer would be a 5x10 matrix, where each row would correspond to a neuron in the hidden layer and each column would correspond to a neuron in the input layer. Similarly, the weight matrix between the hidden layer and the output layer would be a 3x5 matrix. During training, the values ​​in the weight matrix can be adjusted to optimize the performance of the network on a specific task, such as image recognition. The values ​​in the weight matrix can determine the output of each neuron in the next layer based on the input from the previous layer. In convolutional neural networks or recurrent neural networks, the weight matrix can have a more specialized structure that reflects the specific characteristics of the data being processed.

[0071] At least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal can represent at least first pixel information and second pixel information of an image. In the context of image recognition and / or processing, a digital optical computing processing unit configured to perform analog-digital convolution can be used to perform convolution between digital data representing an image and a weight matrix representing a convolution matrix or kernel. In image processing, a kernel, convolution matrix, or mask is a small matrix used for blurring, sharpening, embossing, edge detection, and the like. This can be accomplished by performing a convolution between the kernel and the image. As described herein, the convolution between the kernel and the image can be achieved by a digital optical computing processing unit configured to perform analog-digital convolution. By using a digital optical computing processing unit, noise can be reduced, thereby obtaining better results than analog.

[0072] At least the first electrical digital signal, the second electrical digital signal and / or the third electrical digital signal may include at least one symbol. At least the first electrical digital signal, the second electrical digital signal and / or the third electrical digital signal may be a multi-level electrical digital signal or a binary signal. Each of them may include at least one symbol, such as 4, 8 or 16 symbols. At least the first electrical digital signal, the second electrical digital signal and / or the third electrical digital signal may be a binary signal. At least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal may represent at least the first input, the second input and the third input, wherein each of at least the first input, the second input and the third input is encoded in a binary word. By encoding at least the first input, the second input and the third input in a binary word, a digital-to-analog converter for the input is not required, and the resolution of the analog-to-digital converter included in the signal post-processing unit can be greatly reduced. The analog optical matrix multiplication system preferably relies on an analog photon multiplication core, wherein the input and weights are derived from an M-bit DAC and an N-bit DAC in the electrical domain. If the order of the matrix is ​​L, a matrix with Lx(2 M -1)x(2 N-1) possible levels of an optical signal. The present disclosure discloses systems and methods in which the input is an electrical digital signal (preferably encoded in a binary word). If the input is encoded in a binary word, an optical signal having Lx(2N-1) possible levels is generated. By encoding the input as a binary word, where the input can be at least a first electrical digital signal, a second electrical digital signal, and a third electrical digital signal, the requirement for the ADC included in the signal post-processing unit can be relaxed to M bits. By eliminating the need for a high-speed digital-to-analog converter at the input, the digital optical processing unit reduces cost, saves energy, and simplifies system complexity. At least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal can represent the input encoded in a binary word, where the length of the binary word can be 8 bits, preferably M bits, where M can be between 2 and 256, preferably between 8 and 128, more preferably between 8 and 1024, and even more preferably between 2 and 2048. In the context of image processing and / or recognition, at least the first electrical digital signal, the second electrical digital signal, and / or the third electrical digital signal can represent pixels of an image. The number of different colors a pixel can represent may depend on the number of bits per pixel (bpp). A 1bpp image uses 1 bit per pixel, so each pixel can be either on or off. Each additional bit doubles the number of available colors, so a 2bpp image can have 4 colors, a 3bpp image can have 8 colors, 8bpp can have 256 colors, 16bpp can have 65,536 colors (also known as high color), and 24bpp can have 16,777,216 colors (also known as true color). For color depths of 15 bits per pixel or more, the depth can typically be the sum of the bits allocated to each of the red, green, and blue components. High color (usually 16bpp) typically has 5 bits each for red and blue, and 6 bits for green, because the human eye is more sensitive to errors in green than the other two primary colors. For applications involving transparency, the 16 bits can be divided into 5 bits each for red, green, and blue, leaving 1 bit for transparency. A 24-bit depth allows for 8 bits per component. On some systems, 32-bit depth is available: this means that each 24-bit pixel has an extra 8 bits to describe its opacity, e.g. for the purpose of combining with another image.

[0073] At least one symbol is processed at a certain symbol data rate and sent to a digital optical computation processing unit. The symbol data rate can be defined based on the bandwidth of the digital optical computation processing unit. Generally speaking, the bandwidth of a system may be limited by the functions / devices included in the system. The electro-optical converter and / or the optical-electrical converter may have an inherent bandwidth, which will limit the symbol data rate that can be achieved by the system. Compared to digital multiplier implementations, the use of silicon photonics can maintain a higher bandwidth, which can allow data to be processed at a higher symbol data rate. At least one symbol included in the first electrical digital signal, the second electrical digital signal, and / or the third electrical digital signal can serve as an input to the radio frequency modulator of the microring resonator. Therefore, the symbol data rate can be maximized based on the bandwidth of the radio frequency modulator of the microring modulator. Preferably, the symbol data rate can be half the bandwidth of the radio frequency modulator of the microring modulator, more preferably two-thirds the bandwidth of the radio frequency modulator, and even more preferably three-quarters the bandwidth of the radio frequency modulator.

[0074] In one embodiment, the first waveguide portion is configured to introduce a wavelength-dependent time delay. A wavelength-dependent time delay is a time delay that depends on the wavelength of the optical signal. If the optical signal includes at least two wavelengths, the first waveguide portion can introduce a time delay that depends on each of the at least two wavelengths included in the optical signal. Therefore, if both wavelengths are included in the optical signal, one wavelength can be delayed by a different time delay than the other wavelength.

[0075] In a preferred embodiment, the first waveguide section is a dispersive waveguide section. A dispersive waveguide is designed to support the propagation of dispersive waves, which travel at different speeds depending on their wavelength or frequency. A dispersive waveguide typically consists of a core region surrounded by a cladding region, both of which can be made of materials with different refractive indices. The core region can be designed to confine the light field, while the cladding region provides a boundary to keep the light within the waveguide. When a light signal is introduced into a waveguide, it experiences a change in speed as it travels through the dispersive medium. This change in speed is due to the dispersive properties of the medium, which can cause different frequency components of the signal to propagate at different speeds. As a result, the signal is distorted and spreads out in time and space as it travels through the waveguide. In a dispersive waveguide, the amount of dispersion can be controlled by adjusting the waveguide design, such as the dimensions of the core and cladding regions or the materials used. By carefully controlling dispersion, various useful effects can be achieved, such as pulse compression, time delay, or frequency filtering. The dispersive waveguide portion can be a photonic crystal or a chirped Bragg grating.

[0076] A chirped Bragg grating (CBG) is an optical device that can be used to manipulate the spectral properties of light. It is essentially a Bragg grating, a periodic structure that reflects light of a specific wavelength while allowing light of other wavelengths to pass through, but with a spatially varying periodicity. A Bragg grating can typically be composed of alternating layers of high-refractive-index and low-refractive-index materials, such as silicon dioxide and germanium. These layers act as a periodic refractive-index structure, which can produce wavelength-dependent light reflection. In a conventional Bragg grating, the period of the structure can be constant across the entire grating length, and the reflected wavelength can be fixed. In contrast, the period of a chirped Bragg grating varies along its length, which can cause the reflected wavelength to shift depending on the position. The period of the grating can typically increase or decrease linearly along the length of the grating, resulting in a linear chirp in the reflected wavelength. This allows a CBG to reflect a range of wavelengths, rather than just a single wavelength.

[0077] The dispersive waveguide portion can be configured to generate a time delay between at least two wavelengths. The time delay can be related to a symbol data period, wherein the symbol data period is the inverse of the symbol data rate described herein.

[0078] Preferably, those skilled in the art will realize any method or system suitable for dispersing the first optical signal so as to generate a time delay between at least two wavelengths.

[0079] The time delay is 100% of the symbol data period, preferably included between 100% and 110% of the symbol data period, more preferably included between 100% and 150% of the symbol data period, even more preferably included between 100% and 200% of the symbol data period. Figure 3C As shown, it may be advantageous to have a time delay close to the symbol data period, for example, so that the summing of different wavelengths in the photodetector can be performed simultaneously in the same clock cycle.

[0080] The light source may be a laser source. The laser source may be configured to generate a first optical signal. The first optical signal may include at least two wavelengths. There are several methods for generating an optical signal including at least two wavelengths: Optical frequency comb: An optical frequency comb may be a tool capable of generating a spectrum of equally spaced frequency components.

[0081] This can be accomplished using a mode-locked laser, which emits ultrashort pulses that are tightly spaced in time. When these pulses propagate through a dispersive element (such as an optical fiber or a grating), which can be included in the optical source, they spread out over time and form a series of equally spaced frequency components. By filtering and amplifying the desired frequency component, an optical signal comprising multiple wavelengths can be generated.

[0082] Supercontinuum generation: Supercontinuum generation involves propagating intense optical pulses through a nonlinear medium, such as a photonic crystal fiber. The pulses are broadened due to self-phase modulation, which causes different frequency components to travel at different speeds, leading to a broadening of the spectrum. This broadened spectrum can be used to generate optical signals comprising multiple wavelengths.

[0083] The digital optical computing processing unit may also include a phase modulator. Using a phase modulator, multiple wavelengths can be generated from a laser source. A phase modulator is a device that can be used to modulate the phase of an optical signal generated by a laser source. To generate an optical signal comprising at least two wavelengths from a laser source including a phase modulator, several steps are performed. The laser source emits light of a single wavelength, and a phase modulator is inserted into the path of the laser beam. The modulator changes the phase of the laser light in a controlled manner, and the modulated laser beam is directed to a mixer, which combines the modulated signal with the original signal to generate a new frequency. The output of the mixer is a beam containing multiple wavelengths. The process of generating multiple wavelengths using a laser source and a phase modulator is called frequency comb generation.

[0084] A laser source and a phase modulator can generate a frequency comb. A frequency comb can consist of a series of equally spaced spectral lines (or "teeth") that are evenly spaced in frequency. Each tooth in the comb corresponds to a specific frequency, or wavelength, of light. The spacing between the teeth is very precise, typically on the order of tens of gigahertz.

[0085] The at least two wavelengths are separated from each other by a frequency interval. The frequencies between each of the at least two wavelengths can be equal or different. For example, the two wavelengths can be separated by a first frequency interval, while two other wavelengths included in the same optical signal can be separated by a second frequency interval, where the first frequency interval and the second frequency interval are different. The first frequency interval can be equal to the second frequency interval. In this specific case, the frequency interval can be defined as a comb-like frequency interval.

[0086] At least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal may be multi-level digital signals and / or binary signals.

[0087] In a preferred embodiment, the digital optical computing processing unit further comprises a carrier generator. The carrier generator may be configured to generate at least one carrier wave. Each of the at least one carrier wave has an oscillation frequency. The oscillation frequency may be comprised between 1 Hz and a maximum allowable oscillation frequency, which may be defined by the following equation:

[0088]

[0089] Where BWOE is the bandwidth of the optical-to-electrical converter used in the digital optical computing processing unit, and BWsig is the bandwidth of the electrical digital signal used to modulate the optical signal. The maximum oscillation frequency can be found according to equation (2), where a few parameters of the system can modify the value of the maximum oscillation frequency.

[0090] The carrier generator can be an electronic oscillator, such as a feedback oscillator or a negative resistance oscillator. A feedback oscillator is an electronic oscillator that generates a sinusoidal waveform by feeding part of the output signal back to the input of an amplifier. The feedback loop is designed to provide positive feedback, meaning that the output signal is in phase with the input signal and amplifies it. The circuit can be based on active components such as transistors or operational amplifiers, or it can be based on passive components such as inductors and capacitors. A common example of a feedback oscillator is the Colpitts oscillator.

[0091] A negative resistance oscillator (NRO) is an electronic oscillator that generates a sinusoidal waveform using a negative resistance device, such as a tunnel diode or Gunn diode. This device creates a resonant circuit with a positive feedback loop, which causes the circuit to oscillate. The negative resistance offsets losses in the circuit and provides the energy needed to sustain oscillation. Because NROs can generate extremely high frequencies, they are often used in microwave applications. An example of a NRO is the Gunn oscillator.

[0092] There are many other types of electronic oscillators that could potentially be used, such as phase-shift oscillators, crystal oscillators, and LC-tank oscillators.

[0093] The digital optical computing processing unit may further include at least one digital-to-analog converter, wherein the at least one digital-to-analog converter may be configured to convert at least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal into at least the first electrical analog signal, the second electrical analog signal, and the third electrical analog signal. At least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal may need to be converted into analog signals in order to be modulated by at least one carrier. The carrier wave may generally be a high-frequency wave with a fixed frequency value, and due to the limitation of the communication frequency source, it may generally be used to up-convert the baseband signal to a fixed frequency value in communication. The baseband signal may generally be low-frequency and have a variable frequency width, such as at least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal. In order to transmit the baseband signal, the baseband signal is multiplied by a high-frequency carrier wave to increase the signal frequency so that it passes through the bandwidth of the communication channel.

[0094] At least the first electrical analog signal, the second electrical analog signal, and the third electrical analog signal can be modulated by at least one carrier wave, thereby generating at least a first carrier-modulated electrical signal, a second carrier-modulated electrical signal, and a third carrier-modulated electrical signal. By modulating at least the first electrical analog signal, the second electrical analog signal, and the third electrical analog signal with at least one carrier wave, some electrical analog signals can be modulated using the first carrier wave, while other electrical analog signals can be modulated using the second carrier wave. This allows the first optical signal to be modulated along with at least the first electrical analog signal, the second electrical analog signal, and the third electrical analog signal on at least one carrier wave, thereby increasing the bandwidth of the first optical signal. For example, by having two carrier waves, the total bandwidth of the system can be multiplied by two. In other words, for the same bandwidth, the digital optical processing unit can process twice the amount of data.

[0095] At least the first carrier modulated electrical signal, the second carrier modulated electrical signal, and the third carrier modulated electrical signal are modulated in quadrature or orthogonal fashion relative to one another. By modulating at least the first carrier modulated electrical signal, the second carrier modulated electrical signal, and the third carrier modulated electrical signal in quadrature or orthogonal fashion relative to one another, the signal format increases from a one-dimensional format to a two-dimensional format. Therefore, more possible states can be used to represent the signal, such as with quadrature amplitude modulation (QAM).

[0096] Figure 7 A schematic diagram of an embodiment of a digital optical processing unit is shown, wherein at least a first electrical digital signal, a second electrical digital signal, and a third electrical digital signal are modulated by a carrier. An FPGA is arranged at a transmitter (Tx) and outputs at least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal to a mixer, wherein the mixer is configured to modulate at least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal onto a carrier. At least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal are converted by a DAC into at least a first electrical analog signal, a second electrical analog signal, and a third electrical analog signal, and then injected into a microring modulator array, wherein the microring modulator array is configured to modulate an optical signal from a laser source / comb source using at least the first electrical analog signal, the second electrical analog signal, and the third electrical analog signal. An avalanche photodiode (APD) is configured to convert the optical signal modulated by at least the first electrical analog signal, the second electrical analog signal, and the third electrical analog signal into a first electrical output signal. The first electrical output signal is then converted into a first electrical digital output signal by an ADC, and a mixer in combination with a low-pass filter is configured to demodulate the carrier to obtain a baseband signal. The baseband signal is then output to an FPGA at the receiving end (Rx), which is configured to further process the baseband signal. In the digital optical computation processing unit described in the present disclosure, the baseband signal may include a convolution result or a product between two binary words.

[0097] The digital optical computing processing unit may further include an IQ modulation system. The IQ modulation system may be used to perform phase (I) and quadrature (Q) component modulation on at least the first electrical digital signal, the second electrical digital signal, and the third electrical digital signal, or on at least the first carrier modulated electrical signal, the second carrier modulated electrical signal, and the third carrier modulated electrical signal. For example, at least the first electrical digital signal may be phase modulated, while at least the second electrical digital signal and the third electrical digital signal may be quadrature modulated. The IQ modulation system may be configured to perform orthogonal or quadrature modulation on at least the first carrier modulated electrical signal, the second carrier modulated electrical signal, and the third carrier modulated electrical signal.

[0098] Therefore, the first electrical output signal can be modulated according to quadrature amplitude modulation (QAM). The quadrature amplitude modulation can be quadrature phase shift keying (QPSK) modulation, preferably 8-QAM, more preferably 16-QAM, even more preferably 32-QAM, most preferably 64-QAM, further preferably 128-QAM, further more preferably 256-QAM, further even more preferably 512-QAM, further most preferably 2N-QAM, where N is an integer.

[0099] Figure 8A A chart showing a statistical summary of the correlation between the signal-to-noise ratio (SNR) and the signal error rate (SER) for PAM-16 and QAM-16 modulation schemes is shown. When the signal-to-noise ratio (SNR) rises from 15dB to 35dB, the SER values ​​of both PAM-16 and QAM-16 signals drop sharply, with the QAM-16 signal dropping at a steeper rate. When the SNR is greater than 22dB, the SER value of the QAM-16 signal remains at 0. At the same time, the SER of PAM-16 remains relatively large. Unless the SNR exceeds 33dB, the SER can be reduced to near zero. However, systems with a signal-to-noise ratio exceeding 30dB may be extremely difficult to implement. The chart shows the advantages of QAM modulation over PAM modulation, wherein for any SNR, the SER of QAM modulation is superior.

[0100] Figure 8B A graph presenting a statistical summary of the correlation between the root mean square error (RMSE) and SNR for analog, PAM-16, and QAM-16 modulation schemes is shown. Compared to PAM-16 or analog, QAM-16 exhibits an advantage, with the RMSE for QAM-16 consistently lower for any value of SNR. For the same number of levels (i.e., 16 levels), this is a clear advantage of the QAM modulation scheme over both the PAM modulation scheme and the analog scheme. The analog scheme is the least efficient. Advantageously, the present disclosure can utilize only electrical digital signals, thereby improving the SER and / or RMSE of the digital optical processing unit disclosed herein for any SNR.

[0101] The first electrical output signal can be modulated by at least one carrier wave. The first electrical output signal can be processed by an optoelectronic converter to generate the signal based on the first optical signal. Since the first optical signal can be modulated by at least a first electrical digital signal, a second electrical digital signal, and a third electrical digital signal that can be modulated on at least one carrier wave, the first electrical output signal can be modulated by at least one carrier wave. The digital optical computation processing unit may introduce some impairments in the system that may slightly shift the frequency of at least one carrier wave, but may perform some processing to obtain baseband information from the first optical signal.

[0102] The digital optical calculation processing unit may further include an IQ demodulation system. The IQ demodulation system may be configured to demodulate the first electrical output signal into at least a first electrical digital output signal, a second electrical digital output signal, and a third electrical digital output signal.

[0103] The at least one signal processing unit may include an analog-to-digital converter (ADC) configured to convert the first electrical output signal into a first electrical digital output signal.

[0104] At least one signal processing unit may also include a digital signal processing filter and / or an analog or digital equalizer. These features may be included in at least one signal processing unit because they can better reconstruct the signal to better detect the information carried by the first electrical output signal. The first electrical output signal may have significant distortion and / or uneven distribution. The total bandwidth of the digital optical processing unit may also be lower than the required bandwidth. In some cases, the symbol data period may be higher than the total bandwidth required to distinguish the information carried by the first electrical output signal. Preferably, the equalizer (analog and / or digital) can be configured to equalize the bandwidth, thereby equalizing the first electrical signal. By equalizing the first electrical signal, the level can be better distinguished, thereby facilitating signal processing by the at least one signal processing unit.

[0105] The at least one signal processing unit may further include an electrical signal processing unit, wherein the electrical signal processing unit includes at least one multiplier and at least one full adder. The at least one multiplier and at least one full adder may be configured to generate a binary signal from a multi-level electrical digital signal. Figures 6A to 6C A diagram showing different steps of processing by at least one signal processing unit is shown. Figure 6A The first electrical output signal is shown. The first electrical output signal is a multi-level signal. Preferably, the first electrical output signal is split into groups of 8 symbols for processing. Then, as shown in FIG. Figure 6BAs shown, the first electrical output signal is arranged in time, wherein the difference between consecutive symbols is one clock cycle. Then, the first electrical output signal is processed in a PAM decoder, which can convert the PAM signal into an 8-bit digital signal and keep the difference between consecutive symbols being one clock cycle. Then, as shown in FIG. Figure 6C As shown, all 8-bit signals are added using a full adder similar to an 8-bit multiplier. The PAM decoder may include and / or operate in conjunction with a shift-add circuit (such as at least one multiplier and / or at least one full adder described herein). Preferably, the PAM decoder may be an analog-to-digital converter. At least one signal processing unit may include an analog-to-digital converter followed by at least one shift-and-add circuit. The at least one shift-and-add circuit may be at least one shift-add circuit. Those skilled in the art will appreciate that the at least one shift-add circuit is a common electronic circuit such as a shift-add multiplier.

[0106] If NxN 8-bit convolution matrix multiplication were handled by digital circuit logic, perhaps implemented in electronic circuit hardware, then N multipliers and N-1 full adders would be required. However, with the digital optical processing unit disclosed herein, only one digital logic multiplier, implemented in electronic circuit hardware, might be required. This can advantageously save hardware, thereby reducing power consumption.

[0107] Optical neural networks can be used in many fields, such as autonomous driving. Many studies predict that autonomous vehicles will be mature within 10 years. Using images to accurately identify cars and people in real time for traffic monitoring is one of the biggest challenges. Recently, several deep learning methods based on convolutional neural networks (CNNs) have been proposed for real-time recognition and classification in computer vision. The two most representative CNN models are Faster R-CNN and YOLOv3 and several subsequent derivatives. The required convolution kernel size is 1x1 or 3x3. In these cases, the weight values ​​need to have high precision, from 16 bits to 32 bits. For digital optical computing processing units, it can be challenging to process so many possible weight values ​​while maintaining a reasonable number of levels that can be distinguished from each other in multi-level signals.

[0108] Preferably, the digital optical calculation processing unit may further include a lookup table (LUT). The LUT method is widely used in optical communications, in which the LUT method is used to map detection values ​​to actual data values. The LUT method may be relatively simple and may save energy. The LUT may be implemented in a field programmable gate array (FPGA). By implementing the LUT in the FPGA, the LUT is made flexible so that the detection values ​​can be mapped to different actual data values ​​in the LUT table. If a logic circuit is to be used as the LUT, the output of the logic circuit that maps the detection values ​​to the actual data values ​​will be fixed by the architecture and connections of the logic circuit. The FPGA can provide the LUT with flexibility that the logic circuit cannot provide.

[0109] In a preferred embodiment, the LUT is configured to store at least the first weight and the second weight.Preferably, the LUT may be configured to store a digital representation of at least the first weight and the second weight.

[0110] In one embodiment, the LUT is configured to match at least the first weight and the second weight with at least the first normalized weight value and the second normalized weight value.

[0111] Advantageously, at least the first normalized weight value and the second normalized weight value can be arranged so as to be equally spaced from each other within a specific weight range. Thus, the multilevel signal obtained by convolution can have equally spaced levels. This advantageously simplifies level detection of the multilevel signal while enabling the LUT to match at least the first weight and the second weight with at least the first normalized weight value and the second normalized weight value. The LUT can allow for a higher resolution of possible weight values ​​while maintaining the multilevel signal having equally spaced levels.

[0112] At least the first normalized weight value and the second normalized weight value may be distributed between -1 and 1, preferably uniformly distributed between -1 and 1, and more preferably uniformly distributed between 0 and 1. If the normalized weight values ​​are uniformly distributed between 0 and 1, the electrical output signal may be a PAM signal having equal spaces between levels thereof, thereby conforming to the standard of a PAM signal, which may simplify further processing of the electrical output signal in the processing unit.

[0113] Advantageously, a plurality of convolution kernels can be mapped using a LUT. Based on the first electrical output, a plurality of convolution kernel weight values ​​can be mapped, so that at least the first digital signal and the second digital signal can be convolved using the plurality of convolution kernel weight values ​​stored in the memory and then mapped using the LUT.

[0114] Figures 9A to 9L An example of the working principle of the LUT that can be used in the disclosed digital optical processing unit is shown. Figure 9AAs shown, an optical signal is modulated onto two wavelengths, where each of the two wavelengths is modulated using a first electrical digital signal x1 and a second electrical digital signal x2, and where x1 and x2 modulate the optical signal in two microring resonators having weights w1 and w2, respectively. Assume that the amplitude of x1 is the same as the amplitude of x2. Therefore, the first electrical digital signal and the second electrical digital signal have the same peak-to-peak amplitude. The two modulated optical signals are then added to generate a PAM-4 signal, as shown in FIG. Figure 9B If w1=1 and w2=2, the levels of the PAM-4 signal are equally spaced in the voltage level distribution, which can simplify the detection and further signal processing described in this disclosure. The PAM-4 signal can then be processed to convert it into a binary signal. Figure 9C shows the mapping relationship that may exist between weight values ​​and PAM-4 signals, such as Figures 9E to 9F If different levels of the PAM-4 signal are detected, corresponding weight values ​​may be mapped to the corresponding levels. Figure 9G A convolution kernel matrix of size 3x3 is shown. In many applications, it is necessary to process multiple kernels on the same input data. Therefore, Figure 9G Also shown are multiple convolution kernels to be used on the same input data. 1 W. 2 W and 3 Since the normalized weight value can be mapped using the LUT for the PAM-4 signal representing the first electrical output, multiple weight values ​​can be mapped using the LUT for the PAM-4 signal representing the first electrical output, such as Figures 9H to 9I shown. Figures 9J to 9L An example of mapping multiple convolution kernels to a PAM-4 signal representing a first electrical output is shown. A single input address in the LUT can correspond to multiple outputs, where these outputs are weights. The number of weights mapped from a single LUT input can be large, depending on the amount of memory available in the digital optical processing unit. Advantageously, the output data can be of arbitrary length, i.e., 4-bit, 8-bit, 16-bit, 32-bit, or 64-bit digital word lengths.

[0115] With this LUT approach, the electronic components may need to handle LUTs and summation operations without having to handle multiplication operations in matrix multiplication, which may consume the most time and energy in electrical matrix multiplication. Advantageously, all core calculations in a layer can be obtained by processing only one first optical signal modulated by at least a first electrical digital signal and a second electrical digital signal. By updating the weight data stored in the LUT, the weight matrix can be easily adjusted. Advantageously, this may be very suitable for the training process of a neural network (where weight values ​​may change frequently).

[0116] Compared to electronic digital computing solutions such as Google's TPU, a digital optical processing unit including LUTs may only require at least one LUT and adder, but may not require a multiplier, which may exceed the adder in terms of transistor size and power consumption. For example, an 8-bit adder requires 448 CMOS gates, while an 8-bit multiplier requires 3303 CMOS gates, which is 8 times that of a single adder.

[0117] The digital optical computing processing unit configured to perform analog-to-digital convolution can further process at least a fourth electrical digital signal and a fifth electrical digital signal, and the digital optical computing processing unit configured to perform analog-to-digital convolution can also include: a second waveguide; a second light source, configured to provide a second optical signal including at least two wavelengths to the second waveguide; at least a third electro-optical converter and a fourth electro-optical converter arranged in series and included in the second waveguide, configured to be biased by at least a third weight and a fourth weight, wherein at least the fourth electrical digital signal and the fifth electrical digital signal modulate the second optical signal in at least the third electro-optical converter and the fourth electro-optical converter, respectively, and wherein the third electro-optical converter is configured to modulate the third wavelength, and the fourth electro-optical converter is configured to modulate the fourth wavelength; a second photoelectric converter, arranged at the second output end of the second waveguide, and wherein the second photoelectric converter is configured to output a second electrical output signal based on the second optical signal.

[0118] Those skilled in the art will appreciate that the features described in the previous paragraph are similar to those previously described in this disclosure. By having these additional features, the digital optical computation processing unit configured to perform analog-to-digital convolution can further process at least the fourth electrical digital signal and the fifth electrical digital signal. The digital optical computation processing unit configured to perform analog-to-digital convolution can then perform the convolution on two parallel and distinct paths, thereby increasing the speed by a factor of two. This will also apply by having additional groups of similar features, which will increase the processing / computation speed of the digital optical computation processing unit by a factor proportional to the similar features added to the system.

[0119] Figures 11A to 11B Figures and schematics show embodiments of a digital optical processing unit with parallel waveguides. Figure 11A A 3D view of an embodiment of a digital optical processing unit is shown, wherein multiple waveguides are included in the digital optical processing unit to parallelize the convolution and / or multiplication operations described in the present disclosure. Figure 11BA schematic diagram of an embodiment of a digital optical processing unit with parallel waveguides is shown, wherein the main functions are divided into various categories, namely, a multi-wavelength laser, a microring array, a photon detector, and an electrical interface. The multi-wavelength laser is the light source described in this disclosure, the microring array is at least the first electro-optical converter, the second electro-optical converter, the third electro-optical converter, and the fourth electro-optical converter described in this disclosure, the photon detector is the first photoelectric converter and the second photoelectric converter, and the electrical interface is the at least one processing unit and the at least one signal processing unit described in this disclosure.

[0120] Figure 10 A schematic diagram of an embodiment of a digital optical computing processing unit configured to perform analog-digital continuous convolution is shown. A carrier wave oscillating at a frequency f1 is included in the digital optical computing processing unit. An IQ modulation / demodulation system is also included in the digital optical computing processing unit, thereby allowing electrical digital signals to be modulated on the carrier wave and orthogonal to each other. The electrical digital signals are pixels of an image. For example, X1 and X2 are modulated by the carrier wave and are orthogonal to each other, X1 is modulated by cos(f1*2πt), and X2 is modulated by sin(f1*2πt). The electrical digital signals X1, X2, ... X n Respectively with W1, W2, ...W n Multiply, where W1, W2, ...W n are normalized weights, as described in the present disclosure. These signals generated by multiplying the electrical digital signals with the weights modulate the optical signals generated by the comb laser. The product obtained between X1 and W1 modulates the optical signal at one wavelength, the product obtained between X2 and W2 modulates the optical signal at a second wavelength, and so on. The optical signal is detected by a photodiode at the end of the waveguide. The photodiode converts the optical signal into an electrical output signal, which is then processed by a processing unit such as an FPGA and a QAM output is obtained. The LUT can match the obtained normalized weights with the actual weights to generate the resulting digital data. The processing unit then processes the resulting digital data, converting it into binary data to generate the final result.

[0121] The digital optical calculation processing unit configured to perform analog-digital continuous convolution can also process at least a fourth electrical digital signal, a fifth electrical digital signal, and a sixth electrical digital signal, and the digital optical calculation processing unit configured to perform analog-digital continuous convolution can also include: a second waveguide; a second light source configured to provide a second optical signal including at least two wavelengths to the second waveguide; an auxiliary electro-optical converter included in the second waveguide, wherein the second optical signal is modulated by at least the fourth electrical digital signal in the auxiliary electro-optical converter; at least a third electro-optical converter and a fourth electro-optical converter, preferably arranged in series and included in the second waveguide, and also arranged in series with the auxiliary electro-optical converter, wherein at least the third electro-optical converter and the fourth electro-optical converter are configured to be biased or loaded by at least a third weight and a fourth weight, wherein at least a fifth electrical digital signal and a sixth electrical digital signal modulate a second optical signal in at least a third electro-optical converter and a fourth electro-optical converter, respectively, and wherein the third electro-optical converter is configured to modulate a third wavelength, and the fourth electro-optical converter is configured to modulate a fourth wavelength; a second waveguide portion is configured to disperse the second optical signal or introduce a wavelength-dependent time delay into the second optical signal, wherein the second waveguide portion is included in the second waveguide and is preferably arranged before at least the third electro-optical converter and the fourth electro-optical converter; a second optoelectronic converter is arranged at a second output end of the second waveguide, and wherein the second optoelectronic converter is configured to output a second electrical output signal based on the second optical signal.

[0122] The digital optical calculation processing unit configured to perform digital-digital multiplication can also process at least a fourth electrical digital signal, a fifth electrical digital signal and a sixth electrical digital signal, and the digital optical calculation processing unit configured to perform digital-digital multiplication can also include: a second waveguide; a second light source configured to provide a second optical signal including at least two wavelengths to the second waveguide; an auxiliary electro-optical converter included in the second waveguide, wherein the second optical signal is modulated by at least the fourth electrical digital signal in the auxiliary electro-optical converter; at least a third electro-optical converter and a fourth electro-optical converter, preferably arranged in series and included in the second waveguide, and also arranged in series with the auxiliary electro-optical converter, wherein at least the third electro-optical converter and the fourth electro-optical converter can be configured The optical fiber transmission device is configured to bias or load the optical fiber by at least a third weight and a fourth weight, wherein the at least third electrical digital signal and the fourth electrical digital signal modulate the second optical signal in the at least third electro-optical converter and the fourth electro-optical converter, respectively, and wherein the third electro-optical converter is configured to modulate a third wavelength and the fourth electro-optical converter is configured to modulate a fourth wavelength; a second waveguide portion is configured to disperse the second optical signal or introduce a wavelength-dependent time delay into the second optical signal, wherein the second waveguide portion is included in the second waveguide and is arranged after the at least third electro-optical converter and the fourth electro-optical converter; a second optoelectronic converter is arranged at a second output end of the second waveguide, and wherein the second optoelectronic converter is configured to output a second electrical output signal based on the second optical signal.

[0123] The second waveguide may have at least one characteristic or any combination of characteristics described herein with respect to the first waveguide. The second electrical output signal may have at least one characteristic or any combination of characteristics described herein with respect to the first electrical output signal. The auxiliary electro-optical converter may have at least one characteristic or any combination of characteristics described herein with respect to the primary electro-optical converter. The second light source may be the first light source. The first wavelength may be the same wavelength as the third wavelength, and / or the second wavelength may be the same wavelength as the fourth wavelength. The second optical signal may have at least one characteristic or any combination of characteristics described herein with respect to the first optical signal. The at least third and fourth electro-optical converters may have at least one characteristic or any combination of characteristics described herein with respect to the at least first and second electro-optical converters.

[0124] The digital optical computation processing unit may further include an optical splitter configured to split light from a light source. For example, the first optical signal and the second optical signal may be generated from the light source. Preferably, the light source may be configured to provide optical signals to the first waveguide and the second waveguide, wherein the optical signals are the first optical signal and / or the second optical signal. Advantageously, this saves additional light sources when parallelizing multiple waveguides on the same digital optical computation processing unit.

[0125] Preferably, the second light source may be the first light source.

[0126] The at least fourth, fifth, and sixth electrical digital signals can have at least one characteristic or any combination of characteristics described herein with respect to the at least first, second, and third electrical digital signals. The second waveguide portion can have at least one characteristic or any combination of characteristics described herein with respect to the first waveguide portion.

[0127] Figures 12A to 12G Validation of the modified National Institute of Standards and Technology (MNIST) dataset using the DLP is shown. To evaluate the accuracy of the DLP architecture, handwriting recognition and image classification tasks were performed. The MNIST task is a preferred tool for validating image recognition. The task involves distinguishing between 10 handwritten digits for classification. The model is pre-trained, so that the kernel matrix is ​​known, and is implemented in the DLP. Figure 12A The MNIST test dataset of 10,000 handwritten digits is shown for convolutional layer processing, where each image has a pixel resolution of 28x28. The results of the image processing are shown in Figure 12B , where the pixel resolution is reduced to 26x26 due to the stride of the convolution. In this example, four 3x3 kernel matrices are used, resulting in a 26x26x4 pixel matrix for the 10,000 images. The rest of the network is processed using electronic hardware, such as Figure 12C As shown, the decision is made at the electronic hardware. The result is then compared with the result obtained from a processing unit such as a computer. 12D to 12E As shown, the results are basically the same in each prediction classification. The deviation is about one thousandth. Extract the diagonal prediction accuracy data and compare them, as shown in Figure 12F As shown, the diagonal prediction accuracy data are basically the same. Figure 12G A comparison between standard expected results and results calculated by a digital optical processing unit is shown. The resulting distribution converges linearly, with only a few points located outside the distribution. These results demonstrate that the accuracy of the digital optical processing unit is comparable to that of processing units such as computers.

[0128] In one aspect, the present disclosure may relate to a method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, the method comprising the following steps: providing an optical signal having ixj different optical carriers (such as ixj different wavelengths); modulating the optical signal with ixj digital data signals such that an nth optical carrier among the ixj optical carriers is modulated with an nth digital data signal among the ixj digital data signals, wherein the ixj digital data signals are digital representations of ixj elements of the input data matrix; modulating the optical signal with a kernel matrix such that the kernel matrix applies weights to the ixj optical carriers; and summing the ixj optical carriers at each symbol level in a wavelength-independent optoelectronic converter to generate an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

[0129] The optical-to-electrical converter may be wavelength-independent and may be configured to optically convert all wavelengths of an optical signal passing therethrough into electrical signals using the same modulation data.

[0130] In a preferred embodiment, i is greater than j, preferably less than j, and more preferably equal to j.

[0131] The ixj different wavelengths may be equally spaced from each other. The ixj different wavelengths may have different corresponding frequencies, and these different frequencies may define a frequency comb.

[0132] The ixj digital data signals may be binary words.

[0133] The input data matrix may contain coded elements from an image, wherein the coded elements may be digital representations of pixels. The digital representations of pixels may be binary codes. The binary codes may be 2-bit binary codes, preferably 4-bit binary codes, more preferably 8-bit binary codes, even more preferably 16-bit binary codes, most preferably 32-bit binary codes, and even more preferably 64-bit binary codes.

[0134] The kernel matrix may be a convolution matrix, wherein the convolution matrix may be a matrix for blurring, sharpening, embossing or edge detection. The kernel matrix may be a 3x3 matrix, preferably a 4x4 matrix, more preferably a 6x6 matrix, and even more preferably a 16x16 matrix.

[0135] The optical signal may be generated by a laser source. The laser source may be configured to generate at least two optical wavelengths.

[0136] The ixj digital data signals may include at least two symbols. The ixj digital data signals may be binary data signals, preferably multi-level digital data signals, such as PAM signals. PAM signals may include two, four, or more levels. The binary data signals may be binary signals containing data.

[0137] The weight may be comprised between -1 and 1. Preferably, the weight may be any value comprised between -1 and 1, inclusive.

[0138] The wavelength-independent photoelectric converter may be a photodiode. Preferably, the wavelength-independent photoelectric converter may be any type of photodetector.

[0139] The method for hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj may be performed by an optical digital processing unit configured to perform analog-to-digital convolution as described herein.

[0140] In another aspect, the present disclosure may relate to a method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, the method comprising the following steps: providing an optical signal having ixj different optical carriers (such as ixj different wavelengths); modulating the optical signal with ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix, and causing each of the ixj optical carriers to be modulated with the ixj digital data signals, thereby obtaining ixj optical data signals; providing a time delay between the ixj optical data signals, such that each of the ixj optical data signals is delayed in time with respect to each other; modulating the ixj optical data signals with the kernel matrix, such that the nth element of the kernel matrix applies an nth weight to the nth optical carrier of the ixj optical carriers; and summing the ixj optical carriers at each symbol level in a wavelength-independent optoelectronic converter, thereby generating an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

[0141] The optical-to-electrical converter may be wavelength-independent. The optical-to-electrical converter may be configured to modulate all wavelengths of the optical signal passing therethrough with the same modulation data.

[0142] The ixj digital data signals may be binary words.

[0143] i may be greater than j, preferably less than j, and more preferably equal to j.

[0144] The ixj different wavelengths may be equally spaced from each other. The ixj different wavelengths may have different corresponding frequencies, and these different frequencies may define a frequency comb.

[0145] The time delay may be related to the symbol data rate. j symbols and / or k symbols may modulate i optical carriers at the symbol data rate. The symbol data rate may be the inverse of the symbol data period. The time delay may be 100% of the symbol data period, preferably between 100% and 110% of the symbol data period, more preferably between 100% and 150% of the symbol data period, and even more preferably between 100% and 200% of the symbol data period.

[0146] The input data matrix may contain coded elements from an image. The coded elements may be digital representations of pixels. The digital representations of pixels may be binary codes. The binary codes may be 2-bit binary codes, preferably 4-bit binary codes, more preferably 8-bit binary codes, even more preferably 16-bit binary codes, most preferably 32-bit binary codes, and even more preferably 64-bit binary codes.

[0147] The kernel matrix may be a convolution matrix, wherein the convolution matrix may be a matrix for blurring, sharpening, embossing or edge detection. The kernel matrix may be a 3x3 matrix, preferably a 4x4 matrix, more preferably a 5x5 matrix, even more preferably a 6x6 matrix, and most preferably a 16x16 matrix.

[0148] The optical signal may be generated by a laser source. The laser source may be configured to generate at least two optical wavelengths.

[0149] The ixj digital data signals may include at least two symbols. The ixj digital data signals may be binary data signals, preferably multi-level digital data signals, such as PAM signals. PAM signals may include two, four, or more levels. The binary data signals may be binary signals containing data.

[0150] The weight may be comprised between -1 and 1. Preferably, the weight may be any value comprised between -1 and 1, inclusive.

[0151] The wavelength-independent photoelectric converter may be a photodiode. Preferably, the wavelength-independent photoelectric converter may be any type of photodetector.

[0152] The method for hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj may be performed by a digital optical processing unit as described herein configured to perform analog-to-digital continuous convolution.

[0153] In another aspect, the present disclosure may relate to a method for hybrid optoelectronic multiplication of a first digital signal comprising j symbols and a second digital signal comprising k symbols, the method comprising the following steps: providing an optical signal having i different optical carriers (such as i different wavelengths), and wherein i is greater than or equal to j, and wherein j is greater than or equal to k; modulating the optical signal with a first digital signal such that each of the i optical carriers is modulated with the first digital signal, thereby defining i optical digital signals; modulating each of the i optical digital signals with an nth symbol out of k symbols on an nth carrier out of the i optical carriers; providing a time delay between the i optical digital signals such that each of the i optical digital signals is separated in time; and summing the i optical carriers at each symbol level in a wavelength-independent optoelectronic converter to generate an electrical output signal, wherein the electrical output signal represents a product between the first digital signal and the second digital signal.

[0154] The optical-to-electrical converter may be wavelength-independent. The optical-to-electrical converter may be configured to optically convert all wavelengths of an optical signal passing therethrough into electrical signals using the same modulation data.

[0155] The i different wavelengths may be equally spaced from each other. The i different wavelengths may have different corresponding frequencies, and these different frequencies may define a frequency comb.

[0156] The time delay may be related to the symbol data rate. j symbols and / or k symbols may modulate i optical carriers at the symbol data rate. The symbol data rate may be the inverse of the symbol data period. The time delay may be 100% of the symbol data period, preferably between 90% and 110% of the symbol data period, more preferably between 50% and 150% of the symbol data period, and even more preferably between 50% and 200% of the symbol data period.

[0157] The first digital signal and / or the second digital signal may be a multi-level digital signal or a binary signal. Each of the i optical digital signals may be a multi-level digital signal or a binary signal. The optical signal may be generated by a laser source. The laser source may be configured to generate at least two optical wavelengths.

[0158] The first digital signal may be a multiplier and the second digital signal may be a multiplicand. The first digital signal may be a multiplicand and the second digital signal may be a multiplier. The j symbols and / or the k symbols may be multi-level symbols or bits, which may be two-level symbols.

[0159] The wavelength-independent photoelectric converter may be a photodiode. Preferably, the wavelength-independent photoelectric converter may be any type of photodetector.

[0160] The method for hybrid optical-electrical multiplication may be performed by a digital optical processing unit as described herein configured to perform digital-to-digital multiplication.

[0161] More details

[0162] 1. A digital optical computation processing unit configured to perform analog-to-digital convolution by processing at least a first electrical digital signal and a second electrical digital signal and a first optical signal comprising at least two wavelengths, the digital optical computation processing unit comprising:

[0163] a first waveguide, configured to carry a first optical signal;

[0164] At least a first electro-optical converter and a second electro-optical converter included in a first waveguide are configured to be biased by at least a first weight and a second weight, wherein at least a first electrical digital signal and a second electrical digital signal modulate a first optical signal in at least the first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate a first wavelength and the second electro-optical converter is configured to modulate a second wavelength.

[0165] 2. The digital optical computation processing unit according to item 1, further comprising a light source configured to provide a first optical signal comprising at least two wavelengths to the first waveguide.

[0166] 3. The digital optical computation processing unit according to any one of items 1-2, further comprising a first photoelectric converter arranged at the first output end of the first waveguide, and wherein the first photoelectric converter is configured to output a first electrical output signal based on the first optical signal.

[0167] 4. The digital optical computation processing unit according to any one of items 1 to 3, wherein the at least first and second electrical digital signals represent at least the first and second inputs encoded in binary words.

[0168] 5. The digital optical calculation processing unit according to any one of items 1 to 4, wherein at least the first electro-optical converter and the second electro-optical converter are arranged in series.

[0169] 6. A digital optical computation processing unit configured to perform analog-to-digital continuous convolution by processing at least a first electrical digital signal, a second electrical digital signal, and a third electrical digital signal and a first optical signal comprising at least two wavelengths, the digital optical computation processing unit comprising:

[0170] a first waveguide, configured to carry a first optical signal;

[0171] a primary electro-optical converter comprised in the first waveguide, wherein the first optical signal is modulated in the primary electro-optical converter by at least a first electrical digital signal;

[0172] at least a first electro-optical converter and a second electro-optical converter included in the first waveguide and further arranged in series with the main electro-optical converter, wherein the at least first electro-optical converter and the second electro-optical converter are configured to be biased by at least a first weight and a second weight, and wherein the at least first electrical digital signal and the second electrical digital signal modulate the first optical signal in the at least first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate the first wavelength and the second electro-optical converter is configured to modulate the second wavelength;

[0173] The first waveguide section is configured to introduce a wavelength-dependent time delay in the first optical signal.

[0174] 7. The digital optical computation processing unit according to item 6, further comprising a light source configured to provide a first optical signal including at least two wavelengths to the first waveguide.

[0175] 8. The digital optical computation processing unit according to any one of items 6-7, further comprising a first photoelectric converter arranged at the first output end of the first waveguide, and wherein the first photoelectric converter is configured to output a first electrical output signal based on the first optical signal.

[0176] 9. The digital optical computation processing unit according to any one of items 6 to 8, wherein the first waveguide portion is included in the first waveguide and is arranged before at least the first electro-optical converter and the second electro-optical converter and after the main electro-optical converter.

[0177] 10. The digital optical calculation processing unit according to any one of items 6 to 9, wherein at least the first electro-optical converter and the second electro-optical converter are arranged in series.

[0178] 11. A digital optical computation processing unit configured to perform digital-digital multiplication by processing at least a multiplier and a multiplicand and a first optical signal comprising at least two wavelengths, wherein the multiplicand comprises at least a first electrical digital signal and the multiplier comprises at least a second electrical digital signal and a third electrical digital signal, the digital optical computation processing unit comprising:

[0179] a first waveguide, configured to carry a first optical signal;

[0180] a primary electro-optical converter comprised in the first waveguide, wherein the first optical signal is modulated in the primary electro-optical converter by at least a first electrical digital signal;

[0181] at least a first electro-optical converter and a second electro-optical converter included in the first waveguide and further arranged in series with the main electro-optical converter, wherein the at least second electrical digital signal and the third electrical digital signal modulate the first optical signal in the at least first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate the first wavelength and the second electro-optical converter is configured to modulate the second wavelength;

[0182] The first waveguide section is configured to introduce a wavelength-dependent time delay in the first optical signal.

[0183] 12. The digital optical calculation processing unit according to item 11, wherein at least the first electro-optical converter and the second electro-optical converter are arranged in series.

[0184] 13. The digital optical computation processing unit according to item 11, further comprising a light source configured to provide a first optical signal including at least two wavelengths to the first waveguide.

[0185] 14. The digital optical computation processing unit according to any one of items 11 to 13, further comprising a first photoelectric converter arranged at the first output end of the first waveguide, and wherein the first photoelectric converter is configured to output a first electrical output signal based on the first optical signal.

[0186] 15. The digital optical computation processing unit according to any one of items 11 to 14, wherein at least the first electrical digital signal is encoded in binary words.

[0187] 16. The digital optical calculation processing unit according to any one of items 11 to 15, wherein the at least second electrical digital signal and the third electrical digital signal are at least a first bit and a second bit of the multiplier, respectively.

[0188] 17. The digital optical computation processing unit according to any one of items 11 to 16, wherein the wavelength-dependent time delay is a time delay between at least two wavelengths.

[0189] 18. A digital optical computation processing unit according to any one of items 11 to 17, wherein the waveguide portion is comprised in the first waveguide and is arranged after the main electro-optical converter.

[0190] 19. The digital optical computation processing unit according to any one of items 11 to 18, wherein the waveguide portion is included in the first waveguide and is arranged after at least the first electro-optical converter and the second electro-optical converter.

[0191] 20. A digital optical computing processing unit according to any one of items 11-19, wherein at least the first electro-optical converter and the second electro-optical converter are configured to be biased by at least the second electrical digital signal and the third electrical digital signal, thereby modulating the first optical signal using the at least second electrical digital signal and the third electrical digital signal.

[0192] 21. The digital optical computation processing unit according to any one of items 1 to 20, wherein the first electrical output signal is a first electrical digital output signal.

[0193] 22. The digital optical calculation processing unit according to any one of items 1 to 21, wherein the digital optical calculation processing unit further comprises at least one signal processing unit, and the at least one signal processing unit is configured to convert the first electrical digital output signal into a binary signal.

[0194] 23. The digital optical calculation processing unit according to any one of items 1 to 22, wherein the output terminal of the first electro-optical converter is connected to the input terminal of the second electro-optical converter.

[0195] 24. The digital optical computation processing unit according to any one of items 1 to 23, wherein the primary electro-optical converter is wavelength independent.

[0196] 25. The digital optical computation processing unit according to any one of items 1 to 24, wherein the main electro-optical converter is a Mach-Zehnder modulator (MZM) or an electro-absorption modulator (EAM).

[0197] 26. The digital optical computation processing unit according to any one of items 1 to 25, wherein at least the first electro-optical converter and the second electro-optical converter are wavelength dependent.

[0198] 27. A digital optical computation processing unit according to any of the preceding items, wherein at least the first electro-optical converter and the second electro-optical converter are configured to modulate a range of wavelengths.

[0199] 28. A digital optical computing processing unit according to any of the preceding items, wherein at least the first electro-optical converter and the second electro-optical converter are configured to have a passband bandwidth between 5 and 100 GHz, preferably between 10 and 70 GHz, more preferably between 20 and 60 GHz, and even more preferably between 30 and 50 GHz.

[0200] 29. The digital optical computation processing unit according to any of the preceding items, wherein at least the first electro-optical converter and the second electro-optical converter are at least a first microring resonator and a second microring resonator or at least a first cascaded Mach-Zehnder and a second cascaded Mach-Zehnder.

[0201] 30. The digital optical computation processing unit of item 29, wherein at least the first microring resonator and the second microring resonator comprise at least a first heater modulator and a second heater modulator.

[0202] 31. A digital optical calculation processing unit according to any of the preceding items, wherein at least the first weight and the second weight are normalized and comprised between -1 and +1.

[0203] 32. A digital optical computation processing unit according to any of the preceding items, wherein at least the first weight and the second weight are represented by at least a first electrical digital weight signal and a second electrical digital weight signal.

[0204] 33. A digital optical computing and processing unit according to any one of the preceding items, wherein the digital optical computing and processing unit further comprises a digital-to-analog converter (DAC), and the digital-to-analog converter (DAC) is configured to convert at least a first electrical digital weight signal and a second electrical digital weight signal into at least a first electrical analog weight signal and a second electrical analog weight signal.

[0205] 34. The digital optical computation processing unit according to item 33, wherein at least the first electrical analog weight signal and the second electrical analog weight signal are at least a first bias signal and a second bias signal.

[0206] 35. A digital optical computation processing unit according to any of the preceding items, wherein at least the first bias signal and the second bias signal are configured to bias at least the first heater modulator and the second heater modulator of the at least first electro-optical converter and the second electro-optical converter, respectively.

[0207] 36. A digital optical computation processing unit according to any of the preceding items, wherein at least a first heater modulator and a second heater modulator are configured to modulate at least a first amplitude and a second amplitude of a first wavelength and a second wavelength of the first optical signal, respectively.

[0208] 37. A digital optical computation processing unit according to any of the preceding items, wherein at least the first weight and the second weight are weights of a weight matrix of a neural network.

[0209] 38. A digital optical computation processing unit according to any one of items 1 to 37, wherein at least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal represent at least first pixel information and second pixel information of an image.

[0210] 39. The digital optical computation processing unit according to any one of items 1 to 38, wherein at least the first electrical digital signal, the second electrical digital signal and / or the third electrical digital signal comprises at least one symbol.

[0211] 40. The digital optical computation processing unit of item 39, wherein at least one symbol is processed into the digital optical computation processing unit at a symbol data rate.

[0212] 41. A digital optical computation processing unit according to any one of items 1 to 40, wherein the first waveguide section is a dispersive waveguide section.

[0213] 42. The digital optical computing processing unit according to item 41, wherein the dispersive waveguide portion is a photonic crystal or a chirped Bragg grating.

[0214] 43. A digital optical computation processing unit according to any one of items 41-42, wherein the dispersive waveguide section is configured to generate a time delay between at least two wavelengths.

[0215] 44. A digital optical computation processing unit according to any of the preceding items, wherein the time delay is related to a symbol data period, wherein the symbol data period is the inverse of the symbol data rate.

[0216] 45. A digital optical computation processing unit according to item 43 or 44, wherein the time delay is 100% of the symbol data period, preferably included between 90% and 110% of the symbol data period, more preferably included between 50% and 150% of the symbol data period, and even more preferably included between 50% and 200% of the symbol data period.

[0217] 46. ​​The digital optical computing processing unit according to any one of the preceding items, wherein the light source is a laser source.

[0218] 47. The digital optical computation processing unit according to any of the preceding items, wherein the first optical signal comprises at least two wavelengths.

[0219] 48. The digital optical computing and processing unit according to any of the preceding items, wherein the digital optical computing and processing unit further comprises a phase modulator.

[0220] 49. The digital optical computation processing unit of item 48, wherein the phase modulator is configured to generate a frequency comb light source.

[0221] 50. A digital optical computation processing unit according to any of the preceding items, wherein at least two wavelengths are separated from each other by a frequency interval.

[0222] 51. The digital optical calculation processing unit according to item 50, wherein the frequency interval is a comb-shaped frequency interval.

[0223] 52. The digital optical computation processing unit according to any of the preceding items, wherein at least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal are multi-level digital signals and / or binary signals.

[0224] 53. The digital optical calculation processing unit according to any one of the preceding items, wherein the digital optical calculation processing unit further comprises a carrier generator.

[0225] 54. The digital optical computation processing unit according to item 53, wherein the carrier generator is configured to generate at least one carrier.

[0226] 55. The digital optical calculation processing unit according to item 54, wherein each of the at least one carrier wave has an oscillation frequency.

[0227] 56. The digital optical calculation processing unit according to item 55, wherein the oscillation frequency is comprised between 1 Hz and the maximum oscillation frequency.

[0228] 57. The digital optical computation processing unit according to item 53, wherein the carrier generator is an electronic oscillator, such as a feedback oscillator or a negative resistance oscillator.

[0229] 58. The digital optical computation processing unit according to any of the preceding items, wherein at least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal are modulated in quadrature or in an orthogonal manner with respect to each other.

[0230] 59. The digital optical computation processing unit according to any of the preceding items, wherein each of at least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal is modulated by one of at least one carrier wave.

[0231] 60. A digital optical computing and processing unit according to any one of the preceding items, wherein the digital optical computing and processing unit further comprises at least one digital-to-analog converter, wherein the at least one digital-to-analog converter is configured to convert at least a first electrical digital signal, a second electrical digital signal, and a third electrical digital signal into at least a first electrical analog signal, a second electrical analog signal, and a third electrical analog signal.

[0232] 61. A digital optical computing processing unit according to any one of the preceding items, wherein at least the first electrical analog signal, the second electrical analog signal and the third electrical analog signal are modulated by at least one carrier, thereby generating at least the first carrier modulated electrical signal, the second carrier modulated electrical signal and the third carrier modulated electrical signal.

[0233] 62. The digital optical computing processing unit according to item 61, wherein at least the first carrier modulated electrical signal, the second carrier modulated electrical signal and the third carrier modulated electrical signal are modulated in an orthogonal or quadrature manner with respect to each other.

[0234] 63. The digital optical computing and processing unit according to any of the preceding items, wherein the digital optical computing and processing unit further comprises an IQ modulation system.

[0235] 64. The digital optical computation processing unit according to item 63, wherein the IQ modulation system is configured to modulate at least the first carrier modulated electrical signal, the second carrier modulated electrical signal and the third carrier modulated electrical signal in a mutually orthogonal or quadrature manner.

[0236] 65. A digital optical computation processing unit according to any of the preceding items, wherein the first electrical output signal is modulated according to Quadrature Amplitude Modulation (QAM).

[0237] 66. The digital optical computing processing unit according to item 65, wherein the quadrature amplitude modulation is quadrature phase shift keying (QPSK) modulation, preferably 8-QAM, more preferably 16-QAM, even more preferably 32-QAM, most preferably 64-QAM, further preferably 128-QAM, further more preferably 256-QAM, further even more preferably 512-QAM, further most preferably 2 N -QAM, where N is an integer.

[0238] 67. The digital optical computation processing unit according to any of the preceding items, wherein the first electrical output signal is modulated by at least one carrier wave.

[0239] 68. The digital optical computing and processing unit according to any of the preceding items, wherein the digital optical computing and processing unit further comprises an IQ demodulation system.

[0240] 69. The digital optical computation processing unit according to item 68, wherein the IQ demodulation system is configured to demodulate the first electrical output signal into at least a first electrical digital output signal, a second electrical digital output signal, and a third electrical digital output signal.

[0241] 70. A digital optical computation processing unit according to any of the preceding items, wherein at least one signal processing unit comprises an analog-to-digital converter (ADC), the analog-to-digital converter (ADC) being configured to convert the first electrical output signal into a first electrical digital output signal.

[0242] 71. A digital optical computation processing unit according to any of the preceding items, wherein at least one signal processing unit further comprises a digital signal processing filter and / or an analog or digital equalizer.

[0243] 72. The digital optical computation processing unit according to any of the preceding items, wherein at least one signal processing unit further comprises an electrical signal processing unit, wherein the electrical signal processing unit comprises at least one multiplier and at least one full adder.

[0244] 73. The digital optical computation processing unit according to any of the preceding items, wherein the digital optical computation processing unit further comprises a lookup table (LUT).

[0245] 74. A digital optical computation processing unit according to any of the preceding items, wherein the LUT is configured to store at least a first weight and a second weight.

[0246] 75. A digital optical computation processing unit according to any one of items 71-74, wherein the LUT is implemented in a field programmable gate array (FPGA).

[0247] 76. A digital optical computation processing unit according to any of the preceding items, wherein the LUT is configured to match at least the first weight and the second weight with at least the first normalized weight value and the second normalized weight value.

[0248] 77. The digital optical calculation processing unit according to item 76, wherein at least the first normalized weight value and the second normalized weight value are distributed between -1 and 1, preferably uniformly distributed between -1 and 1, and more preferably uniformly distributed between 0 and 1.

[0249] 78. The digital optical computation and processing unit according to any of the preceding items, wherein the digital optical computation and processing unit is configured to further process at least the fourth electrical digital signal and the fifth electrical digital signal and the second optical signal comprising at least two wavelengths, the digital optical computation and processing unit further comprising:

[0250] a second waveguide, configured to carry a second optical signal;

[0251] At least a third electro-optical converter and a fourth electro-optical converter included in the second waveguide are configured to be biased by at least a third weight and a fourth weight, wherein at least a fourth electrical digital signal and a fifth electrical digital signal modulate the second optical signal in the at least third electro-optical converter and the fourth electro-optical converter, respectively, and wherein the third electro-optical converter is configured to modulate a third wavelength and the fourth electro-optical converter is configured to modulate a fourth wavelength.

[0252] 79. The digital optical calculation processing unit according to item 78, wherein at least the third electro-optical converter and the fourth electro-optical converter are arranged in series.

[0253] 80. The digital optical computation processing unit according to item 78, further comprising a second light source configured to provide a second optical signal comprising at least two wavelengths to the second waveguide.

[0254] 81. The digital optical computation processing unit according to item 78, further comprising a second photoelectric converter arranged at the second output end of the second waveguide, and wherein the second photoelectric converter is configured to output a second electrical output signal based on the second optical signal.

[0255] 82. The digital optical computation and processing unit according to any of the preceding items, wherein the digital optical computation and processing unit is configured to further process at least a fourth electrical digital signal, a fifth electrical digital signal, and a sixth electrical digital signal and a second optical signal comprising at least two wavelengths, the digital optical computation and processing unit further comprising:

[0256] a second waveguide, configured to carry a second optical signal;

[0257] an auxiliary electro-optical converter included in the second waveguide, wherein the second optical signal is modulated by at least a fourth electrical digital signal in the auxiliary electro-optical converter;

[0258] at least a third electro-optical converter and a fourth electro-optical converter included in the second waveguide and further arranged in series with the auxiliary electro-optical converter, wherein the at least third electro-optical converter and the fourth electro-optical converter are configured to be biased by at least third and fourth weights, wherein the at least fifth and sixth electrical digital signals modulate the second optical signal in the at least third and fourth electro-optical converters, respectively, and wherein the third electro-optical converter is configured to modulate a third wavelength and the fourth electro-optical converter is configured to modulate a fourth wavelength;

[0259] The second waveguide portion is configured to introduce a wavelength-dependent time delay in the second optical signal.

[0260] 83. The digital optical calculation processing unit according to item 82, wherein at least the third electro-optical converter and the fourth electro-optical converter are arranged in series.

[0261] 84. The digital optical computation processing unit according to item 82, further comprising a second light source configured to provide a second optical signal comprising at least two wavelengths to the second waveguide.

[0262] 85. The digital optical computation processing unit according to item 82, further comprising a second photoelectric converter arranged at the second output end of the second waveguide, and wherein the second photoelectric converter is configured to output a second electrical output signal based on the second optical signal.

[0263] 86. The digital optical calculation and processing unit according to any of the preceding items, wherein the digital optical calculation and processing unit is configured to further process at least a fourth electrical digital signal, a fifth electrical digital signal, and a sixth electrical digital signal and a second light comprising at least two wavelengths, the digital optical calculation and processing unit further comprising:

[0264] a second waveguide, configured to carry a second optical signal;

[0265] an auxiliary electro-optical converter included in the second waveguide, wherein the second optical signal is modulated by at least a fourth electrical digital signal in the auxiliary electro-optical converter;

[0266] at least a third electro-optical converter and a fourth electro-optical converter included in the second waveguide and further arranged in series with the auxiliary electro-optical converter, wherein the at least third electro-optical converter and the fourth electro-optical converter are configured to be biased by at least third and fourth weights, wherein the at least third and fourth electrical digital signals modulate the second optical signal in the at least third and fourth electro-optical converters, respectively, and wherein the third electro-optical converter is configured to modulate a third wavelength and the fourth electro-optical converter is configured to modulate a fourth wavelength;

[0267] The second waveguide portion is configured to introduce a wavelength-dependent time delay in the second optical signal.

[0268] 87. The digital optical calculation processing unit according to item 86, wherein at least the third electro-optical converter and the fourth electro-optical converter are arranged in series.

[0269] 88. The digital optical computation processing unit according to item 86, further comprising a second light source configured to provide a second optical signal comprising at least two wavelengths to the second waveguide.

[0270] 89. The digital optical computation processing unit according to item 86, further comprising a second photoelectric converter arranged at the second output end of the second waveguide, and wherein the second photoelectric converter is configured to output a second electrical output signal based on the second optical signal.

[0271] 90. A digital optical computation processing unit according to any of the preceding items, wherein the second waveguide has at least one property or any combination of properties presented with respect to the first waveguide.

[0272] 91. A digital optical computation processing unit according to any of the preceding items, wherein the second electrical output signal has at least one characteristic or any combination of characteristics presented by the first electrical output signal.

[0273] 92. A digital optical computation processing unit according to any of the preceding items, wherein the auxiliary electro-optical converter has at least one characteristic or any combination of characteristics presented with respect to the primary electro-optical converter.

[0274] 93. The digital optical computation processing unit according to any of the preceding items, wherein the second light source is the first light source.

[0275] 94. The digital optical computation processing unit according to any of the preceding items, wherein the first wavelength has the same wavelength as the third wavelength and / or the second wavelength has the same wavelength as the fourth wavelength.

[0276] 95. A digital optical computation processing unit according to any of the preceding items, wherein the second optical signal has at least one characteristic or any combination of characteristics presented with respect to the first optical signal.

[0277] 96. A digital optical computation processing unit according to any of the preceding items, wherein at least the third electro-optical converter and the fourth electro-optical converter have at least one characteristic or any combination of characteristics exhibited by at least the first electro-optical converter and the second electro-optical converter.

[0278] 97. The digital optical computing and processing unit according to any of the preceding items, wherein the digital optical computing and processing unit further comprises a spectrometer configured to split the light source, for example, a first light signal and a second light signal are generated from the light source.

[0279] 98. The digital optical computation processing unit according to any of the preceding items, wherein the second light source is the first light source.

[0280] 99. A digital optical computing processing unit according to any of the preceding items, wherein at least the fourth electrical digital signal, the fifth electrical digital signal and the sixth electrical digital signal have at least one characteristic or any combination of characteristics presented with respect to at least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal.

[0281] 100. A digital optical computation processing unit according to any of the preceding items, wherein the second waveguide portion has at least one property or any combination of properties presented with respect to the first and second waveguide portions.

[0282] 101. A method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, comprising the following steps:

[0283] - providing an optical signal having ixj different optical carriers (such as ixj different wavelengths);

[0284] - modulating the optical signal using ixj digital data signals such that an nth optical carrier among the ixj optical carriers is modulated by an nth digital data signal among the ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix;

[0285] - Modulate the optical signal using a kernel matrix such that the kernel matrix applies weights to the ixj optical carriers;

[0286] - summing the ixj optical carriers at each symbol level in an optical-to-electrical converter to generate an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

[0287] 102. The method of item 101, wherein the photoelectric converter is wavelength independent.

[0288] 103. The method according to any one of items 101-102, wherein i is greater than j, preferably less than j, and more preferably equal to j.

[0289] 104. The method of any one of items 101-103, wherein the ixj different wavelengths are equally spaced apart from each other.

[0290] 105. The method of any one of items 101-104, wherein the ixj different wavelengths have different corresponding frequencies, and wherein the different frequencies define a frequency comb.

[0291] 106. A method according to any one of items 101-105, wherein the input data matrix contains coding elements from an image.

[0292] 107. The method of item 106, wherein the coding element is a digital representation of a pixel.

[0293] 108. The method of item 107, wherein the digital representation of the pixel is a binary code.

[0294] 109. The method according to item 108, wherein the binary code is a 2-bit binary code, preferably a 4-bit binary code, more preferably an 8-bit binary code, even more preferably a 16-bit binary code, most preferably a 32-bit binary code, and further preferably a 64-bit binary code.

[0295] 110. A method according to any one of items 101-109, wherein the kernel matrix is ​​a convolution matrix, wherein the convolution matrix is ​​a matrix used for blurring, sharpening, embossing or edge detection.

[0296] 111. The method according to any one of items 101-110, wherein the kernel matrix is ​​a 3x3 matrix, preferably a 4x4 matrix, more preferably a 6x6 matrix, and even more preferably a 16x16 matrix.

[0297] 112. The method of any one of items 101-111, wherein the optical signal is generated by a laser source.

[0298] 113. The method of item 112, wherein the laser source is configured to generate at least two wavelengths of light.

[0299] 114. A method according to any one of items 101-113, wherein the ixj digital data signals include at least two symbols.

[0300] 115. The method according to any one of items 101 to 114, wherein the ixj digital data signals are binary data signals, preferably multi-level digital data signals.

[0301] 116. The method of any one of items 101-115, wherein the weight is comprised between -1 and 1.

[0302] 117. The method of any one of items 101-116, wherein the wavelength-independent photoelectric converter is a photodiode.

[0303] 118. The method according to any one of items 101-117, wherein the method is performed by the system according to item 1-100.

[0304] 119. A method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, comprising the following steps:

[0305] - providing an optical signal having ixj different optical carriers (such as ixj different wavelengths);

[0306] - modulating the optical signal with ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix, and causing each of the ixj optical carriers to be modulated with the ixj digital data signals, thereby obtaining ixj optical data signals;

[0307] - providing a time delay between the ixj optical data signals such that each of the ixj optical data signals is delayed in time with respect to each other;

[0308] - modulating ixj optical data signals using a kernel matrix such that an nth element of the kernel matrix applies an nth weight to an nth optical carrier among the ixj optical carriers;

[0309] - summing the ixj optical carriers at each symbol level in an optical-to-electrical converter to generate an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

[0310] 120. The method of item 119, wherein the photoelectric converter is wavelength independent.

[0311] 121. A method according to any one of items 119-120, wherein i is greater than j, preferably less than j, and more preferably equal to j.

[0312] 122. The method of any one of items 119-121, wherein the ixj different wavelengths are equally spaced apart from each other.

[0313] 123. A method according to any of items 119-122, wherein the ixj different wavelengths have different corresponding frequencies, and wherein the different frequencies define a frequency comb.

[0314] 124. A method according to any of items 119-123, wherein the time delay is related to the symbol data rate.

[0315] 125. A method according to any of items 119-124, wherein j symbols and / or k symbols modulate i optical carriers at a symbol data rate.

[0316] 126. The method of item 125, wherein the symbol data rate is the inverse of the symbol data period.

[0317] 127. A method according to any one of items 119-126, wherein the time delay is 100% of the symbol data period, preferably included between 90% and 110% of the symbol data period, more preferably included between 50% and 150% of the symbol data period, and even more preferably included between 50% and 200% of the symbol data period.

[0318] 128. A method according to any one of items 119-127, wherein the input data matrix contains coding elements from an image.

[0319] 129. A method according to any of items 119-128, wherein the coding element is a digital representation of a pixel.

[0320] 130. The method of any one of items 119-129, wherein the digital representation of the pixel is a binary code.

[0321] 131. The method according to item 130, wherein the binary code is a 2-bit binary code, preferably a 4-bit binary code, more preferably an 8-bit binary code, even more preferably a 16-bit binary code, most preferably a 32-bit binary code, and further preferably a 64-bit binary code.

[0322] 132. A method according to any one of items 119-131, wherein the kernel matrix is ​​a convolution matrix, wherein the convolution matrix is ​​a matrix used for blurring, sharpening, embossing or edge detection.

[0323] 133. A method according to any one of items 119-132, wherein the kernel matrix is ​​a 3x3 matrix, preferably a 4x4 matrix, more preferably a 5x5 matrix, even more preferably a 6x6 matrix, and most preferably a 16x16 matrix.

[0324] 134. A method according to any one of items 119 to 133, wherein the optical signal is generated by a laser source.

[0325] 135. The method of item 134, wherein the laser source is configured to generate at least two wavelengths of light.

[0326] 136. A method according to any of items 119-135, wherein the ixj digital data signals include at least two symbols.

[0327] 137. The method according to any of items 119 to 136, wherein the ixj digital data signals are binary data signals, preferably multi-level digital data signals.

[0328] 138. A method according to any one of items 119-137, wherein the weight is comprised between -1 and 1.

[0329] 139. The method of any one of items 119-138, wherein the wavelength-independent photoelectric converter is a photodiode.

[0330] 140. The method according to any one of items 119-139, wherein the method is performed by a system according to item 1-100.

[0331] 141. A method for hybrid optoelectronic multiplication of a first digital signal comprising j symbols and a second digital signal comprising k symbols, comprising the following steps:

[0332] - providing an optical signal having i different optical carriers (such as i different wavelengths), and wherein i is greater than or equal to j, and wherein j is greater than or equal to k;

[0333] - modulating the optical signal with the first digital signal, such that each of the i optical carriers is modulated with the first digital signal, thereby defining i optical digital signals;

[0334] - modulating each of the i optical digital signals on the nth optical carrier among the i optical carriers using the nth symbol among the k symbols;

[0335] - providing a time delay between the i optical digital signals so that each of the i optical digital signals is separated in time;

[0336] - summing the i optical carriers at each symbol level in a wavelength-independent optical-to-electrical converter to generate an electrical output signal, wherein the electrical output signal represents the product between the first digital signal and the second digital signal.

[0337] 142. The method of item 141, wherein the photoelectric converter is wavelength independent.

[0338] 143. The method of any one of items 141-142, wherein the i different wavelengths are equally spaced apart from each other.

[0339] 144. A method according to any of items 141-143, wherein the i different wavelengths have different corresponding frequencies, and wherein the different frequencies define a frequency comb.

[0340] 145. A method according to any of items 141-144, wherein the time delay is related to the symbol data rate.

[0341] 146. A method according to any of items 141-145, wherein j symbols and / or k symbols modulate i optical carriers at a symbol data rate.

[0342] 147. The method of item 146, wherein the symbol data rate is the inverse of the symbol data period.

[0343] 148. A method according to any one of items 141-147, wherein the time delay is 100% of the symbol data period, preferably included between 90% and 110% of the symbol data period, more preferably included between 50% and 150% of the symbol data period, and even more preferably included between 50% and 200% of the symbol data period.

[0344] 149. The method according to any one of items 141 to 148, wherein the first digital signal and / or the second digital signal is a multi-level digital signal or a binary signal.

[0345] 150. The method according to any one of items 141 to 149, wherein each of the i optical digital signals is a multi-level digital signal or a binary signal.

[0346] 151. A method according to any one of items 141-150, wherein the optical signal is generated by a laser source.

[0347] 152. The method of item 151, wherein the laser source is configured to generate at least two wavelengths of light.

[0348] 153. The method of any one of items 141-152, wherein the wavelength-independent photoelectric converter is a photodiode.

[0349] 154. The method according to any one of items 141-153, wherein the method is performed by the system according to item 1-100.

Claims

1. A digital optical computation processing unit configured to perform analog-to-digital convolution by processing at least a first electrical digital signal and a second electrical digital signal and a first optical signal comprising at least two wavelengths, the digital optical computation processing unit comprising: a first waveguide, configured to carry the first optical signal; At least a first electro-optical converter and a second electro-optical converter included in the first waveguide are configured to be biased by at least a first weight and a second weight, wherein at least the first electrical digital signal and the second electrical digital signal modulate the first optical signal in at least the first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate a first wavelength and the second electro-optical converter is configured to modulate a second wavelength.

2. The digital optical calculation processing unit according to claim 1, wherein: At least the first electro-optical converter and the second electro-optical converter are arranged in series.

3. The digital optical calculation processing unit according to claim 1 , further comprising a first photoelectric converter arranged at the first output end of the first waveguide, and wherein, The first photoelectric converter is configured to output a first electrical output signal based on the first optical signal. 4 . The digital optical computation processing unit according to claim 1 , further comprising a light source configured to provide the first optical signal.

5. A digital optical calculation processing unit according to any one of the preceding claims, wherein: At least the first and second electrical digital signals represent at least a first input and a second input encoded in binary words.

6. The digital optical computation processing unit according to any one of the preceding claims, further comprising a primary electro-optical converter contained in the first waveguide, wherein The first optical signal is modulated by at least a third electrical digital signal in the main electro-optical converter, and wherein the main electro-optical converter is arranged in series and before at least the first electro-optical converter and the second electro-optical converter.

7. A digital optical calculation processing unit according to any one of the preceding claims, wherein: At least the third electrical digital signal represents at least a third input encoded in a binary word.

8. The digital optical computation processing unit according to any one of the preceding claims, further comprising a first waveguide section configured to introduce a wavelength-dependent time delay in the first optical signal.

9. The digital optical calculation processing unit according to claim 8, wherein: The first waveguide portion is included in the first waveguide and is arranged before at least the first and second electro-optical converters and after the main electro-optical converter.

10. A digital optical calculation processing unit according to any one of the preceding claims, wherein: The first electrical output signal is a first electrical digital output signal, and wherein the digital optical calculation processing unit further comprises at least one signal processing unit configured to convert the first electrical digital output signal into a binary signal.

11. A digital optical calculation processing unit according to any one of the preceding claims, wherein: The main electro-optical converter is independent of wavelength, such as a Mach-Zehnder modulator MZM or an electro-absorption modulator EAM, and wherein at least the first electro-optical converter and the second electro-optical converter are dependent on wavelength, such as a microring resonator or a cascaded Mach-Zehnder modulator MZM, and / or at least the first electro-optical converter and the second electro-optical converter are independent of wavelength.

12. A digital optical calculation processing unit according to any one of the preceding claims, wherein: The first waveguide portion is a dispersive waveguide portion, such as a photonic crystal or a chirped Bragg grating, and / or wherein the first waveguide portion is configured to generate a time delay between the at least two wavelengths.

13. A digital optical calculation processing unit according to any one of the preceding claims, wherein: At least the first electrical digital signal, the second electrical digital signal and / or the third electrical digital signal comprises at least one symbol, and wherein the at least one symbol is processed into the digital optical processing unit at a symbol data rate, and wherein the time delay is 100% of the symbol data period.

14. A digital optical computation processing unit according to any one of the preceding claims, wherein: The digital optical calculation processing unit further includes a carrier generator configured to generate at least one carrier.

15. A digital optical calculation processing unit according to any one of the preceding claims, wherein: At least the first electrical digital signal, the second electrical digital signal and the third electrical digital signal are modulated in quadrature or orthogonal manner with respect to each other, and / or at least each of the first electrical digital signal, the second electrical digital signal and the third electrical digital signal is modulated by one of the at least one carrier wave.

16. A digital optical computation processing unit according to any one of the preceding claims, wherein: The first electrical output signal is modulated according to Quadrature Amplitude Modulation (QAM), and wherein the QAM is 16-QAM.

17. A digital optical computation processing unit according to any one of the preceding claims, wherein: The digital optical calculation processing unit also includes a lookup table LUT, and wherein the LUT is configured to store at least the first weight and the second weight, and wherein the LUT is configured to match at least the first weight and the second weight with at least a first normalized weight value and a second normalized weight value, wherein at least the first normalized weight value and the second normalized weight value are uniformly distributed between 0 and 1.

18. A method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, comprising the steps of: - providing an optical signal having ixj different optical carriers, such as ixj different wavelengths; - modulating the optical signal using ixj digital data signals such that an nth optical carrier among the ixj optical carriers is modulated by an nth digital data signal among the ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix; - modulating the optical signal using the kernel matrix so that the kernel matrix applies weights to the ixj optical carriers; - summing the ixj optical carriers at each symbol level in an optical-to-electrical converter to generate an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

19. The method according to claim 18, wherein The i×j digital data signals are binary words.

20. The method according to any one of claims 18-19, wherein The optical-to-electrical converter is independent of wavelength.

21. The method according to any one of claims 18 to 20, wherein The optical signal is generated by a laser source.

22. The method according to claim 21, wherein The laser source is configured to generate at least two wavelengths of light.

23. The method according to any one of claims 18 to 22, wherein: The ixj digital data signals include at least two symbols.

24. The method according to any one of claims 18 to 23, wherein: The ixj digital data signals are binary data signals, preferably multi-level digital data signals.

25. The method according to any one of claims 18 to 24, wherein The weight is comprised between -1 and 1.

26. The method according to any one of claims 18 to 25, wherein The wavelength-independent photoelectric converter is a photodiode.

27. A method for performing hybrid optoelectronic convolution between an input data matrix of size ixj and a kernel matrix of size ixj, comprising the steps of: - providing an optical signal having ixj different optical carriers, such as ixj different wavelengths; - modulating the optical signal with ixj digital data signals, wherein the ixj digital data signals are digital representations of the ixj elements of the input data matrix, and causing each of the ixj optical carriers to be modulated with the ixj digital data signals, thereby obtaining ixj optical data signals; - providing a time delay between the ixj optical data signals such that each of the ixj optical data signals is delayed in time with respect to each other; - modulating the ixj optical data signals using the kernel matrix, so that an nth element of the kernel matrix applies an nth weight to an nth optical carrier among the ixj optical carriers; - summing the ixj optical carriers at each symbol level in an optical-to-electrical converter to generate an electrical output signal, wherein the electrical output signal represents the convolution between the input data matrix and the kernel matrix.

28. The method according to claim 27, wherein The ixj digital data signals are binary words.

29. The method according to any one of claims 27-28, wherein The optical-to-electrical converter is independent of wavelength.

30. The method according to any one of claims 27 to 29, wherein The time delay is related to the symbol data rate.

31. The method according to any one of claims 27 to 30, wherein: The j symbols and / or the k symbols modulate the i optical carriers at the symbol data rate.

32. The method according to claim 31, wherein The symbol data rate is the inverse of the symbol data period.

33. The method according to any one of claims 27 to 32, wherein: The time delay is 100% of the symbol data period, preferably included between 90% and 110% of the symbol data period, more preferably included between 50% and 150% of the symbol data period, and even more preferably included between 50% and 200% of the symbol data period.

34. The method according to any one of claims 27 to 33, wherein: The optical signal is generated by a laser source.

35. The method according to claim 34, wherein The laser source is configured to generate at least two wavelengths of light.

36. The method according to any one of claims 27 to 35, wherein The ixj digital data signals include at least two symbols.

37. The method according to any one of claims 27 to 36, wherein The ixj digital data signals are binary data signals, preferably multi-level digital data signals.

38. The method according to any one of claims 27 to 37, wherein The weight is comprised between -1 and 1.

39. The method according to any one of claims 27 to 38, wherein The wavelength-independent photoelectric converter is a photodiode.

40. A digital optical computation processing unit configured to perform digital-to-digital multiplication by processing at least a multiplier and a multiplicand and a first optical signal comprising at least two wavelengths, wherein: The multiplicand includes at least a first electrical digital signal, the multiplier includes at least a second electrical digital signal and a third electrical digital signal, and the digital optical calculation processing unit includes: a first waveguide, configured to carry the first optical signal; a main electro-optical converter included in the first waveguide, wherein the first optical signal is modulated in the main electro-optical converter by at least the first electrical digital signal; at least a first electro-optical converter and a second electro-optical converter included in the first waveguide and further arranged in series with the main electro-optical converter, wherein at least the second electrical digital signal and the third electrical digital signal modulate the first optical signal in at least the first electro-optical converter and the second electro-optical converter, respectively, and wherein the first electro-optical converter is configured to modulate a first wavelength and the second electro-optical converter is configured to modulate a second wavelength; The waveguide portion is configured to introduce a wavelength-dependent time delay into the first optical signal.

41. The digital optical calculation processing unit according to claim 40, wherein: At least the first electro-optical converter and the second electro-optical converter are arranged in series.

42. The digital optical calculation processing unit according to claim 40, further comprising a first photoelectric converter arranged at the first output end of the first waveguide, and wherein, The first photoelectric converter is configured to output a first electrical output signal based on the first optical signal.

43. The digital optical computation processing unit according to any one of claims 40 to 42, further comprising a light source configured to provide a first optical signal comprising at least two wavelengths to the first waveguide.

44. The digital optical calculation processing unit according to any one of claims 40 to 43, wherein: At least the first electrical digital signal is encoded in binary words.

45. The digital optical calculation processing unit according to any one of claims 40 to 44, wherein: At least the second electrical digital signal and the third electrical digital signal are respectively at least a first bit and a second bit of the multiplier.

46. ​​The digital optical calculation processing unit according to any one of claims 40 to 45, wherein: The wavelength-dependent time delay is the time delay between the at least two wavelengths.

47. The digital optical calculation processing unit according to any one of claims 40 to 46, wherein: The waveguide portion is comprised in the first waveguide and is arranged after the main electro-optical converter.

48. The digital optical calculation processing unit according to any one of claims 40 to 47, wherein: The waveguide portion is included in the first waveguide and is arranged after at least the first electro-optical converter and the second electro-optical converter.

49. The digital optical calculation processing unit according to any one of claims 40 to 48, wherein: At least the first and second electrical-to-optical converters are configured to be biased by at least the second and third electrical digital signals, thereby modulating the first optical signal with at least the second and third electrical digital signals.

50. A method for hybrid optoelectronic multiplication of a first digital signal comprising j symbols and a second digital signal comprising k symbols, comprising the following steps: - providing an optical signal having i different optical carriers, such as i different wavelengths, and wherein i is greater than or equal to j, and wherein j is greater than or equal to k; - modulating the optical signal using the first digital signal so that each of the i optical carriers is modulated using the first digital signal, thereby defining i optical digital signals; - modulating each of the i optical digital signals on the n-th optical carrier among the i optical carriers using the n-th symbol among the k symbols; - providing a time delay between the i optical digital signals so that each of the i optical digital signals is separated in time; - summing the i optical carriers at each symbol level in a wavelength-independent optical-to-electrical converter to generate an electrical output signal, wherein the electrical output signal represents the product between the first digital signal and the second digital signal.

51. The method of claim 50, wherein: The optical-to-electrical converter is independent of wavelength.

52. The method according to any one of claims 50-51, wherein The time delay is related to the symbol data rate.

53. The method according to any one of claims 50 to 52, wherein: The j symbols and / or the k symbols modulate the i optical carriers at the symbol data rate.

54. The method of claim 53, wherein: The symbol data rate is the inverse of the symbol data period.

55. The method according to any one of claims 50 to 54, wherein The time delay is 100% of the symbol data period, preferably included between 90% and 110% of the symbol data period, more preferably included between 50% and 150% of the symbol data period, and even more preferably included between 50% and 200% of the symbol data period.

56. The method according to any one of claims 50 to 55, wherein The first digital signal and / or the second digital signal is a multi-level digital signal or a binary signal.

57. The method according to any one of claims 50 to 56, wherein Each of the i optical digital signals is a multi-level digital signal or a binary signal.

58. The method according to any one of claims 50 to 57, wherein The optical signal is generated by a laser source.

59. The method of claim 58, wherein The laser source is configured to generate at least two wavelengths of light.

60. The method according to any one of claims 50 to 59, wherein The wavelength-independent photoelectric converter is a photodiode.

Images