LINEAR PHOTONIC PROCESSORS AND RELATED METHODS

Abstract

Photonic processors are described. The photonic processors described herein are configured to perform matrix-matrix (e.g., matrix-vector) multiplication. Some embodiments relate to photonic processors arranged according to a dual-rail architecture, in which numeric values are encoded in the difference between a pair optical signals (e.g., in the difference between the powers of the optical signals). Relative to other architectures, these photonic processors exhibit increased immunity to noise. Some embodiments relate to photonic processors including modulatable detector-based multipliers. Modulatable detectors are detectors designed so that the photocurrent can be modulated according to an electrical control signal. Photonic processors designed using modulatable detector-based multipliers are significantly more compact than other types of photonic processors.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application is a Continuation claiming the benefit under 35 U.S.C. § 120 of U.S. application Ser. No. 17/398,384, filed Aug. 10, 2021, under Attorney Docket No. L0858.70023US03, entitled “LINEAR PHOTONIC PROCESSORS AND RELATED METHODS,” which is hereby incorporated herein by reference in its entirety.

U.S. application Ser. No. 17/398,384 is a Division claiming the benefit under 35 U.S.C. § 120 of U.S. application Ser. No. 17/101,415, filed Nov. 23, 2020, under Attorney Docket No. L0858.70023US02, entitled “LINEAR PHOTONIC PROCESSORS AND RELATED METHODS,” which is hereby incorporated herein by reference in its entirety.

U.S. application Ser. No. 17/101,415 claims the benefit under 35 U.S.C. § 119(e) of U.S. Provisional Patent Application Ser. No. 62/939,480, entitled “SYSTEMS AND METHODS FOR ANALOG COMPUTING,” filed on Nov. 22, 2019 under Attorney Docket No. L0858.70016US01, U.S. Provisional Patent Application Ser. No. 62/962,759, entitled “MODULATABLE DETECTOR-BASED MULTIPLIERS,” filed on Jan. 17, 2020 under Attorney Docket No. L0858.70023US00, U.S. Provisional Patent Application Ser. No. 62/963,315, entitled “DUAL-RAIL PHOTONIC MULTIPLIER SYSTEM WITH APPLICATIONS TO LINEAR PHOTONIC PROCESSOR,” filed on Jan. 20, 2020 under Attorney Docket No. L0858.70022US00, U.S. Provisional Patent Application Ser. No. 62/970,360, entitled “MODULATABLE DETECTORS,” filed on Feb. 5, 2020 under Attorney Docket No. L0858.70026US00, and U.S. Provisional Patent Application Ser. No. 62/978,181, entitled “MODULATABLE DETECTORS,” filed on Feb. 18, 2020 under Attorney Docket No. L0858.70026US01, each of which is hereby incorporated herein by reference in its entirety.

BACKGROUND

Deep learning, machine learning, latent-variable models, neural networks and other matrix-based differentiable programs are used to solve a variety of problems, including natural language processing and object recognition in images. Solving these problems with deep neural networks typically requires long processing times to perform the required computation. The most computationally intensive operations in solving these problems are often mathematical matrix operations, such as matrix multiplication.

SUMMARY OF THE DISCLOSURE

Some embodiments relate to a method for performing a mathematical operation comprising: receiving an input optical signal; obtaining a first numeric value and a second numeric value; generating first and second encoded optical signals by modifying the input optical signal using the first numeric value; generating first and second encoded output signals using the second numeric value and the first and second encoded optical signals; and obtaining a result of the mathematical operation using the first and second encoded output signals.

In some embodiments, the first and second encoded output signals are optical signals.

In some embodiments, a difference between the first and second encoded output signals is proportional to a difference between the first and second encoded optical signals.

In some embodiments, the difference between the first and second encoded output signals is proportional to the second numeric value.

In some embodiments, the difference between the first and second encoded optical signals is proportional to the first numeric value.

In some embodiments, the first and second encoded optical signals are orthogonal to each other.

In some embodiments, respective phases of the first and second encoded optical signals are uncorrelated.

In some embodiments, the first and second encoded optical signals have different carrier frequencies.

In some embodiments, the first and second encoded optical signals have a constant quadrature phase difference.

In some embodiments, wherein receiving the input optical signal comprises receiving the input optical signal from an incoherent light source.

In some embodiments, the first and second encoded optical signals have orthogonal polarizations.

In some embodiments, the first and second encoded optical signals are temporally non-overlapping.

In some embodiments, generating the first and second encoded optical signals comprises passing the input optical signal through an optical modulator.

In some embodiments, generating the first and second encoded optical signals comprises setting a characteristic of the optical modulator based on the first numeric value.

In some embodiments, obtaining the result comprises subtracting the first encoded output signal from the second encoded output signal or subtracting the second encoded output signal from the first encoded output signal.

In some embodiments, generating the first and second encoded output signals comprises detecting the first and second encoded optical signals using one or more modulatable detectors.

In some embodiments, generating the first and second encoded output signals comprises setting a characteristic of the one or more modulatable detectors based on the second numeric value.

In some embodiments, the first and second encoded optical signals are confined within different optical waveguides.

In some embodiments, obtaining the result comprises obtaining a product of the first numeric value times the second numeric value.

Some embodiments relate to a photonic processor comprising: a plurality of differential optical encoders including first and second differential optical encoders; a first set of differential multipliers coupled to the first differential optical encoder; a second set of differential multipliers coupled to the second differential optical encoder; a first receiver coupled to the first set of differential multipliers; and a second receiver coupled to the second set of differential multipliers.

In some embodiments, at least one of the plurality of differential encoders comprises an optical modulator having a pair of optical output ports.

In some embodiments, at least one differential multiplier of the first and second sets of differential multipliers comprises an optical modulator having a pair of optical output ports.

In some embodiments, the first and second sets of differential multipliers comprise modulatable detectors.

In some embodiments, the photonic processor further comprises an optical orthogonalization unit placed between the first differential optical encoder and the first set of differential multipliers.

In some embodiments, the optical orthogonalization unit comprises a serpentine-shaped optical waveguide.

In some embodiments, the optical orthogonalization unit comprises an optical polarization rotator.

In some embodiments, the photonic processor further comprises a light source and an optical splitter tree coupling the light source to the first set of differential multipliers.

In some embodiments, the optical splitter tree lacks waveguide crossings.

In some embodiments, the photonic processor further comprises a controller configured to control the plurality of differential optical encoders, wherein the controller comprises a plurality of transistors, and wherein the plurality of transistors and the plurality of differential optical encoders share at least one layer of a semiconductor substrate.

Some embodiments relate to a method for fabricating a photonic processor comprising: obtaining a semiconductor substrate; forming, on the semiconductor substrate: a plurality of differential optical encoders including first and second differential optical encoders; a first set of differential multipliers coupled to the first differential optical encoder; a second set of differential multipliers coupled to the second differential optical encoder; a first receiver coupled to the first set of differential multipliers; and a second receiver coupled to the second set of differential multipliers.

In some embodiments, forming the plurality of differential encoders comprises forming a plurality of optical modulators each having a pair of optical output ports.

In some embodiments, the method further comprises forming a plurality of transistors on the semiconductor substrate.

In some embodiments, the plurality of transistors and the plurality of differential optical encoders share at least one layer of the semiconductor substrate.

Some embodiments relate to a method for performing a mathematical operation comprising: receiving an input optical signal; obtaining a first numeric value and a second numeric value; generating an encoded optical signal by modifying the input optical signal using the first numeric value; generating a photocurrent at least in part by: detecting the encoded optical signal using a modulatable detector, and setting a characteristic of the modulatable detector based on the second value; and obtaining a result of the mathematical operation using the photocurrent.

In some embodiments, the modulatable detector comprises a photodetector, and wherein setting the characteristic of the modulatable detector based on the second value comprises setting a responsivity of the photodetector based on the second value.

In some embodiments, obtaining the result comprises obtaining a product of the first numeric value times the second numeric value.

In some embodiments, the modulatable detector comprises a control capacitor, and wherein setting the characteristic of the modulatable detector comprises setting a voltage applied to the control capacitor.

In some embodiments, the control capacitor comprises a metal-oxide-semiconductor capacitor (MOS cap), and wherein setting the voltage applied to the control capacitor comprises setting the voltage applied to the MOS cap.

In some embodiments, setting the characteristic of the modulatable detector comprises producing carrier avalanche.

In some embodiments, generating the encoded optical signal comprises passing the input optical signal through an optical modulator.

In some embodiments, the modulatable detector comprises a photodetector and a transistor, and wherein setting the characteristic of the modulatable detector comprises setting a voltage applied to the transistor.

In some embodiments, the modulatable detector comprises a photodetector and a gain stage, and wherein setting the characteristic of the modulatable detector based on the second value comprises setting a current gain of the gain stage based on the second value.

Some embodiments relate to a photonic device configured to perform a mathematical operation comprising: an optical encoder; a modulatable detector coupled to an output of the optical encoder; and a controller coupled to both the optical encoder and the modulatable detector, the controller being configured to: obtain a first numeric value and a second numeric value, control the optical encoder to generate an encoded optical signal by modifying an input optical signal using the first numeric value, control the modulatable detector to generate a photocurrent in response to receiving the encoded optical signal, wherein controlling the modulatable detector comprises setting a characteristic of the modulatable detector based on the second numeric value, and obtain a result of the mathematical operation using the photocurrent.

In some embodiments, the modulatable detector comprises a photodetector, and wherein setting the characteristic of the modulatable detector based on the second value comprises setting a responsivity of the photodetector based on the second value.

In some embodiments, the modulatable detector comprises a photo-absorption region and a control capacitor positioned adjacent to the photo-absorption region.

In some embodiments, the control capacitor comprises a metal-oxide-semiconductor capacitor (MOS cap).

In some embodiments, the modulatable detector further comprises an electron avalanche region positioned adjacent to the MOS cap.

In some embodiments, the modulatable detector comprises: a first photodetector and a second photodetector; first and second transistors both coupled to the first photodetector; and third and fourth transistors both coupled to the second photodetector.

In some embodiments, the first photodetector is coupled to respective sources of the first and second transistors, and wherein the first transistor and the third transistor have drains that are coupled to each other.

In some embodiments, the first and third transistors are arranged as an inverter, and wherein the first photodetector is coupled to respective sources of the first and second transistors.

In some embodiments, the first photodetector is further coupled to the third and fourth transistors and the second photodetector is further coupled to the first and second transistors, and wherein the first transistor and the second transistor have drains that are coupled to each other and sources that are coupled to each other.

In some embodiments, the modulatable detector comprises: a first photodetector and a second photodetector; first and second transistors both coupled to the first photodetector; and a node coupled to the first and second photodetectors and further coupled to the first and second transistors.

In some embodiments, the modulatable detector comprises a photodetector and a plurality of transistors, and wherein the photodetector and the plurality of transistors are formed on a common semiconductor substrate.

In some embodiments, the modulatable detector comprises a plurality of balanced photodetectors and a plurality of transistors arranged differentially.

Some embodiments relate to a photonic processor comprising: a plurality of differential optical encoders including first and second differential optical encoders; a first pair of modulatable detectors coupled to the first differential optical encoder; a second pair of modulatable detectors coupled to the second differential optical encoder; a first differential receiver coupled to the first pair of modulatable detectors; and a second differential receiver coupled to the second pair of modulatable detectors.

In some embodiments, at least one of the first pair of modulatable detectors comprises a photo-absorption region and a control capacitor positioned adjacent to the photo-absorption region.

In some embodiments, the control capacitor comprises a metal-oxide-semiconductor capacitor (MOS cap).

In some embodiments, at least one of the first pair of modulatable detectors comprises: a first photodetector and a second photodetector; first and second transistors both coupled to the first photodetector; and third and fourth transistors both coupled to the second photodetector.

In some embodiments, the first photodetector is coupled to respective sources of the first and second transistors, and wherein the first transistor and the third transistor have drains that are coupled to each other.

In some embodiments, at least one of the first pair of modulatable detectors comprises: a first photodetector and a second photodetector; first and second transistors both coupled to the first photodetector; and a node coupled to the first and second photodetectors and further coupled to the first and second transistors.

In some embodiments, at least one of the first pair of modulatable detector comprises a photodetector and a plurality of transistors, wherein the photodetector and the plurality of transistors are formed on a common semiconductor substrate.

Some embodiments relate to a method for fabricating a photonic processor comprising: obtaining a semiconductor substrate; forming, on the semiconductor substrate: a plurality of differential optical encoders including first and second differential optical encoders; a first pair of modulatable detectors coupled to the first differential optical encoder; a second pair of modulatable detectors coupled to the second differential optical encoder; a first differential receiver coupled to the first pair of modulatable detectors; and a second differential receiver coupled to the second pair of modulatable detectors.

In some embodiments, forming the plurality of differential encoders comprises forming a plurality of optical modulators each having a pair of optical output ports.

In some embodiments, forming the first pair of modulatable detectors comprising forming a plurality of transistors.

In some embodiments, the plurality of transistors and the plurality of differential optical encoders share at least one layer of the semiconductor substrate.

BRIEF DESCRIPTION OF THE DRAWINGS

Various aspects and embodiments of the application will be described with reference to the following figures. It should be appreciated that the figures are not necessarily drawn to scale. Items appearing in multiple figures are indicated by the same reference number in the figures in which they appear.

FIG. 1A is a block diagram illustrating a dual-rail optical multiplier, in accordance with some embodiments.

FIG. 1B is a flow chart illustrating a method for performing multiplications using a dual-rail optical multiplier, in accordance with some embodiments.

FIG. 2A is a diagram illustrating a tunable directional coupler, in accordance with some embodiments.

FIG. 2B is a diagram illustrating a portion of the tunable directional coupler of FIG. 2A in additional detail, in accordance with some embodiments.

FIG. 3 is a diagram illustrating a Mach-Zehnder interferometer, in accordance with some embodiments.

FIG. 4A-4D are block diagrams illustrating additional dual-rail optical multiplier, in accordance with some embodiments.

FIG. 5 is a representation of a matrix-vector multiplication, in accordance with some embodiments.

FIG. 6 is a block diagram of a photonic processor arranged according to a 2×2 configuration, in accordance with some embodiments.

FIG. 7A is a block diagram of a photonic processor arranged according to a 4×4 configuration, in accordance with some embodiments.

FIG. 7B is a diagram illustrating an optical splitter tree, in accordance with some embodiments.

FIG. 8 is a diagram illustrating another optical splitter tree, in accordance with some embodiments.

FIG. 9A is a block diagram illustrating a modulatable detector, in accordance with some embodiments.

FIG. 9B is a block diagram illustrating a pair of modulatable detectors arranged differentially, in accordance with some embodiments.

FIG. 9C-9D are block diagrams illustrating the interior of modulatable detectors, in accordance with some embodiments.

FIG. 10A is a block diagram illustrating an optical multiplier including the modulatable detector of FIG. 9A, in accordance with some embodiments.

FIG. 10B is a flow chart illustrating a method for performing multiplications, in accordance with some embodiments.

FIG. 11 is a block diagram illustrating a differential optical multiplier including the modulatable detectors of FIG. 9B, in accordance with some embodiments.

FIG. 12A is a block diagram of another photonic processor arranged according to a 2×2 configuration, in accordance with some embodiments.

FIG. 12B is a block diagram of another photonic processor arranged according to a 4×4 configuration, in accordance with some embodiments.

FIG. 12C is a diagram illustrating an optical splitter tree, in accordance with some embodiments.

FIG. 13 is a diagram illustrating a photoconductor, in accordance with some embodiments.

FIG. 14A is a top view of a photodetector including a metal-oxide-semiconductor capacitor (MOS cap), in accordance with some embodiments.

FIG. 14B is a cross sectional view of the modulatable detector of FIG. 14A taken along the BB line, in accordance with some embodiments.

FIG. 14C is an energy band diagram plotted along the CC line of FIG. 14B, in accordance with some embodiments.

FIGS. 15A-15E are plots illustrating example responsivities for the modulatable detector of FIG. 14A, in accordance with some embodiments.

FIGS. 16A-16B are plots illustrating photocarrier concentration over time, in accordance with some embodiments.

FIG. 17 is a plot illustrating example current/voltage characteristics of a modulatable detector, in accordance with some embodiments.

FIGS. 18A-18H are circuit diagrams illustrating additional modulatable detectors, in accordance with some embodiments.

FIGS. 19A-19B are plots illustrating example responsivities associated with the modulatable detector of FIG. 18E, in accordance with some embodiments.

FIGS. 20A-20B are cross sectional views illustrating substrates including modulatable detectors, in accordance with some embodiments.

DETAILED DESCRIPTION

I. Overview

Conventional electronic processors face severe speed and efficiency limitations primarily to the inherent presence of impedance in electronic interconnects. Connecting multiple processor cores and/or connecting a processor core to a memory involves use of conductive traces. Large values of impedance limit the maximum rate at which data can be transferred through the trace with a negligible bit error rate. For processing that requires billions of operations, these delays can result in a significant loss of performance. In addition to electrical circuits' inefficiencies in speed, the heat generated by the dissipation of energy caused by the impedance of the circuits is also a barrier in developing electronic processors.

The inventors have recognized and appreciated that using optical signals (instead of, or in combination with, electrical signals) overcomes the aforementioned problems with electronic computing. Optical signals travel at the speed of light in the medium in which the light is traveling. Thus, the latency of optical signals is far less of a limitation than electrical propagation delay. Additionally, no power is dissipated by increasing the distance traveled by the light signals, opening up new topologies and processor layouts that would not be feasible using electrical signals. Thus, photonic processors offer far better speed and efficiency performance than conventional electronic processors.

The inventors have recognized and appreciated that photonic processors are well-suited for particular types of algorithms. For example, many machine learning algorithms (e.g. support vector machines, artificial neural networks and probabilistic graphical model learning) rely heavily on linear transformations on multi-dimensional arrays/tensors. The simplest linear transformation is a matrix-vector multiplication, which using conventional algorithms has a complexity on the order of O(N²), where N is the dimensionality of a square matrix being multiplied by a vector of the same dimension. The inventors have recognized and appreciated that a photonic processor can perform linear transformations, such as matrix multiplication, in a highly parallel manner by propagating a particular set of input optical signals through a configurable array of active optical components. Using such implementations, matrix-vector multiplication of dimension N=512 can be completed in hundreds of picoseconds, as opposed to the tens to hundreds of nanoseconds using conventional electronic circuit-based processing.

General matrix-matrix (GEMM) operations are ubiquitous in software algorithms, including those for graphics processing, artificial intelligence, neural networks and deep learning. GEMM calculations in today's computers are typically performed using transistor-based systems such as GPU systems or systolic array systems. Matrix-vector multiplication using photonics arrays can be highly power efficient when compared to their electronic counterparts as optical signals can propagate within a semiconductor substrate with a minimal amount of loss.

However, the inventors have recognized and appreciated a number of challenges associated with the use of such photonic arrays. First, just like electronic circuits, photonic arrays are susceptible to noise. Noise arises in photonic arrays due to a variety of mechanisms, including thermal noise and shot noise. The presence of noise reduces the ability of photonic arrays to accurately reproduce and process numeric values, thereby reducing their overall performance.

Second, photonic arrays are susceptible to optical loss. On the one hand, optical sources can produce only limited amounts of optical power. On the other hand, photodetectors are limited by shot noise, meaning that the minimum optical power that a photodetector can detect is limited to a floor level. The result is that the optical power budget is limited, and therefore, every decibel of optical power counts. Optical loss negatively affects the power budgets, and can arise due to a variety of reasons. For example, optical loss arises as a result of optical modulation. In conventional optical systems, optical power is intentionally attenuated—thereby increasing optical loss—to produce modulation. For example, in amplitude-based modulation schemes, a logical 0 is reproduced by suppressing the power of an optical signal.

Third, conventional photonic arrays occupy substantially more chip real estate than electronic components such as transistors, thereby limiting the amount of processing capabilities that may be integrated on a single chip. Consider for example Mach-Zehnder interferometers, which are conventionally used to perform optical modulation. To provide sufficient optical modulation, Mach-Zehnder interferometers are generally designed with large lengths, often in the order of several millimeters. One the other hand, a transistor, the core component of an electronic circuit, is several orders of magnitude smaller.

The inventors have developed a novel photonic processing architecture for performing matrix-matrix multiplication (including matrix-vector multiplication, a core component of GEMM operations), that avoids or mitigates the above-described challenges. According to an aspect of the present disclosure, the architecture described herein involves processing optical signals in a differential fashion. Instead of encoding data into a single-ended optical signal as is done for example in digital optical communication systems, data is encoded in the difference between a pair of optical signals (e.g., in the difference between the amplitudes of the optical signals or in the difference between the powers of the optical signals). This architecture is referred to herein as “dual-rail.” One rail carries an optical signal, another rail carries another optical signal. Rails can be implemented in any of numerous ways, including for example using a pair of distinct optical waveguides, where each waveguide carries one optical signal. It should be appreciated, however, that the rails need not be physically separate channels. In some embodiments, for example, both rails are implemented on the same physical waveguide, and the optical signals are distinguishable from one another by virtue of their polarization, wavelength, or other optical characteristics.

The dual-rail architectures described herein reduce photonic processors'susceptibility to noise. Statistically, when noise is present on the first rail, there is a high likelihood that noise having substantially the same characteristics is also present on the second rail. This is especially true if the rails are defined on optical waveguides positioned in close proximity to one another. Decoding data encoded on the difference between the optical signals of the rails involves performing a subtraction, meaning that the signal (including noise) present at the first rail is subtracted from the signal (including noise) present at the second rail. If the noise present at the first rail has characteristics substantially similar to the noise present at the second rail (e.g., there is a relatively high correlation), the overall noise is reduced when the subtraction is performed.

In addition to improving the susceptibility to noise, the dual-rail architectures described herein reduce modulation-induced optical loss. Unlike in conventional optical digital communication systems, in which optical modulation is achieved by introducing optical attenuation, optical modulation according to the present architectures involves a rotation of the optical vector (phase and/or polarization). In other words, modulation involves manipulating the relative phase of the optical signals present at the two rails. This results in a substantial reduction of the optical loss.

According to another aspect of the present disclosure, the photonic architectures described herein involve multipliers implemented based on “modulatable detectors.” Modulatable detectors are optical detectors having at least one characteristic that can be controlled by a user using one or more electric control signals. These detectors are designed so that application of a control signal (e.g., a voltage or a current) alters a characteristic of the detector, such as the detector's responsivity, gain, impedance, etc. The result is that the detector's photocurrent (the current that the modulatable detector produces in response to light) depends not only on the optical power incident on the detector, but also on the control signal applied to the detector. This may be achieved in any of numerous ways, as described in detail further below. In one example, a modulatable detector is designed to include a control capacitor. The control capacitor is designed so that a change in the voltage applied to it results in a change in the responsivity of the modulatable detector. The control capacitor may be implemented, among other possible configurations, using a pair of electrodes separated by a dielectric material or using a metal-oxide-semiconductor capacitor (MOS cap). In another example, a modulatable detector includes a photodetector and a gain stage coupled to the photodetector. The gain stage may be connected to the photodetector so that a change in the voltage applied to the gain stage produces a change in the photocurrent generated by the modulatable detector. Several examples of controllable gain stages are described in detail further below.

Photonic arrays that use modulatable detector-based multipliers are substantially more compact than other types of photonic arrays. This is because modulatable detectors are far more compact than optical devices traditionally used to perform multiplication in the optical domain, such as phase shifters (e.g., Mach-Zehnder interferometers), attenuators and amplifiers.

II. Dual-Rail Optical Multipliers

FIG. 1A is a block diagram illustrating an optical multiplier implemented based on a dual-rail architecture, in accordance with some embodiments. This optical multiplier includes a light source 10, a differential optical encoder 12, a differential optical multiplier 14, a differential receiver 16 and a controller 17. Controller 17 includes a pair of digital-to-analog converters (D/A) 19 and a numeric value unit 18.

Numeric value unit 18 produces a pair of scalar numeric values: x and m. Numeric value m is also referred to as “weight” or “weight parameter” and numeric value x is also referred to herein as “input data,” “input value” or “input parameter”. These numeric values may be produced based on data received by the controller, including data obtained from a memory internal to controller 17 and/or data provided to controller 17 from another computing system. These numeric values may be represented using any digital representation, including fixed-point or floating-point representations. The first D/A 19 converts numeric value x to an electric signal representative of x. In this example, the D/A produces a voltage V_x. The second D/A 19 converts numeric value m to an electric signal representative of m. In this example, the D/A produces a voltage V_m. In some embodiments, V_x is proportional to x. In some embodiments, V_m is proportional to m. The dual-rail optical multiplier of FIG. 1A is configured to multiply these numeric values to one another, thereby producing the result x×m.

Light source 10 may be implemented using a coherent source, e.g., a laser. Alternatively, light source 10 may be implemented using an incoherent source, e.g., a light-emitting diode, a source of amplified spontaneous emission, or a source of stimulated emission having a relatively large linewidth. As used herein, the terms “coherence” and “coherent” refers to temporal coherence. The optical power produced by light source 10 is identified as “P_in.”

Differential optical encoder 12 receives a voltage V_x, and in response, produces a pair of optical signals. The labels “P_t” and “P_b” identify the optical powers of these optical signals, respectively. Differential optical encoder 12 receives voltage V_x, and encodes the optical signal received from light source 10 based on V. More specifically, differential optical encoder 12 produces a pair of optical signals in such a way that the difference between the powers of these optical signals (P_t−P_b) is proportional to both x and Pin (in FIG. 1A, the symbol “∝” means “is proportional to”). It should be appreciated that this architecture is referred to as “dual-rail” in that x is encoded in the difference between two optical signals (the difference in the signal powers as in this example, or in other examples, the difference between the signal amplitudes). Differential optical encoder 12 may be implemented using any suitable optical modulator, including an optical interferometer (such as a tunable directional coupler or a Mach-Zehnder interferometer), a resonant modulator, a Franz-Keldysh modulator, etc. Examples of differential optical encoder 12 are described in detail further below.

As discussed above, P_t identifies the power of the optical signal at the top rail and P_b identifies the power of the optical signal at the bottom rail. In some embodiments, the rails are defined in terms of physical channels. In one example, the top rail is defined in a first optical waveguide and the bottom rail is defined in a second optical waveguide that is physical distinct from the first optical waveguide. In another example, the top rail is defined in a first free-space optical channel and the bottom rail is defined in a second free-space optical channel that is spatially separated from the first free-space optical channel. In other embodiments, however, the top and bottom rails may be defined by a common physical channel. That is, the optical signals generated by differential optical encoder 12 share the same optical waveguide or free-space channel. In these embodiments, the optical signal of each rail is distinguishable from the optical signal of the other rail by a certain optical characteristic, such as by the time bin, polarization or wavelength. In one example, the optical signal at the first rail is defined by a first polarization mode of an optical waveguide (e.g., the optical waveguide's TE₀₀-mode) and the optical signal at the second rail is defined by a second polarization mode of the same optical waveguide (e.g., the optical waveguide's TE₀₁-mode or TM₀₀-mode). In another example, the optical signal at the first rail is defined by a first wavelength and the optical signal at the second rail is defined by a second wavelength.

It should be appreciated that, while FIG. 1A depicts an architecture in which the signals representing the numeric values to be multiplied are voltages (V_x and V_m), in other embodiments, the numeric values may be represented using other types of electrical signals, such as electric currents or electric charges.

Differential optical multiplier 14 receives the optical signals produced by differential optical encoder 12 and produces a pair of output optical signals based on the voltage V_m. The powers of the output optical signals are labeled “P_t′” and “P_b′,” respectively. The optical signals produced by differential optical multiplier 14 are such that the difference between their powers (P_t′−P_b′) is proportional to both m and the difference P_t−P_b. The result is that the quantity P_t′-P_b′ is proportional to each of m, x and P_in. Being proportional to both x and m, P_t′−P_b′, in essence, is encoded with the product of numeric value x times numeric value m.

Differential optical multiplier 14 may be implemented using any suitable photonic device, including any suitable optical interferometer such as an adjustable directional coupler or a Mach-Zehnder interferometer, as described in detail further below.

Differential receiver 16 detects optical signals P_t′ and P_b′ and, in response, produces a numeric value y that is equal to the product x×m. To perform this operation, receiver 16 may include for example a pair of balanced photodetectors, a differential trans-impedance amplifier configured to generate an output voltage proportional to P_t′−P_b′, and an analog-to-digital converter configured to convert the output voltage to a numeric value y.

FIG. 1B is a flow chart illustrating a method for performing mathematical operations (e.g., multiplications), in accordance with some embodiments. Method 20 can be performed using any suitable optical device, including the dual-rail architecture depicted in FIG. 1A. Method 20 begins at step 22, in which an optical device receives an input optical signal. Referring for example to the architecture of FIG. 1A, at step 22, differential optical encoder 12 receives the optical signal P_in.

At step 24, the photonic device obtains a first numeric value and a second numeric value. These are the numeric values to be multiplied. The numeric values need not be obtained at the same time. Referring for example to the architecture of FIG. 1A, at step 24, numeric value unit 18 generates numeric values x and m based on data stored in a memory of controller 17 and/or on data obtained from another computing system. These numeric values may represent any type of information, such as text, audio, video, etc. The numeric values may be real or complex, positive or negative.

At step 26, the photonic device generates a pair of encoded optical signals by modifying the input optical signal using the first numeric value. In some embodiments, the pair of encoded optical signals are encoded so that the difference between the optical signals (e.g., the difference between the powers of the optical signals or the difference between the amplitudes of the optical signals) is proportional to the first numeric value. Referring for example to the architecture of FIG. 1A, at step 26, differential optical encoder 12 generates optical signals with powers P_t and P_b, respectively, based on numeric value x.

At step 28, the photonic device generates a pair of encoded output signals using the second numeric value and the pair of encoded optical signals generated at step 26. In some embodiments, the difference (in terms of power or amplitude) between the first and second encoded output signals is proportional to the difference (again, in terms of power or amplitude) between the first and second encoded input signals. In some such embodiments, the difference between the first and second encoded output signals is proportional to both the second numeric value and the first numeric value. Referring for example to the architecture of FIG. 1A, at step 28, differential optical multiplier 14 generates optical signals with powers P_t′ and P_b′, respectively, based on numeric value m and P_t−P_b.

At step 30, the photonic device obtains a result of a mathematical operation (e.g., the product of the first numeric value times the second numeric value) using the pair of encoded output signals. In some embodiments, step 30 involves i) detecting the pair of encoded output signals with a pair of balanced photodetectors to obtain a pair of photocurrents, ii) receiving the photocurrents with a differential trans-impedance amplifier to obtain an output voltage, and iii) converting the output voltage with an analog-to-digital converter to obtain a numeric value representing the result (e.g., the product). Referring for example to the architecture of FIG. 1A, at step 30, differential receiver 16 generates numeric value y, which equals the product of x times m.

As discussed above, differential optical multiplier 14 may be implemented using any suitable optical device. One such device is the tunable directional coupler depicted in FIG. 2A, in accordance with some embodiments. The directional coupler includes a pair of input optical waveguides, where one input optical waveguide receives optical signal P_t and the other input optical waveguide receives optical signal P_b. The directional coupler further includes a pair of output optical waveguides, where one output optical waveguide outputs optical signal P_t′ and the other output optical waveguide outputs optical signal K. The region between the pair of input of optical waveguides and the pair of output optical waveguides is labeled “A.” This is the region in which the input optical signals are combined to produce the output optical signals.

Region A is depicted in additional detail in FIG. 2B. In this region, the waveguides are sufficiently close to one another to produce evanescent coupling. The coupling coefficient with which the waveguides couple to one another depends, among other parameters, on the spatial distance d between the waveguides. The closer the waveguides, the larger the coupling coefficient. In some embodiments, the directional coupler may be designed so that d is adjustable based on an input signal. This may be achieved using nano-opto-electro-mechanical system (NOEMS) techniques. For example, the waveguides may be suspended in air in the region A, and the position of the suspended waveguides in the lateral direction (the direction in the plane of the waveguides perpendicular to the propagation axis) can be controlled with an external voltage. Referring again to FIG. 1A, voltage V_m may control the distance between the waveguides in region A.

Referring back to FIG. 2A, the letter “t” identifies the transmission coefficient and the letter “k” identifies the cross-coupling coefficient (from one waveguide to the other). The coefficients t and k are expressed in terms of amplitude transmission, and must obey the relationship |t|²+|k|²≤1 in order to provide for energy conservation. The equality is achieved when the multiplication device is lossless. Coefficients t and k depend on the distance between the waveguides d. Accordingly, voltage V_m controls the coefficients t and k.

The power P_t′ is equal to the fraction of P_t that travels on the top waveguide (equal to |t|² P_t), plus the fraction of P_b that couples to the top waveguide (equal to |k|² P_b), plus a cross term resulting from the interference of the signal at the top waveguide with the signal at the bottom waveguide. It should be appreciated that the interference term is non-zero if the input optical fields are coherent to one another (in other words, their phases are correlated). However, if the input optical signals are not mutually coherent, the interference term goes to zero. This is because mutually incoherent optical signals, having uncorrelated phases, do not interfere with each other. Therefore, assuming that the signals are not mutually coherent, the output power at the top waveguide is simply given by |t|²P_t +|k|²P_b. Similarly, the power P_b′ is equal to the fraction of P_b that travels on the bottom waveguide (equal to |t|²P_b), plus the fraction of P_t that couples to the bottom waveguide (equal to |k|²P_t), plus a cross term resulting from the interference of the signal at the bottom waveguide with the signal at the top waveguide. Again, the interference term is non-zero if the input optical fields are coherent to one another. However, if the input optical signals are not mutually coherent, the interference term goes to zero. Therefore, assuming that the signals are not mutually coherent, the output power at the bottom waveguide is simply given by |t|²P_b+|k|²P_t.

As discussed above, in dual-rail architectures of the types described herein, numeric values are encoded in the difference (power difference or amplitude difference) between optical signals. When the input signals are mutually incoherent, the difference P_t′−P_b′ is equal to (|t|²−|k|²)(P_t−P_b). It should be noted that, when the input signals are mutually incoherent, the directional coupler of FIG. 2A operates as an optical multiplier in that it outputs the product of the quantity |t|²−|k|²times the quantity P_t−P_b. However, the directional coupler does not output the product if the input signals are mutually coherent. Therefore, it is critical that the input signals be mutually incoherent.

Mutual incoherence may be achieved in a variety of ways. First, mutual incoherence may be achieved by using an incoherent light source, such as an LED, a source of amplified spontaneous emission or a source of stimulated emission with a relatively large linewidth. Second, mutual incoherence may be achieved by allowing one waveguide (e.g., the top waveguide) to travel a distance greater that the distance traveled by the other waveguide (e.g., the bottom waveguide), where the difference between the distances is greater that the coherence length of the light source (as discussed in detail below in connection with FIG. 4C).

Additionally, or alternatively, differential optical encoder 12 may implemented in some embodiments using the directional coupler of FIG. 2A. In such embodiments, the top input waveguide receives P_in, and the bottom input waveguides receives no optical signal (though the opposite is also possible). The output waveguides output, respectively, P_t=|t|²P_in and P=|k|² P_in, where t and k are controlled using voltage V_x (that is, by controlling distance d).

Another implementation of differential optical multiplier 14 is depicted in FIG. 3, in accordance with some embodiments. This embodiments includes a Mach-Zehnder interferometer. In this example, the coefficients t and k are controlled based on the phases (pi and introduced by the phase shifters (which in turn are controlled based on voltage V_m). The differential output optical power is in general given by: ΔP ∝sin (θ_v) cos (θ_w)+2 cos φ[cos (θ_w/2) sin (θ_w/2) (cos²(θ_v/2)−sin² (θ_v/2))], where φ is the difference between φ₁ and φ₂. However, when an incoherent light source is used or when the light source has a coherence length shorter than the optical delay line, the phase term cos (φ) time-averages to zero. This leaves ΔP ∝sin (θ_v) cos (θ_w)Therefore, the difference between the powers of the output optical signals is proportional to a product.

Both the implementations of FIG. 2A and FIG. 3 produce differential optical multiplication assuming that the input optical signals are mutually incoherent. This is because, when the signals are mutually incoherent, the interference term is zero. It should be noted that there are other ways to cause the interference term to be zero. For example, the interference term can be set to zero by defining the input optical signals on mutually orthogonal polarizations. Additionally, or alternatively, the interference term can be set to zero by defining the input optical signals on non-overlapping time bins. Additionally, or alternatively, the interference term can be set to zero by defining the input optical signals on different carrier frequencies.

The architecture of FIG. 4A is designed to cause the signals P_t and P_b to be combined to one another so that the interference term is zero. In addition to the architecture of FIG. 1A, this architecture further includes optical orthogonalization unit 13. Optical orthogonalization unit 13 is configured to render signals P_t and P_b orthogonal to one another. Here, two signals are orthogonal if, when the signals are combined, the resulting interference term is substantially zero. In the example of FIG. 4B, optical orthogonalization unit 13 includes optical polarization rotator 43. In this example, optical signals P_t and P_b are orthogonal in that they are defined on orthogonal polarizations.

In the example of FIG. 4C, optical orthogonalization unit 13 includes an optical delay line 45 (e.g., a serpentine-shaped waveguide). Here, the optical delay line is introduced to delay signal P_t relative to signal P_b by a sufficient amount to render the phases of the signals mutually uncorrelated. To achieve this result, the additional path introduced by optical delay line 45 must be greater than the coherence length of the light source. In some embodiments, the optical delay line may be sized to provide a constant quadrature phase difference.

Alternatively, or additionally, the optical delay line is introduced to delay signal P_t relative to signal P_b by a sufficient amount so that the signals do not overlap in time (are defined in different time bins). To achieve this result, the additional delay introduced by optical delay line 45 must be greater than the duration of the signal pulses.

In another example, optical orthogonalization unit 13 includes a device for defining the first rail on one carrier frequency and the second rail on another carrier frequency. This may be achieved, in some embodiments, using an optical non-linear medium. Alternatively, distinct carrier frequencies may be obtained by using two distinct lasers that emit at different wavelengths. The first laser emits at a first wavelength and supports the first rail, the second laser emits at a second wavelength and supports the second rail.

FIG. 4D depicts an implementation of the architecture of FIG. 4C, in accordance with some embodiments. Here, both differential optical encoder 12 and differential optical multiplier 14 are implemented using Mach-Zehnder interferometers 70. Each Mach-Zehnder interferometer includes a phase shifter 72. The phase shifter of differential optical encoder 12 is controlled using voltage V_x and the phase shifter of differential optical multiplier 14 is controlled using voltage V_m.

III. Photonic Processors Using Dual-Rail Multipliers

The dual-rail multiplier of FIG. 1A performs scalar multiplications (x times m). However, many machine learning algorithms rely on matrix-matrix (e.g., matrix-vector) multiplication. Some embodiments relate to photonic processors configured to perform matrix-matrix (e.g., matrix-vector) multiplications using the dual-rail optical multipliers described above. These photonic processors are configured to multiply a matrix M by a vector X to produce a vector Y. Matrix M is also referred to herein as “weight matrix,” vector X is also referred to herein as “input vector” and vector Y is also referred to herein as “output vector.” FIG. 5 illustrates an example of such a multiplication. In this example, M is an N×N matrix, though embodiments of the present application are not limited to square matrices or to any specific dimension.

An example of a dual-rail photonic processor is depicted in FIG. 6, in accordance with some embodiments. In this implementation, the photonic processor is configured to multiply a 2×2 matrix (M) by a 2×1 input vector (X) to obtain a 2×1 output vector (Y). A first light source 10 provides an input optical signal having power P_in1 and a second light source 10 provides an input optical signal having power P_in2 (although the same light source may be used in some embodiments). P_in2 may be equal to or different from P_in1. Each light source is followed by a differential optical encoder 12, which operates in the manner described in connection with the differential optical encoder of FIG. 1A. The first differential optical encoder receives voltage V_x1, which is representative of numeric value x₁. This differential optical encoder encodes the received input optical signal using voltage V_x1 to generate a pair of encoded optical signals having powers P_t1 and P_b1. These optical signals are provided as inputs to two differential optical multipliers 14. The top differential optical multiplier receives voltage V_M11, which represents numeric value M₁₁. The bottom differential optical multiplier receives voltage V_M21, which represents numeric value M₂₁. Both differential optical multipliers operate in the manner described in connection with the differential optical multiplier of FIG. 1A. The top differential optical multiplier outputs optical signals Pb₁₁′ and Pt_11,′ and the bottom differential optical multiplier outputs optical signals Pb₂₁′ and Pt₂₁′. The difference between Pb₁₁′ and Pt₁₁′ is proportional to both M₁₁ and the difference between Pb₁ and Pt₁, and accordingly, is proportional to the product M₁₁x₁. Similarly, the difference between Pb_21′ and Pt₂₁ is proportional to both M₂₁ and the difference between Pb₁ and Pt₁, and accordingly, is proportional to the product M₂₁x₁. The four output optical signals are detected using respective photodetectors 60.

The second differential optical encoder receives voltage V_x2, which is representative of numeric value x₂. This differential optical encoder encodes the received input optical signal using voltage V_x2 to generate a pair of encoded optical signals having powers P_t2 and P_b2. These optical signals are provided as inputs to two differential optical multipliers 14. The top differential optical multiplier receives voltage V_M12, which represents numeric value M₁₂. The bottom differential optical multiplier receives voltage V_M22, which represents numeric value M₂₂. Both differential optical multipliers operate in the manner described in connection with the differential optical multiplier of FIG. 1A. The top differential optical multiplier outputs optical signals Pb₁₂′ and Pt₁₂′, and the bottom differential optical multiplier outputs optical signals Pb_22′ and Pt_22′. The difference between Pb₁₂′ and Pt₁₂′ is proportional to both Mu and the difference between Pb₂ and Pt₂, and accordingly, is proportional to the product M₁₂x₂. Similarly, the difference between Pb₂₂′ and Pt₂₂′ is proportional to both M₂₂ and the difference between Pb₂ and Pt₂, and accordingly, is proportional to the product M₂₂x₂. The four output optical signals are detected using respective photodetectors 60.

As shown in FIG. 6, the outputs of the photodetectors are combined (see, e.g., node 62), thereby allowing the photocurrents to be added to one another. Receivers 64 receive the photocurrents produced by detectors 60. Receivers 64 include a differential trans-impedance amplifier (or other circuits for subtracting the first input current from the second input current) and an analog-to-digital converter. The top receiver 64 outputs numeric value y₁₌M₁₁x₁+M₁₂x₂. The bottom receiver 64 outputs numeric value y₂₌M₂₁x₁+M₂₂x₂.

An example of a 4×4 dual-rail photonic processor is illustrated in FIG. 7A. This processor includes four light sources 10 (not shown), four differential encoders 12 (which encode numeric values x₁, x₂, x₃ and x₄ into respective pairs of dual-rail optical signals), four optical splitter trees 75 (shown in additional detail in FIG. 7B), sixteen differential optical multipliers 14 (which multiply the respective inputs by M₁₁, M₂₁, M₃₁, M₄₁, M₁₂, M₂₂, M₃₂, M₄₂, M₁₃, M₂₃, M₃₃, M₄₃, M₁₄, M₂₄, M₃₄ and M₄₄), thirty-two detectors 60 and four receivers 64.

FIG. 7B illustrates a portion of the photonic processor of FIG. 7A in additional detail. More specifically, FIG. 7B illustrates an optical splitter tree 75. Optical splitter tree 75 delivers the optical signals produced by a differential optical encoder 12 to multiple differential optical multipliers 14. In this implementation, the optical splitter tree includes multiple 3db optical splitters 76 and multiple optical waveguide crossings 77. Each crossing may be implemented using two levels of waveguides. For example, one waveguide level (e.g., the bottom level) may be made of silicon and another waveguide level (e.g., the top level) may be made of silicon nitride.

FIG. 8 illustrates an alternative optical splitter tree that does not include optical waveguide crossings. The inventors have recognized that passing light through a waveguide crossing leads to optical loss, thereby negatively affecting the overall performance of the photonic processor. The tree of FIG. 8 improves the performance of the photonic processor because waveguide crossings are omitted (another example of such an optical splitter tree is described below in connection with FIG. 12C).

IV. Optical Multipliers and Processors Based on Modulatable Detectors

The inventors have appreciated that some optical multipliers occupy substantial chip real estate due to the presence of lengthy optical interferometers. This limits the number of multipliers that can be integrated on a single chip, thus limiting the computational capabilities of photonic processors that employ these multipliers. Some embodiments relate to compact optical multipliers that are based on modulatable detectors. Modulatable detectors are optical detectors having at least one characteristic that can be controlled by a user using an electric control signal. These detectors are designed so that varying the magnitude of a control signal (e.g., a voltage or a current) alters a characteristic of the detector, such as the detector's responsivity, gain, impedance, conductance, etc. The result is that the detector's photocurrent depends not only on the optical power that the detector receives, but also on the control signal applied to the detector. Optical multipliers based on modulatable detectors are designed so that one of the factors to be multiplied modulates the modulatable characteristic. For example, in some of the embodiments in which the modulatable characteristic is the detector's responsivity, the responsivity may be controlled based on a weight parameter m.

FIG. 9A is a block diagram illustrating a modulatable detector, in accordance with some embodiments. Modulatable detector 90 receives as input an optical signal having power P, and in response, produces a photocurrent i (the dark current of the modulatable detector will be neglected from this discussion). The photocurrent i is related to the input optical power by the following expression: i=P/R, where R is the responsivity of the modulatable detector. In addition to being dependent upon power P, the photocurrent is also dependent upon control voltage V (or, in other embodiments, another type of electric signal such a control current). This detector is preceded by the term “modulatable” to indicate that the photocurrent can be modulated through the application of a control electric signal. Example implementations of modulatable detector 90 are described in detail further below.

FIG. 9B is a block diagram illustrating a pair of modulatable detectors arranged according to a dual-rail architecture, in accordance with some embodiments. The top modulatable detector receives optical power P₊ and control voltage V₁. Photocurrent ii depends upon optical power P₊ and control voltage V₁. Similarly, the bottom modulatable detector receives optical power P₋ and control voltage V₂. Photocurrent i₂ depends upon optical power P₋ and control voltage V₂.

FIG. 9C is a block diagram illustrating the interior of a modulatable detector, in accordance with some embodiments. Modulatable detector 90 includes a photodetector 91 and a gain stage 92. Photodetector 91 may be implemented using a photoconductor, a photodiode, an avalanche photodiode, a phototransistor, a photomultiplier (e.g., a tube), a superconducting detector, a graphene-based detector, or any other suitable type of photo-sensitive device. Gain stage 92 can include, for example, a current amplifier for amplifying the current generated by photodetector 91. In some embodiments, modulating a characteristic of modulatable detector 90 may involve modulating, using voltage V (or using a control current), the responsivity of photodetector 91 and/or the gain (e.g., current gain) or impedance of gain stage 92. FIG. 9D is similar to the diagram of FIG. 9C, but illustrates a pair of modulatable detectors arranged in a dual-rail configuration. As shown, each photodetector 91 is coupled to both gain stages 92.

FIG. 10A is a block diagram illustrating an optical multiplier based on a modulatable detector, in accordance with some embodiments. Just like the multiplier of FIG. 1A, this multiplier includes a light source 10 and a controller 17, which includes a pair of D/As 19 and a numeric value unit 18.

Optical encoder 82 generates an encoded optical signal having power P, that is proportional to both numeric value x and input power P_in. Optical encoder 82 may be implemented using any optical modulator, including an optical interferometer (e.g., a tunable directional coupler or a Mach-Zehnder interferometer), a resonant modulator, a Franz-Keldysh modulator, etc.

Modulatable detector 90 multiplies numeric value x by numeric value m. This is achieved by producing a photocurrent i that is proportional to both P_x and m (by way of voltage V_m), and accordingly, is proportional to P_{in, x} and m. Receiver 96 includes a trans-impedance amplifier and an analog-to-digital converter. Receiver 96 produces an output numeric value y that is equal to the product of x times m.

FIG. 10B is a flow chart illustrating a method for performing mathematical operations (e.g., multiplications), in accordance with some embodiments. Method 200 can be performed using any suitable optical device, including the modulatable detector-based optical multiplier depicted in FIG. 10A.

Method 200 begins at step 202, in which an optical device receives an input optical signal. Referring for example to the architecture of FIG. 10A, at step 202, optical encoder 82 receives optical signal P_in.

At step 204, the photonic device obtains a first numeric value and a second numeric value. These are the numeric values to be multiplied. The numeric values need not be obtained at the same time. Referring for example to the multiplier of FIG. 10A, at step 204, numeric value unit 18 generates numeric values x and m based on data stored in a memory of controller 17 and/or on data obtained from another computing system. These numeric values may represent any type of information, such as text, audio, video, etc. The numeric value may be real or complex, positive or negative.

At step 206, the photonic device generates an encoded optical signal by modifying the input optical signal using the first numeric value. Referring for example to the multiplier of FIG. 10A, at step 206, optical encoder 82 produces an encoded optical signal having a power proportional to both P_in and x.

At step 208, the photonic device generates a photocurrent. The generation involves detecting the encoded signal using a modulatable detector and setting a characteristic of the modulatable detector based on the second numeric value. Referring for example to the multiplier of FIG. 10A, at step 208, modulatable detector 90 produces photocurrent i, which involves detecting optical signal P_x and setting a characteristic of the modulatable detector based on voltage V_m.

At step 210, the photonic device obtains a result of the mathematical operation using the photocurrent. In some embodiments, the result represents the product of the first numeric value times the second numeric value. In some embodiments, this step involves generating a voltage based on the photocurrent generated at step 208, and converting the voltage to the digital domain. Referring for example to the multiplier of FIG. 10A, step 210 may be performed using receiver 96.

The optical multiplier of FIG. 10A is based on a single-rail architecture. In some embodiments, however, modulatable detector-based optical multipliers are arranged in accordance with a dual-rail architecture. In this way, the multiplier further leverages the benefits described above in connection with FIG. 1A, including increased immunity to noise and reduced optical loss.

FIG. 11 is a block diagram of a dual-rail modulatable detector-based optical multiplier, in accordance with some embodiments. This optical multiplier combines the architecture of FIG. 1A with the architecture of FIG. 10A. Differential optical encoder 12 produces a pair of encoded optical signals having powers P_t and P_b, respectively. The difference between the optical signals (e.g., the difference in signal power or signal amplitude) is proportional to both numeric value x and input power P_in. This optical multiplier includes a pair of modulatable detectors 90. The top modulatable detector produces a photocurrent i_t, the bottom modulatable detector produces a photocurrent i_b. Photocurrent it depends on both optical power P_t and voltage V_m. Similarly, photocurrent i_b depends on both optical power P_b and voltage −V_m. The difference between it and i_b is proportional to both m and V_t−V_b, and accordingly, is proportional to P_in, x and m. In some embodiments, the two modulatable detectors can be coupled together as described in FIG. 9D such that both photocurrents it and ib depend on both optical powers Pt and Pb and Vm. In this configuration, the modulatable detectors steer the photocurrents to the right output rail to maintain the large photocurrents generated at the detectors themselves. Differential receiver 96 includes a differential trans-impedance amplifier (or other circuits for subtracting the first photocurrent from the second photocurrent) and an analog-to-digital converter.

The modulatable detector-based multipliers of FIG. 10A and FIG. 11 perform scalar multiplications (x times m). However, some embodiments relate to photonic processors including modulatable detectors designed to compute matrix-matrix (e.g., matrix-vector) multiplications. These photonic processors may be implemented using single-rail or dual-rail architectures.

An example of a modulatable detector-based dual-rail photonic processor is depicted in FIG. 12A, in accordance with some embodiments. In this implementation, the photonic processor is configured to multiply a 2×2 matrix (M) by a 2×1 input vector (X) to obtain a 2×1 output vector (Y). This processor operates in a manner similar to the multiplier described in connection with FIG. 11. A first light source 10 provides an input optical signal having power P_in1 and a second light source 10 provides an input optical signal having power P_in2 (although the same light source may be used in some embodiments). P_in2 may be equal to or different from P_in1. Each light source is followed by a differential optical encoder 12, which operates in the manner described in connection with the differential optical encoder of FIG. 11. The first differential optical encoder receives voltage V_x1, which is representative of numeric value x₁. This differential optical encoder encodes the received input optical signal using voltage V_x1 to generate a pair of encoded optical signals having powers P_t1 and P_b1. These optical signals are provided as inputs to a set of modulatable detectors 90. The top pair of modulatable detectors receives voltages V_M11 and −V_M11, which represent numeric value M₁₁. The bottom pair of modulatable detectors receives voltages V_M21 and −V_M21, which represent numeric value M₂₁. Both differential pairs of modulatable detectors operate in the manner described in connection with FIG. 11. The top pair of modulatable detectors outputs photocurrents ibis and itii, and the bottom pair of modulatable detectors outputs photocurrents i_b21 and i_t21. The difference between ibis and itis is proportional to both M₁₁ and the difference between Pb₁ and Pt₁, and accordingly, is proportional to the product M₁₁x₁. Similarly, the difference between i_b21 and i_t21 is proportional to both M₂₁ and the difference between Pb₁ and Pt₁, and accordingly, is proportional to the product M₂₁x₁.

The second differential optical encoder receives voltage V_x1, which is representative of numeric value x₂. This differential optical encoder encodes the received input optical signal using voltage V_x2 to generate a pair of encoded optical signals having powers P_t2 and P_b2. These optical signals are provided as inputs to a set of modulatable detectors 90. The top pair of modulatable detectors receives voltages V_M12 and −V_M11, which represent numeric value M₂₁. The bottom pair of modulatable detectors receives voltages V_M22 and −V_M22, which represent numeric value M₂₂. The top pair of modulatable detectors outputs photocurrents i_b11 and i_t11, and the bottom pair of modulatable detectors outputs photocurrents i_b22 and i_t22. The difference between i_b12 and i_t12 is proportional to both M₁₂ and the difference between P_b2 and P_t2, and accordingly, is proportional to the product M₁₂x₂. Similarly, the difference between i_b22 and i_t22 is proportional to both M₂₂ and the difference between P_b2 and P_t2, and accordingly, is proportional to the product M₂₂x₂.

The outputs of the photodetectors are combined (see, e.g., node 62), thereby allowing the photocurrents to be added to one another. Receivers 96 receive the photocurrents produced by detectors 90. Receivers 96 include a differential trans-impedance amplifier (or other circuits for subtracting the first input current from the second input current) and an analog-to-digital converter. The top receiver 96 outputs numeric value y₁=M₁₁x₁+M₁₂x₂. The bottom receiver 96 outputs numeric value y₂=M₂₁x₁+M₂₂x₂.

An example of a 4×4 modulatable detector-based dual-rail photonic processor is illustrated in FIG. 12B. This processor includes four light sources 10 (not shown), four differential encoders 12 (which encode numeric values x₁, x₂, x₃ and x₄ into respective pairs of dual-rail optical signals), four optical splitter trees 75, thirty-two modulatable detectors 90 (which multiply the respective inputs by M₁₁, M₂₁, M₃₁, M₄₁, M₁₂, M₂₂, M₃₂, M₄₂, M₁₃, M₂₃, M₃₃, M₄₃, M₁₄, M₂₄, M₃₄ and M₄₄), and four receivers 96.

Although not expressly illustrated in FIGS. 11, 12A and 12B, modulatable detectors 90 may be arranged according to the scheme depicted in FIG. 9D. Accordingly, the photodetector 91 of a pair of modulatable detectors control both the gain stage that is directly coupled to the photodetector and the gain stage that is directly coupled to the other photodetector of the pair.

FIG. 12C illustrates an alternative optical splitter tree that does not include optical waveguide crossings 77. As in the tree of FIG. 12A, each differential optical encoder feeds four modulatable detectors. In the tree of FIG. 12A, these four modulatable detectors 90 are all oriented in the same direction. To the contrary, in the tree of FIG. 12C, half of the modulatable detectors 90 are oriented in one direction and half of the modulatable detectors 90 are oriented in another direction (e.g., the opposite direction). The arrangement with modulatable detectors oriented in opposite directions is shown in the inset of FIG. 12C. Having sets of modulatable detectors oriented in different directions allows for optical splitter trees that omit waveguide crossings, and as a result, reduces optical loss.

It should be noted that the photonic processors described in connection with FIGS. 12A-12B have relatively short optical paths, especially if compared to the photonic processors of FIGS. 6 and 7A. Compare for example the photonic processor of FIG. 12A with the photonic processor of FIG. 6. In the photonic processor of FIG. 12A, the optical paths extend only from the light sources to the modulatable detectors. By contrast, the photonic processor of FIG. 6 has longer optical paths which encompass, in addition to the optical paths of FIG. 12A, the differential optical amplifiers.

Despite the shorter optical paths, the photonic processor of FIG. 12A takes substantial advantage of the physical nature of optical signal processing. The inventors have appreciated that analog accelerators that are implemented to operate entirely in the electrical domain suffer from unwanted coupling between the different stages of the accelerator. In analog electronic accelerators, in addition to the forward electrical path, there is an unwanted backward path that couples electrical signals in the backward direction. This results in a variety of negative effects, including noise increase and speed decrease. By contrast, optical signals by nature travel only in one direction, the forward direction. Accordingly, there are generally no optical signals traveling in the backward direction which may otherwise negatively affect the performance of the photonic processor (although there may be optical back-reflections in some embodiments, the extent of these reflections can be reduced substantially using proper optical terminations). The inventors have appreciated that encoding numeric values in the optical domain as for example shown in FIG. 12A effectively isolates the encoder from the multiplier, thereby avoiding the negative effects that exist in fully-electrical accelerators.

V. Examples of Modulatable Detectors

As discussed above, modulatable detectors of the types described herein are optical detectors having at least one characteristic that can be controlled by a user using one or more electric control signals. Therefore, modulatable detectors have at least one electrical control terminal. These detectors are designed so that application of a control signal (e.g., a voltage, current or charge) alters a characteristic of the detector, such as the detector's responsivity, gain, impedance, etc. The result is that the detector's photocurrent depends not only on the optical power incident on the detector, but also on the control signal applied to the detector.

As discussed above in connection with FIG. 9C-9D, in some embodiments, modulating a characteristic of a modulatable detector may involve modulating the responsivity of one or more photodetectors 91. FIG. 13 illustrates an example of a photodetector 91, in accordance with some embodiments. This photodetector is configured to operate as a photoconductor—incident light produces electron-hole pairs which cause current to flow under the presence of an electric field. The photoconductor includes an optical waveguide 1300 (e.g., a silicon waveguide), a photo-absorption region 1302 including germanium (though absorbing materials other than germanium may be used in some embodiments), and electrodes 1304 and 1306. Optical waveguide 1300 provides light to germanium region 1302. Assuming that the wavelength is below the germanium's absorption cut-off wavelength (about 1.9 μm in some embodiments), the incident light is absorbed by the germanium and electron-hole pairs are generated. If a voltage is applied between electrodes 1304 and 1306, an electric field is established that sweeps the photogenerated carriers out of the germanium region, thereby giving rise to a photocurrent. In this implementation, the modulatable characteristic is the photoconductor's responsivity. The inventors have appreciated, in fact, that the responsivity R (measured in amperes per watts) of this photoconductor depends on the voltage V applied to the electrodes. In particular, the responsivity is given by the following expression:

$R=\frac{q}{\mathrm{hv}}\frac{{\tau }_n{\mu }_{n}V}{w^2}{\eta }_i\left(1-e^{-\alpha l}\right)$

where τ_n is the electron recombination time, μ_n the electron mobility, w is the device electrode spacing, η_i the intrinsic quantum efficiency, α is the absorption coefficient of the detector material, v is the frequency of the incident light and l is the device depth in the longitudinal direction with respect to the light propagation direction. It should be noted that, because the responsivity R is proportional to voltage V, the photocurrent is also proportional to V. Therefore, the photocurrent can be controlled by varying V.

FIGS. 14A-14B illustrate another example of a photodetector 91, in accordance with some embodiments. FIG. 14A is a top view and FIG. 14B is a cross sectional view taken along the BB line of FIG. 14A. This photodetector includes a waveguide 100 (e.g., a silicon or silicon nitride waveguide), a highly doped n region (n+ region) 102, an intrinsic region (i region) 103, a p region 104, an oxide layer (e.g., silicon dioxide) 105, a poly-silicon layer (or a layer made of another conductive material) 106, an intrinsic region (i region) 107, a photo-absorption region 108 including germanium (or including another absorbing material) and a highly doped p region (p+ region) 110. In some embodiments, the germanium region is also shaped to form an optical waveguide.

Waveguide 100 abuts against germanium region 108. In this way, light traveling down waveguide 100 is transmitted to germanium region 108, and as a result, is absorbed. Germanium region 108 is positioned on top of intrinsic region 107. For example, germanium region 108 is grown epitaxially on silicon. The highly doped regions 102 and 110 are connected to respective electrodes. P region 104 is positioned adjacent to germanium region 108. Oxide layer 105 is positioned on top of p region 104, and poly-silicon layer 106 is positioned on top of oxide layer 105.

Collectively, p region 104, oxide layer 105 and poly-silicon layer 106 form a metal-oxide-semiconductor capacitor (MOS cap). It should be appreciated that control capacitors other than the MOS cap may be used in some embodiments, including for example a Shottky junction-capacitor or a graphene-based capacitor. FIG. 14C illustrates the energy band diagram of the photodetector of FIG. 14B along the CC line. The diagram illustrates four bands. The top bands represent the conduction band when a low bias voltage is applied to the MOS cap and the conduction band when a large bias voltage is applied to the MOS cap. The bottom bands represent the valance band when a low bias voltage is applied to the MOS cap and the valence band when of large a bias voltage is applied to the MOS cap.

The voltage applied between the n+ and p+ regions controls the electric field along the x-axis. In the diagram of FIG. 14C, this voltage produces a reverse bias. The reverse bias, in turn, produces an electric field oriented along the x-axis. This electric field sweeps photogenerated carriers away from the germanium region, assuming that the photocarriers have sufficient energy to overcome the energy barrier existing at the interface between the Ge region 108 and the p region 104.

The voltage applied to the MOS cap or other control capacitors (referred to as the gate voltage) determines the extent of the electron and hole energy barriers at the interface between the Ge region 108 and the p region 104. When the bias voltage applied to the MOS cap is low, both the electron and the hole energy barriers are relatively large. Under these conditions, carriers photogenerated in the germanium region are blocked, and as a result, do not produce a significant photocurrent. To the contrary, when the bias voltage applied to the MOS cap is large, both the electron and the hole energy barriers are relatively low. Under these conditions, carriers photogenerated in the germanium region have sufficient energy to overcome the respective barriers, and as a result, produce a photocurrent. Thus, the voltage applied to the MOS cap controls the responsivity of the photodetector.

In some embodiments, a photodetector includes an avalanche region in which photogenerated carriers experience impact ionization, thereby producing gain. In the example of FIGS. 14A-14B, i region 103 forms an avalanche region. In other embodiments, the avalanche region may include a quasi-i region (e.g., a region with doping concentration less than 10⁻¹⁴ cm⁻ 3). Impact ionization occurs in the avalanche region, which amplifies the photocurrent. Thus, the presence of the avalanche region increases the sensitivity of the photodetector. In some embodiments, the responsivity of the photodetector may be modulated by controlling the gain associated with the avalanche region.

FIG. 15A is a plot illustrating the responsivity of a photodetector of the type depicted in FIGS. 14A-14B. In this example, the germanium region has a width (measured in the direction parallel to the x-axis) of 500 nm. As shown, varying the voltage applied to the MOS cap (the gate voltage) results in a change in the responsivity of the photodetector. As discussed above, this change in responsivity occurs because the voltage controls the extent of the energy barriers.

FIG. 15B is another plot illustrating the responsivity of a photodetector of the type depicted in FIGS. 14A-14B. Again, the width of the germanium region is 500 nm. In this case, the power of the input optical signal is also varied. As shown, varying the input power also causes a change in the responsivity, which is undesirable because it negatively effects the linearity of the photodetector. FIGS. 15C and 15D illustrate similar plots. In the plot of FIG. 15C, the width of the germanium region is 1000 nm. In the plot of FIG. 15D, the width of the germanium region is 1500 nm. These plots illustrate that the responsivity's dependence upon the input power can be mitigated by properly selecting the width of the germanium region.

FIG. 15E further illustrates that the responsivity exhibits a weak temperature-dependence. As for the power-dependence, the temperature-dependence may be mitigated by properly selecting the width of the germanium region.

The inventors have appreciated that the speed of the photodetector of FIGS. 14A-14B depends upon the gate voltage applied to the MOS cap. When the gate voltage is high enough to drive the MOS cap into inversion, the photogenerated carriers are quickly swept away from the germanium region. When no gate voltage is applied, the photogenerated carriers are stored in germanium region due to the presence of the energy barrier, and largely decay by carrier recombination processes.

FIG. 16A shows the photocarrier concentration as a function of time for a temporally narrow incident optical pulse. As shown, there are two time constants associated with the photocarriers: one associated with the rise time (Ty) and another associated with the fall time (′rR). The rise time is associated with the creation of photogenerated carriers. The fall time, in the absence of a photocarrier sweeping field, is associated with carrier recombination processes (which are typically on the order of nanoseconds in imperfect semiconductors). This can be significantly sped up by introducing a “sweeping field”. The sweeping field can be applied periodically to remove photogenerated carriers and effectively increase the bandwidth of the device as shown in FIG. 16B. One possible implementation of the sweeping field would be to apply a high voltage pulse to the MOS cap.

As discussed above, the inventors have appreciated that the photodetector may be designed to exhibit avalanche multiplication when an avalanche region is included between the p region 104 and the n+ region 102, as shown in FIG. 14B. FIG. 17 depicts the I-V characteristics in reverse bias of a representative photodetector with varying gate voltage. As shown, avalanche multiplication kicks in at around 10 V in this example.

As discussed above in connection with FIG. 9C-9D, in some embodiments, modulating a characteristic of a modulatable detector may involve modulating the current gain of one or more gain stages 92. FIGS. 18A-18H are circuit diagrams of various modulatable detectors having gain stages characterized by modulatable current gains, in accordance with some embodiments. In these modulatable detectors, gain stage 92 is implemented using transistors. Further, in these modulatable detectors, optical power P_x (and optionally, voltage −P_x) is encoded with numeric value x; voltage V_m (and optionally, voltage −V_m) is encoded with numeric value m and modulates the current gain; and photocurrent i_xm (and optionally, current −i_xm) represents the product of numeric value x times numeric value m. It should be appreciated that, while these figures illustrate gain stages formed using MOS transistors, any other suitable type of transistor may be used, including bipolar junction transistors (BJTs) and junction field effect transistors (JFETs). Thus, as used herein, the term “gate” indicates the gate of a field effect transistor or the base of a bipolar transistor, the term “source” indicates the source of a field effect transistor or the emitter of a bipolar transistor and the term “drain” indicates the drain of a field effect transistor or the collector of a bipolar transistor. It should also be noted that the drain and source of a transistor may be interchangeable. Accordingly, in some embodiments, when the drain of a transistor is described as being coupled to a certain terminal, this includes the configuration in which the source of the transistor is coupled to the terminal (and vice versa).

The modulatable detectors of FIGS. 18A-18H allow for encoding of signed numeric values. In these embodiments, in fact, the sign of a numeric value (positive or negative) is represented based on the direction of the photocurrent(s).

The modulatable detector of FIG. 18A includes a photodetector 91 and a transistor T1 serving as gain stage 92. Photodetector 91 detects an optical signal P_x. The source of transistor T1 is coupled to photodetector 91. The gate of transistor T1 receives a voltage V_m. Voltage V_m controls transistor T1's gate-source voltage, and as a result, the current gain provided by the transistor. Photocurrent i_xm is proportional to both numeric value x and numeric value m, thereby representing the product of these numeric values. The inventors have appreciated that such a single-ended/unbalanced circuit has three limitations. First, this single-ended circuit is susceptible to common-mode and power-supply noise, both of which degrade the performance of the modulatable detector. Second, a low magnitude voltage V_m turns off transistor T1, thus severely limiting the transistor's speed. Third, blocking of some or all the photocurrent results in charge buildup that is dependent on optical power, resulting in an effective responsivity that is dependent not only on the applied voltage V_m but also on the input optical power.

FIG. 18B illustrates another modulatable detector, in accordance with some embodiments. This modulatable detector includes two photodetectors 91 to provide balanced detection of the differential optical input signal P_x with enhanced common-mode rejection ratio (CMRR) on the P_x inputs. The gain stage 92 includes a complementary MOS (CMOS) inverter—a p-type (PMOS) transistor T2 and an n-type (NMOS) transistor T3 enable modulation via a single-ended voltage V_m. The photocurrent 2i_xm represents the product of numeric value x times numeric value m. However, the inventors have appreciated that this modulatable detector suffers from poor CMRR on the V_m input. For the circuit to operate properly, the PMOS and NMOS transistors need to be sized properly to provide equal drive strength. Any process variation between the PMOS and NMOS transistors results in systematic offset error, thereby negatively affecting the circuit's ability to provide equal drive strength. Further, this modulatable detector suffers from the second and third problems described above in connection with the modulatable detector of FIG. 18A.

FIG. 18C illustrates another modulatable detector, in accordance with some embodiments. This embodiment involves balanced detection (see detectors 91) and single-ended modulation (see transistor T4, which serves as gain stage 92). The photocurrent 2i_xm represents the product of numeric value x times numeric value m. However, the inventors have appreciated that this modulatable detector suffers from the same drawbacks described above in connection with the modulatable detector of FIG. 18B.

FIG. 18D illustrates another modulatable detector, in accordance with some embodiments. This modulatable detector includes a gain stage with two transistors (T5 and T6) arranged in a differential pair configuration to provide differential modulation. The inventors have appreciated that this configuration leads to a large CMRR on the V_m input. Further, this modulatable detector circumvents the speed limitation described in connection with FIGS. 18A-18C. Due to the differential nature of the modulating electrical input signal V_m, at least one of the transistors is always turned on, thus enabling high speed operation. In addition, because a current flows continuously through the gain stage, the optical power-dependent charge buildup is negligible. As a result, the overall responsivity of the modulatable detector does not depend on the optical power. However, the inventors have appreciated that this modulatable detector suffers from low CMRR on the P_x input.

The modulatable detectors of FIGS. 18E-18H provide fully differential modulation and balanced detection and address the problems described in connection with FIGS. 18A-18D. FIG. 18E illustrates another modulatable detector, in accordance with some embodiments. This modulatable detector includes a gain stage with two differential pairs (see transistor pair T7-T8 and transistor pair T9-T10) with cross-coupled connections. The drain of transistor T7 is coupled to the drain of transistor T9, and the drain of transistor T8 is coupled to the drain of transistor T10. Photodetectors 91 are coupled to the sources of transistors T7-T10. Voltages V_m and −V_m control the current gain of the gain stage, and photocurrents 2i_xm and −2i_xm represent the product of numeric value x times numeric value m. The modulatable detector of FIG. 18E provides high speed operation, optical-power-independent overall responsivity, robustness to process variation and mismatch, and large CMRR on both P_x and V_m inputs.

FIG. 18F illustrates another modulatable detector, in accordance with some embodiments. This modulatable detector is similar to the modulatable detector of FIG. 18E. However, this modulatable detector includes a gain stage having CMOS inverters with differential pairs of transistors. PMOS transistor T11 and NMOS transistor T12 have drains that are coupled to one another. Similarly, PMOS transistor T13 and NMOS transistor T14 have drains that are coupled to one another. One photodetector 91 is coupled to the sources of transistors T12 and T14. One photodetector 91 is coupled to the sources of transistors T11 and T13. This modulatable detector can be viewed as the differential version of the modulatable detector of FIG. 18B.

FIG. 18G illustrates another modulatable detector, in accordance with some embodiments. This modulatable detector is similar to the modulatable detector of FIG. 18E, but it includes a gain stage with one differential pair of transistors T15-T16. This modulatable detector includes a node N1, which is coupled to both photodetectors 91 and to the sources of transistors T15 and T16. This modulatable detector can be viewed as the fully differential version of the modulatable detector of FIG. 18C.

FIG. 18H illustrates another modulatable detector, in accordance with some embodiments. This modulatable detector includes a gain stage with two CMOS transmission gates. The drains of transistors T17 and T18 are coupled to each other and the sources of transistors T17 and T18 are also coupled to each other. Similarly, the drains of transistors T19 and T20 are coupled to each other and the sources of transistors T19 and T20 are also coupled to each other. NMOS transistor T17 and PMOS transistor T18 form a first complementary transmission gate. NMOS transistor T20 and PMOS transistor T19 form a second complementary transmission gate. Node N2 is coupled to both complementary transmission gates and to both photodetectors 91.

FIG. 19A illustrates the steady-state overall responsivity (measured in terms of photocurrent 2i_xm and input power P_x) for the modulatable detector FIG. 18E plotted as a function of voltage V_m. The other modulatable detectors have similar responses. This plots reflects the fact that voltage V_m modulates the current gain provided collectively by transistors T7-T10. It should be noted that the overall responsivity is nearly independent of the input optical power. FIG. 19B illustrates the transient overall responsivity as a function of time. The transient response exhibits a settling time of less than 500 ps, thus indicating high-speed operation in the gigahertz range.

Photonic processors of the types described herein may include tens if not hundreds of thousands of modulatable detectors. For example, a photonic processor configured to perform multiplications on 256×256 matrices may include, in some embodiments, as many as 131,072 modulatable detectors. The inventors have appreciated that it would be desirable to integrate the photonic processor on a single chip in order to reduce manufacturing costs, increase speed of operation and limit power consumption. Because the photonic processors of the types described herein include both photonic and electronic circuits, integrating a photonic processor on a single chip involves electronic-photonic co-integration. This may be achieved at least in two ways.

The first way involves forming transistors and silicon photonics on the same silicon substrate. For example, silicon photonics and transistors may be formed on the same silicon layer. A representative arrangement is depicted in FIG. 20A, in accordance with some embodiments. In this case, both the transistors and the silicon photonics are formed on substate 300 (e.g., a silicon bulk substrate or a silicon-on-insulator substrate). A cladding 302 (e.g., silicon dioxide layer) is formed between the substrate's handle and silicon layer 304. Transistors 306 (which include, among others, the transistors used in any of the arrangements of FIGS. 18A-18H and, optionally, the transistors that constitute controller 17) are patterned on silicon layer 304. Similarly, silicon photonics 308 (which include, among others, the optical encoders, optical multipliers, splitter trees described above) are also patterned on silicon layer 304. Germanium region 310 (which may be used to form any of the photodetectors described above) is formed on top of silicon layer 304, for example via epitaxial growth.

The second way involves forming the silicon photonics and the transistors on separate substrates and bonding the substrates together. A representative arrangement is depicted in FIG. 20B, in accordance with some embodiments. Here, transistors 306 are formed on substrate 300 and silicon photonics 308 on substrate 400. As in the previous example, silicon photonics 308 are formed on a silicon layer (404), and germanium region 310 is formed on top of silicon layer 404. Substrates 300 and 400 are bonded to one another using flip-chip bonding techniques. For example, the substrates may be placed in electrical communication using through-oxide vias (as shown in FIG. 20B) passing through cladding 414 and/or using through-silicon vias (not shown in FIG. 20B).

VI. Conclusion

Having thus described several aspects and embodiments of the technology of this application, it is to be appreciated that various alterations, modifications, and improvements will readily occur to those of ordinary skill in the art. Such alterations, modifications, and improvements are intended to be within the spirit and scope of the technology described in the application. It is, therefore, to be understood that the foregoing embodiments are presented by way of example only and that, within the scope of the appended claims and equivalents thereto, inventive embodiments may be practiced otherwise than as specifically described. In addition, any combination of two or more features, systems, articles, materials, and/or methods described herein, if such features, systems, articles, materials, and/or methods are not mutually inconsistent, is included within the scope of the present disclosure.

Also, as described, some aspects may be embodied as one or more methods. The acts performed as part of the method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed in an order different than illustrated, which may include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments.

All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and/or ordinary meanings of the defined terms.

The indefinite articles “a” and “an,” as used herein in the specification and in the claims, unless clearly indicated to the contrary, should be understood to mean “at least one.”

The phrase “and/or,” as used herein in the specification and in the claims, should be understood to mean “either or both” of the elements so conjoined, i.e., elements that are conjunctively present in some cases and disjunctively present in other cases.

As used herein in the specification and in the claims, the phrase “at least one,” in reference to a list of one or more elements, should be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each and every element specifically listed within the list of elements and not excluding any combinations of elements in the list of elements. This definition also allows that elements may optionally be present other than the elements specifically identified within the list of elements to which the phrase “at least one” refers, whether related or unrelated to those elements specifically identified.

The terms “approximately” and “about” may be used to mean within ±20% of a target value in some embodiments, within ±10% of a target value in some embodiments, within ±5% of a target value in some embodiments, and yet within ±2% of a target value in some embodiments. The terms “approximately” and “about” may include the target value.

Claims

1. (canceled)

2. A photonic device configured to perform a mathematical operation comprising:

an optical encoder;

a modulatable detector coupled to an output of the optical encoder, the modulatable detector comprising a first photodetector and a first transistor; and

a controller coupled to both the optical encoder and the modulatable detector, the controller being configured to: obtain a first numeric value and a second numeric value, control the optical encoder to generate an encoded optical signal by modifying an input optical signal using the first numeric value, control the first transistor using the second numeric value to generate a photocurrent in response to the first photodetector receiving the encoded optical signal, and obtain a result of the mathematical operation using the photocurrent.

3. The photonic device of claim 2, wherein controlling the first transistor using the second numeric value to generate a photocurrent comprises applying, to a gate of the first transistor, a voltage that is encoded with the second numeric value.

4. The photonic device of claim 2, wherein the modulatable detector further comprises:

a second photodetector and a second transistor,

wherein the first and second photodetectors are configured to provide balanced detection, and the first and second transistors form a first inverter.

5. The photonic device of claim 4, wherein the modulatable detector further comprises:

a third transistor and a fourth transistor forming a second inverter,

wherein the first inverter and the second inverter are in parallel to each other.

6. The photonic device of claim 5, wherein the controller is further configured to control the second, third and fourth transistors using the second numeric value.

7. The photonic device of claim 2, wherein the modulatable detector further comprises:

a second photodetector, wherein the first and second photodetectors are configured to provide balanced detection; and

a node coupled to the first transistor and both the first and second photodetectors.

8. The photonic device of claim 7, wherein the modulatable detector further comprises:

a second transistor, wherein the node is further coupled to the second transistor.

9. The photonic device of claim 8, wherein the controller is further configured to control the second transistor using the second numeric value.

10. The photonic device of claim 2, wherein the modulatable detector further comprises:

a second transistor, wherein the first and second transistors form a first differential pair.

11. The photonic device of claim 10, wherein the modulatable detector further comprises:

a second photodetector; and

third and fourth transistors forming a second differential pair, wherein the first photodetector is coupled to the first differential pair and the second photodetector is coupled to the second differential pair.

12. The photonic device of claim 11, wherein the first and second differential pairs have cross-coupled connections.

13. The photonic device of claim 11, wherein the controller is further configured to control the second, third and fourth transistors using the second numeric value.

14. The photonic device of claim 2, wherein the modulatable detector further comprises:

a second photodetector;

second, third and fourth transistors, wherein the first and second transistors form a first transmission gate and the third and fourth transistors form a second transmission gate; and

a node coupled to both the first and second photodetectors and both the first and second transmission gates.

15. The photonic device of claim 14, wherein the controller is further configured to control the second, third and fourth transistors using the second numeric value.

16. The photonic device of claim 2, wherein obtaining the result of the mathematical operation comprises obtaining a product of the first numeric value times the second numeric value.

17. A photonic device configured to perform a mathematical operation comprising:

an optical encoder;

a modulatable detector coupled to an output of the optical encoder, the modulatable detector comprising a balanced detection unit and a differential modulation unit; and

a controller coupled to both the optical encoder and the modulatable detector, the controller being configured to: obtain a first numeric value and a second numeric value, control the optical encoder to generate an encoded optical signal by modifying an input optical signal using the first numeric value, control the differential modulation unit using the second numeric value to generate a pair of photocurrents in response to the balanced detection unit receiving the encoded optical signal, and obtain a result of the mathematical operation using the pair of photocurrents.

18. The photonic device of claim 17, wherein obtaining the result of the mathematical operation comprises obtaining a product of the first numeric value times the second numeric value.

19. The photonic device of claim 17, wherein controlling the differential modulation unit using the second numeric value to generate a pair of photocurrents comprises applying, to gates of the differential modulation unit, voltages encoded with the second numeric value.

20. The photonic device of claim 17, wherein:

the differential modulation unit comprises first and second transistors forming a first differential pair and third and fourth transistors forming a second differential pair, wherein the first and second differential pairs have cross-coupled connections.

21. The photonic device of claim 17, wherein:

the differential modulation unit comprises first and second transistors forming a first inverter and third and fourth transistors forming a second inverter, wherein the first inverter and the second inverter are in parallel to each other.

22. The photonic device of claim 17, wherein:

the differential modulation unit comprises first and second transistors and the balanced detection unit comprises first and second photodetectors, and

the modulatable detector further comprises a node coupled to both the first and second transistors and both the first and second photodetectors.

23. The photonic device of claim 17, wherein:

the differential modulation unit comprises first, second, third and fourth transistors, wherein the first and second transistors form a first transmission gate and the third and fourth transistors form a second transmission gate.