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.

Abstract: Described herein are techniques for yield enhancement in photonic communications platforms. A photonic communication platform may include a photonic substrate patterned with a plurality of photonic modules including at least first and second photonic modules, wherein the first and second photonic modules are copies of a common template photonic module. Yield enhancement may be accomplished using photonic redundancy and/or electronic redundancy. Photonic redundancy may involve redundant optical lanes provided in parallel to primary optical lanes. Electronic redundancy may involve use of additional electronic circuits or wires running in parallel to electronic circuits or wires. Defective circuits may be disabled to prevent negative impacts on other parts of the electronic system. This can be done by providing power-isolating switches that completely disable and isolate the defective circuits.

Abstract: A method for manipulating an input vector is described. The method involves controlling a plurality of optical switches to obtain a nominal orientation vector or a transpose orientation vector based on a plurality of input optical signals encoding the input vector and received at the plurality of optical switches. The nominal orientation vector and the transpose orientation vector represent transposed versions of one another. A memory system comprising a first section configured to store vectors in accordance with a nominal orientation and a second section configured to store vectors in accordance with a transpose orientation. A controller stores the nominal orientation vector in the first section of the memory system or stores the transpose orientation vector in the second section of the memory system.

Abstract: Described herein are photonic sources and related system architectures that can satisfy the optical power requirements of large photonic accelerators. Some embodiments relate to a computer comprising a photonic accelerator configured to perform matrix multiplication; a fiber array optically coupled to the photonic accelerator; and a photonic source optically coupled to the fiber array. The photonic source comprising a laser array comprising a plurality of monolithically co-integrated lasers, and a coupling lens array comprising a plurality of monolithically co-integrated lenses, the coupling lens array optically coupling the laser array to the fiber array. The laser array is configured to output between 0.1 W and 10 W of optical power.

Abstract: Systems and methods for performing signed matrix operations using a linear photonic processor are provided. The linear photonic processor is formed as an array of first amplitude modulators and second amplitude modulators, the first amplitude modulators configured to encode elements of a vector into first optical signals and the second amplitude modulators configured to encode a product between the vector elements and matrix elements into second optical signals. An apparatus may be used to implement a signed value of an output of the linear processor. The linear photonic processor may be configured to perform matrix-vector and/or matrix-matrix operations.

Abstract: Electronic-photonic packages and related fabrication methods are described. A package may include a plurality of photonic integrated circuits (PICs), where each PIC comprises a photonic accelerator configured to perform matrix multiplication in the optical domain. The package may further include an application specific integrated circuit (ASIC) configured to control at least one of the photonic accelerators. The package further includes an interposer. The plurality of PICs are coupled to a first side of the interposer and the ASIC is coupled to a second side of the interposer opposite the first side. A first thermally conductive member in thermal contact with at least one of the PICs. The first thermally conductive member may include a heat spreader. A second thermally conductive member in thermal contact with the ASIC. The second thermally conductive member may include a lid.

Abstract: Methods and apparatus for training a matrix-based differentiable program using a photonics-based processor. The matrix-based differentiable program includes at least one matrix-valued variable associated with a matrix of values in a Euclidean vector space. The method comprises configuring components of the photonics-based processor to represent the matrix of values as an angular representation, processing, using the components of the photonics-based processor, training data to compute an error vector, determining in parallel, at least some gradients of parameters of the angular representation, wherein the determining is based on the error vector and a current input training vector, and updating the matrix of values by updating the angular representation based on the determined gradients.

Abstract: Aspects relate to a photonic processing system, an integrated circuit, and a method of operating an integrated circuit to control components to modulate optical signals. A photonic processing system, comprising: a photonic integrated circuit comprising: a first electrically-controllable photonic component electrically coupling an input pin to a first output pin; and a second electrically-controllable photonic component electrically coupling the input pin to a second output pin.

Abstract: Methods and apparatus for training a matrix-based differentiable program using a photonics-based processor. The matrix-based differentiable program includes at least one matrix-valued variable associated with a matrix of values in a Euclidean vector space. The method comprises configuring components of the photonics-based processor to represent the matrix of values as an angular representation, processing, using the components of the photonics-based processor, training data to compute an error vector, determining in parallel, at least some gradients of parameters of the angular representation, wherein the determining is based on the error vector and a current input training vector, and updating the matrix of values by updating the angular representation based on the determined gradients.

Abstract: Methods and apparatus for tuning a photonics-based component. An opto-electrical detector is configured to output an electrical signal based on a measurement of light intensity of the photonics-based component, the light intensity being proportional to an amount of detuning of the photonics-based component. Analog-to-digital conversion (ADC) circuitry is configured to output a digital signal based on the electrical signal output from the opto-electrical detector. Feedback control circuitry is configured to tune the photonics-based component based, at least in part, on the digital signal output from the ADC circuitry.

Abstract: Methods and systems for training neural networks using low-bitwidth accelerators are described. The methods described herein use moment-penalization functions. For example, a method comprises producing a modified data set by training a neural network using a moment-penalization function and the data set. The moment-penalization function is configured to penalize a moment associated with the neural network. Training the neural network in turn comprises quantizing the data set to obtain a fixed-point data set so that the fixed-point data set represents the data set in a fixed-point representation, and passing the fixed-point data set through an analog accelerator. The inventors have recognized that training a neural network using a modified objective function augments the accuracy and robustness of the neural network notwithstanding the use of low-bitwidth accelerators.

Abstract: Methods and apparatus for tuning a photonics-based component. An opto-electrical detector is configured to output an electrical signal based on a measurement of light intensity of the photonics-based component, the light intensity being proportional to an amount of detuning of the photonics-based component. Analog-to-digital conversion (ADC) circuitry is configured to output a digital signal based on the electrical signal output from the opto-electrical detector. Feedback control circuitry is configured to tune the photonics-based component based, at least in part, on the digital signal output from the ADC circuitry.

Abstract: Systems and methods for performing signed matrix operations using a linear photonic processor are provided. The linear photonic processor is formed as an array of first amplitude modulators and second amplitude modulators, the first amplitude modulators configured to encode elements of a vector into first optical signals and the second amplitude modulators configured to encode a product between the vector elements and matrix elements into second optical signals. An apparatus may be used to implement a signed value of an output of the linear processor. The linear photonic processor may be configured to perform matrix-vector and/or matrix-matrix operations.

Abstract: A photonic processor uses light signals and a residue number system (RNS) to perform calculations. The processor sums two or more values by shifting the phase of a light signal with phase shifters and reading out the summed phase with a coherent detector. Because phase winds back every 2? radians, the photonic processor performs addition modulo 2?. A photonic processor may use the summation of phases to perform dot products and correct erroneous residues. A photonic processor may use the RNS in combination with a positional number system (PNS) to extend the numerical range of the photonic processor, which may be used to accelerate homomorphic encryption (HE)-based deep learning.

Abstract: A memory device is described. The memory device comprises a plurality of stacked memory layers, wherein each of the plurality of stacked memory layers comprises a plurality of memory cells. The memory device further comprises an optical die bonded to the plurality of stacked memory layers and in electrical communication with the stacked memory layers through one or more interconnects. The optical die comprises an optical transceiver, and a memory controller configured to control read and/or write operations of the stacked memory layers. The optical die may be positioned at one end of the plurality of stacked memory layers. The one or more interconnects may comprise one or more through silicon vias (TSV). The plurality of memory cells may comprise a plurality of solid state memory cells. The memory devices described herein can enable all-to-all, point-to-multipoint and ring architectures for connecting logic units with memory devices.

Abstract: Described herein are techniques of training a machine learning model and performing inference using an analog processor. Some embodiments mitigate the loss in performance of a machine learning model resulting from a lower precision of an analog processor by using an adaptive block floating-point representation of numbers for the analog processor. Some embodiments mitigate the loss in performance of a machine learning model due to noise that is present when using an analog processor. The techniques involve training the machine learning model such that it is robust to noise.

Abstract: Aspects of the present disclosure provide an aligned storage strategy for stripes within a long vector for a vector processor, such that the extra computation needed to track strides between input stripes and output stripes may be eliminated. As a result, the stripe locations are located in a more predictable memory access pattern such that memory access bandwidth may be improved and the tendency for memory error may be reduced.

Abstract: Parallelization and pipelining techniques that can be applied to multi-core analog accelerators are described. The techniques descried herein improve performance of matrix multiplication (e.g., tensor-tensor multiplication, matrix-matrix multiplication or matrix-vector multiplication). The parallelization and pipelining techniques developed by the inventors and described herein focus on maintaining a high utilization of the processing cores. A representative processing systemin includes an analog accelerator, a digital processor, and a controller. The controller is configured to control the analog accelerator to output data using linear operations and to control the digital processor to perform non-linear operations based on the output data.

Abstract: Aspects of the present disclosure are directed to an efficient data transfer strategy in which data transfer is scheduled based on a prediction of the internal memory utilization due to computational workload throughout its runtime. According to one aspect, the DMA transfer may be performed opportunistically: whenever internal buffer memory is available and the additional internal memory usage due to DMA transfer isn't interfering with the processor's ability to complete the workload. In some embodiments, an opportunistic transfer schedule may be found by solving an optimization problem.

Abstract: Techniques for computing matrix operations for arbitrarily large matrices on a finite-sized hybrid analog-digital matrix processor are described. Techniques for gain adjustment in a finite-sized hybrid analog-digital matrix processor are described which enable the system to obtain higher energy efficiencies, greater physical density and improved numerical accuracy. In some embodiments, these techniques enable maximization of the predictive accuracy of a GEMM-based convolutional neural network using low-precision data representations.