Optical Crossbar Compensation for Signed Matrix-Vector Multiplication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical and electrical crossbar arrays face challenges in representing negative vector or matrix values and dealing with light or electrical current leakages, which affect the accuracy of matrix-vector multiplications.
Innovation Solution
A crossbar array apparatus with a compensation device and readout device that generates a compensation output to correct for signal attenuations, allowing for accurate representation of negative entries and compensating for imperfections in signal propagation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If optical crossbar array uses attenuative components (PCMs) to implement MVM by attenuating signals, then the multiplication operation can be performed, but signal leakages and losses occur due to imperfections in the components
Solution Approach 1:
The patent implements a feedback mechanism where the actual output signals from the optical crossbar array are measured and compared against expected values. Compensation signals are then generated based on the difference (error) between actual and expected outputs, and these compensation signals are fed back to correct the multiplication results. This closed-loop feedback approach directly addresses the signal leakage and loss issues by continuously monitoring and correcting for component imperfections.
Solution Approach 2:
The patent performs preliminary characterization of the optical crossbar array components before actual MVM operations. The attenuation characteristics of each PCM component are measured and stored in advance. During operation, these pre-characterized parameters are used to generate compensation signals that account for known component imperfections, thereby proactively correcting for signal losses before they affect the final result accuracy.
2Speed
If optical signals are used to represent vector elements, then high-speed computation is achieved, but negative values cannot be directly represented since light intensity is always positive
Solution Approach 1:
The patent segments the representation of vector elements into two separate optical channels: one channel represents the magnitude (absolute value) of each element, while another channel represents the sign (positive or negative). This segmentation allows the optical system to encode signed numerical values using always-positive light intensities, thereby maintaining high-speed optical computation while gaining the ability to represent negative values through the combined interpretation of magnitude and sign channels.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The apparatus provides accurate matrix-vector multiplications by generating a compensated result that accounts for signal leakages and negative values, enhancing the efficiency and precision of vector-to-vector, matrix-to-vector, and matrix-to-matrix multiplications.
Implementation Method 1
A set of input waveguides is provided, with each input waveguide configured to propagate a different respective one of a set of input optical signals
Implementation Method 2
Each processing element is configured to receive a portion of one of the set of input optical signals and pass a respective controllable fraction of that portion to one of the set of output waveguides. The controllable fractions are based on values of elements of an input matrix to be multiplied with the input vector.
Implementation Method 3
A compensation device is provided. The compensation device is configured to generate a set of compensation optical signals
Implementation Method 4
A readout device is provided. The readout device is configured to: receive the set of output optical signals; receive the compensation optical signals; and provide a compensated result of a multiplication between the input vector and the input matrix
Data Source
AI summary
An apparatus and method for computing multiplication operations involving vectors, matrices or both, using a photonic computing architecture including an optical crossbar array. A compensation device is used to compensate for non-ideal characteristics of devices in the optical crossbar array. Apparatus and methods for representing negative-valued vectors of a multiplier in the multiplication operation are also provided.