Optoelectronic Computing Systems with Optical Matrix Multiplication

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current neuromorphic computing systems face limitations in processing speed and efficiency due to reliance on electronic matrix multiplication units, which are slower compared to optical processing capabilities, especially when handling large datasets and multiple neural network computations.

Innovation Solution

A system that integrates optical and electronic components to perform artificial neural network computations, utilizing a digital-to-analog converter and optical matrix multiplication units to process data optically, with a controller managing operations to enhance throughput and reduce memory access latency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If electronic matrix multiplication units are used for neural network computations, then the system can perform computations using existing electronic infrastructure, but the processing speed is limited compared to optical processing capabilities

Engineering Contradiction:
Improveprocessing speedVSAvoidsystem complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent replaces electronic matrix multiplication operations with optical processing operations. Optical signals are used to perform matrix multiplication in the optical domain, leveraging the inherent parallelism and speed of light propagation. This substitution of electronic computation with optical computation directly addresses the speed limitation while maintaining computational functionality through the use of optical modulators, waveguides, and photodetectors arranged in a matrix multiplication architecture.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If optical processing is used for matrix multiplication, then processing speed increases, but the system requires integration of optical and electronic components increasing complexity

Engineering Contradiction:
ImprovethroughputVSAvoidcomponent integration complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent divides the neural network computation system into distinct functional modules: optical input signal generation, optical matrix multiplication processing, and electronic output signal detection. Each module is optimized for its specific function, with the optical section handling high-speed matrix multiplication and the electronic section handling input/output interfacing. This segmentation allows the system to achieve high throughput through optical processing while managing complexity through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a hybrid optoelectronic system where a single integrated platform performs both optical processing functions (matrix multiplication) and electronic processing functions (input/output signal generation and detection). This multi-functional integration allows the system to leverage the strengths of both optical and electronic domains within a unified architecture, achieving high throughput while consolidating control logic and reducing the need for separate dedicated systems.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Loss of time

If electronic components process all neural network computations, then the system architecture is simpler, but memory access latency increases for large datasets

Engineering Contradiction:
Improvememory access latencyVSAvoidsystem operation simplicity
Core Design Contradiction:
Loss of timeVSEase of operation

Solution Approach 1:

The patent extracts the computationally intensive matrix multiplication operations from the electronic domain and relocates them to the optical domain. By taking out these specific operations and performing them optically, the system eliminates the memory access bottleneck inherent in electronic processing of large datasets. The optical processing occurs in parallel without sequential memory access, dramatically reducing latency for large-scale neural network computations while the electronic components handle only the necessary input/output interfacing.

Inventive Principle:
Principle #2Taking out (Extraction)

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 system achieves faster processing speeds and improved throughput by leveraging optical matrix multiplication, reducing the burden on electronic components and enabling self-contained neural network computations independent of external processors.

Implementation Method 1

utilizing a digital-to-analog converter and optical matrix multiplication units to process data optically

Methodology Applied
Scientific EffectOptical signal processing:

Implementation Method 2

utilizing a digital-to-analog converter and optical matrix multiplication units to process data optically

Methodology Applied
Scientific EffectElectro-optic conversion: Electro-Optic Effects

Data Source

PatentEP3912096B1Optoelectronic computing systems
Publication Date: 2024.12.18 LIGHTELLIGENCE PTE LTD
  • EP3912096B1 patent drawingFigure 1A
  • EP3912096B1 patent drawingFigure 1B~1E
  • EP3912096B1 patent drawingFigure 1F

AI summary

A system includes a first unit for generating modulator control signals, and a processor that includes a light source for providing light outputs and a first set of optical modulators for generating an optical input vector by modulating light from the light source based on digital input values. The processor includes a matrix multiplication unit having a second set of optical modulators. The matrix multiplication unit transforms the optical input vector into an analog output vector based on digital weight values corresponding to a second set of modulator control signals. At least one optical modulator is configured to modulate an optical signal based on a first modulator control signal, and the first unit is configured to shape the first modulator control signal to include bandwidth-enhancement associated with a change in amplitude associated with a corresponding change in successive digital values corresponding to the first modulator control signal.