Optical Neural Network Circuit Using Transparency-Modulated Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional circuits for neural network computing face challenges with signal delay due to RC delay and reduced energy efficiency from resistors, and require reliable and durable components for frequent updates during training.
Innovation Solution
An optical circuit system utilizing laser projecting devices, optical device groups, and power supply devices to generate and adjust optical signals based on transparency parameters, reducing signal delay and improving reliability without significantly affecting energy efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional circuits with conductive lines are used for neural network computing, then the circuit can be implemented, but signal delay occurs due to RC delay
Solution Approach 1:
The patent replaces electrical signal transmission through conductive lines with optical signal transmission through optical fibers or free space. This substitution eliminates the RC delay inherent in electrical circuits because optical signals do not suffer from the same resistive and capacitive effects. The laser projects optical signals that travel through optical waveguides or free space to reach the optical devices, fundamentally changing the transmission medium from electrical to optical domain.
Solution Approach 2:
The patent introduces optical devices (such as electro-optic modulators or transparent electrodes) as intermediaries to convert electrical control signals into optical signals for transmission. These optical devices act as mediators that receive electrical control inputs and generate corresponding optical signals with specific light intensities, thereby enabling reliable and fast signal transmission without direct electrical connections between computing nodes.
2Ease of manufacture
If resistors are added to implement neural networks in traditional circuits, then the neural network functionality is achieved, but energy efficiency is reduced
Solution Approach 1:
The patent replaces resistor-based electrical circuit implementation with optical device-based implementation. Instead of using resistors to implement weight multiplication and accumulation, the patent uses optical devices whose light intensity output is controlled by transparency parameters. This substitution eliminates the continuous power consumption associated with resistive heating while maintaining the computational functionality through optical intensity modulation.
Solution Approach 2:
The patent changes the fundamental operating parameter from electrical resistance to optical transparency. By controlling the transparency parameter of optical devices through voltage control, the system achieves weight multiplication without the energy loss inherent in resistive circuits. The transparency parameter can be dynamically adjusted to represent different weight values in the neural network.
3Productivity
If circuit components operate for long periods to support frequent node updates during training, then training can proceed, but reliability and durability become bottlenecks
Solution Approach 1:
The patent replaces traditional electrical circuit components with optical devices that have superior durability and reliability. Optical devices such as electro-optic modulators and optical waveguides do not suffer from the same degradation mechanisms as electrical components under continuous operation. They can withstand frequent switching and updates during neural network training without significant performance degradation, enabling long-term reliable operation.
4Reliability
If optical signals are used instead of conductive lines, then signal delay is reduced and reliability is improved, but the system complexity increases
Solution Approach 1:
The patent makes the optical devices serve multiple functions: they act as signal receivers, signal modulators, and computational elements simultaneously. Each optical device receives optical signals, modulates them according to its transparency parameter (representing weight), and outputs the modulated signal. This multi-functionality reduces the need for separate components for each function, thereby managing system complexity while achieving the benefits of optical transmission.
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 optical circuit system enhances reliability and reduces signal delay while maintaining energy efficiency, simplifying circuit routing and design complexity by using optical signals instead of conductive lines.
Implementation Method 1
a laser projecting device, a first optical device group and a second optical device group. The laser projecting device is configured to generate a plurality of standard optical signals
Implementation Method 2
Each of the plurality of first optical devices has a transparency parameter and is configured to generate a plurality of first optical signals based on the plurality of standard optical signals and the transparency parameter
Data Source
AI summary
An optical circuit system comprising a laser projecting device, a first optical device group and a second optical device group and configured to perform a neural network computing is provided. The laser projecting device generates multiple standard optical signals. The first optical device group comprises multiple first optical devices. Each first optical device has a transparency parameter and generates multiple first optical signals based on the received standard optical signals and the transparency parameter. The second optical device group comprises multiple second optical devices. Each second optical device has multiple transparency parameters and generates multiple second optical signals based on the received first optical signals and the transparency parameters, thereby generating a combined optical signal. The light intensity of the first/second optical signals generated by one of the first/second optical devices is related to one of multiple neuronal data of a first/second level of a neural network.