Optical Neural Network System Using Nested Light Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical 4f systems for convolutional artificial neural networks are limited by their large volume and energy inefficiency, making them unsuitable for applications like autonomous driving and small IoT devices, where miniaturization and improved energy efficiency are necessary.
Innovation Solution
The optical artificial neural network system incorporates a light insertion unit, spatial light modulators, light path adjustment devices, and a Fourier transform device to process light in a miniaturized 4f configuration, optimizing light paths and using components like polarized beam splitters, quarter wave-plates, and concave mirrors to enhance computation speed and energy efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If an optical 4f system is used for convolutional artificial neural networks, then computation speed and energy efficiency are improved, but the system volume becomes large
Solution Approach 1:
The patent implements a nested optical path configuration where the return light path is embedded within the forward light path. The second light path adjustment device is positioned to receive reflected light and guide it back through the same optical components (spatial light modulators, Fourier transform device) in reverse sequence, effectively nesting the return path within the forward path structure. This reduces the overall system volume by eliminating the need for separate return path components and adjacently positioned optics.
Solution Approach 2:
The patent utilizes the optical reflection dimension to fold the light path back through the system. By employing mirrors and beam splitters to redirect light at different angles and planes, the optical path is folded multiple times within a compact footprint. The light travels forward through the system, reflects off a mirror, and returns through the same components in reverse, effectively using spatial dimensionality to reduce the linear extent of the system while maintaining the required optical path length.
2Productivity
If an optical 4f system is used for convolutional artificial neural networks, then computation speed is improved, but energy efficiency deteriorates
Solution Approach 1:
The patent implements a continuous optical computation process where light travels through the system in a single uninterrupted pass. The incident light is modulated by the spatial light modulators, transformed by the Fourier transform device, and reflected back through the same modulators without interruption or conversion to other forms of energy. This continuous optical action maintains high energy efficiency by avoiding energy-lossy conversions between optical and electrical domains, while the rapid speed of light propagation ensures high computation speed.
Solution Approach 2:
The patent replaces electronic computation mechanisms with optical mechanisms. Instead of using electronic processors that consume significant energy for matrix multiplications and convolutions, the system uses optical interference and diffraction phenomena to perform the same computational functions. The spatial light modulators modulate light intensity and phase directly, and the Fourier transform device performs frequency domain transformations optically, eliminating the need for energy-intensive electronic computation while maintaining high-speed processing.
3Volume of stationary object
If existing electronic computers are used to process image information, then system volume is reduced, but computation speed becomes slow
Solution Approach 1:
The patent substitutes electronic computation with optical computation. The spatial light modulators use liquid crystal or micro-mechanical elements to modulate light directly in the optical domain, and the Fourier transform device uses optical diffraction and interference to perform computations at the speed of light. This optical mechanism maintains a compact form factor while achieving computation speeds orders of magnitude faster than electronic processors, as optical operations occur simultaneously across the entire light field rather than sequentially as in electronic computation.
4Measurement precision
If the light path is extended to include multiple modulation stages, then computation accuracy is improved, but system complexity increases
Solution Approach 1:
The patent merges the forward and return light paths to traverse the same optical components in sequence. The incident light passes through the first spatial light modulator, then the Fourier transform device, then reflects off a mirror and passes back through the Fourier transform device and the first spatial light modulator in reverse, before reaching the second spatial light modulator. This merging of paths allows multiple modulation stages to be implemented within a compact configuration, achieving high computation accuracy through cumulative optical transformations while avoiding the complexity of separate forward and return path components.
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
This configuration achieves computation speed and energy efficiency comparable to electronic computers while reducing the system's volume, enabling its application in autonomous driving and small IoT devices with improved mobility and convenience.
Implementation Method 1
the polarized beam splitter reflects a first direction component of the incident light toward the first spatial light modulator
Implementation Method 2
the quarter wave-plate circularly polarizes the first direction component of the first light reflected by the polarized beam splitter
Implementation Method 3
the first light path adjustment device includes a Faraday rotator configured to rotate a polarization direction of a transmitted light by 45 degrees
Implementation Method 4
a light Fourier transform device that reflects the first light circularly polarized, to generate a second light having a second optical image
Data Source
AI summary
Disclosed is an optical artificial neural network system which includes a first spatial light modulator modulating an incident light to generate a first light having a first optical image, a concave mirror reflecting the first light to generate a second light having a second optical image, a first polarized beam splitter disposed between the first spatial light modulator and the concave mirror, a first quarter wave-plate disposed between the first spatial light modulator and the first polarized beam splitter, a second spatial light modulator generating a third light by modulating the second light reflected by the first polarized beam splitter so as to have a third optical image, a beam splitter disposed between the first spatial light modulator and the first polarized beam splitter, and an imaging device.