Photonic Crossbar Synapse for Low-Energy Neuromorphic Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine vision systems based on the von Neumann architecture face issues such as data redundancy, access delays, high energy consumption, and high fabrication costs due to the physical separation of components, which are exacerbated in autonomous vehicles with limited computing power.
Innovation Solution
A neuromorphic machine vision system using an optical synapse device with a transparent conductive film electrode, a double oxide active layer, and an electrically conductive layer, capable of generating currents in response to light and voltage, arranged in a crossbar configuration for convolution operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional von Neumann architecture is used for machine vision systems, then separate hardware components can be implemented, but data access delay and energy consumption increase due to frequent data shuttling between components
Solution Approach 1:
The patent merges sensing, memory, and processing functions into a single integrated photonic crossbar device. The photoreceptive array directly generates photonic signals that are processed through the crossbar architecture without being shuttled to separate processing units, eliminating data access delays while maintaining system reliability through functional integration.
Solution Approach 2:
The patent replaces electrical data shuttling with photonic signal processing. By using photoreceptive cells that convert light directly into photonic signals for processing, the system eliminates the need for electrical-to-optical conversion and data transmission between separate components, reducing both time loss and energy consumption.
2Adaptability or versatility
If multiple separate hardware components are used in traditional machine vision systems, then distinct functions can be performed, but fabrication costs and device complexity increase
Solution Approach 1:
The photonic crossbar device performs multiple functions within a single architecture: photoreceptive cells detect light, the crossbar structure stores and processes data, and convolution operations are executed optically. This multi-functional integration reduces device complexity while maintaining adaptability for various machine vision tasks through reconfigurable crossbar connections.
Solution Approach 2:
The patent combines sensing, memory, and processing functions into a unified photonic crossbar device. The photoreceptive array, crossbar switch fabric, and processing logic are integrated into a single device structure, reducing fabrication costs and complexity while preserving the functional capabilities needed for machine vision operations.
3Power
If data is shuttled wirelessly to central processing system in autonomous vehicles, then onboard computing power requirements are reduced, but energy consumption from battery supply increases
Solution Approach 1:
The patent replaces wireless data transmission with direct photonic signal processing within the vehicle. By performing convolution operations optically in the crossbar device, the system eliminates the need for wireless communication and centralized processing, reducing energy consumption from the battery while maintaining adequate computing power for autonomous vehicle operations.
Solution Approach 2:
The photonic crossbar device performs processing locally at the sensing location without requiring external computational resources. The device processes visual data autonomously through its integrated architecture, reducing the energy burden on the vehicle's battery system while providing sufficient computing power for real-time decision-making.
4Productivity
If larger crossbar arrays are used to enrich convolution operations with different kernels, then processing capability increases, but fabrication costs and integration density decrease
Solution Approach 1:
The patent implements dynamic reconfiguration of the crossbar array to perform different convolution kernels. By programmatically controlling the crossbar connections and weights, the same physical array can be reconfigured to execute various processing tasks, increasing productivity without requiring larger arrays or additional fabrication resources.
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
Enables efficient broadband sensing and fast data processing with low energy consumption, supporting front-end retinomorphic image sensing, convolutional processing, and back-end neuromorphic computing, reducing fabrication costs and integration density.
Implementation Method 1
A neuromorphic machine vision system uses an optical synapse device with a transparent conductive film electrode, a double oxide active layer, and an electrically conductive layer, capable of generating currents in response to light
Implementation Method 2
a double oxide active layer having resistive switching properties... configured to generate a current in response to the voltage and light
Data Source
AI summary
An optical synapse device which can be used in a neuromorphic machine vision system to enable broadband sensing and fast data processing in applications where machine vision is used, such as for example in real-time video analysis, autonomous vehicles and medical diagnosis. The novel optical synapse device enables image sensing, convolutional processing, and computing. Multiple synaptic plasticity triggered by photons can implement photonic computing and information transmission. Convolutional processing is realized by ultra-low energy kernel generators fully controlled by photons. The optical synapse device shows the ability of conductance modulations under electronic stimulations that implement neuromorphic computing. The novel two-terminal broadband optoelectronic synapse device enables front-end retinomorphic image sensing, convolutional processing, and back-end neuromorphic computing.