Diffractive Photonic Computing Chiplets for Scalable AI Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current photonic integrated circuits (PICs) are limited in scale and computational power, struggling to support large-scale neural network computing due to physical constraints, device errors, and high redundancy, making them inadequate for advanced artificial general intelligence (AGI) tasks.
Innovation Solution
A large-scale distributed photoelectric intelligent computing system utilizing reconfigurable chiplets with integrated photonics, comprising a diffractive compression encoder, reconfigurable interference feature embedder, and diffractive decoder, which employs diffractive coding and reconfigurable interference operations to process high-dimensional inputs efficiently and reduce redundancy, enabling larger-scale neural network computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If traditional electronic devices (GPU) are used for AI big model computing, then computational capability can be achieved, but energy overhead becomes huge and inference speed is limited
Solution Approach 1:
The patent replaces traditional electronic computing systems with a photoelectric integrated computing system that combines photonic circuits for high-speed data transmission and electronic circuits for processing. This substitution leverages the advantages of both optical (high bandwidth, low energy) and electronic (precise control, computation) systems to achieve low energy consumption and high inference speed simultaneously.
Solution Approach 2:
The computing system is divided into distinct photonic and electronic functional modules. The photonic circuit handles data encoding and transmission, while the electronic circuit performs computation operations. This segmentation allows each subsystem to operate in its optimal performance regime, resolving the contradiction between energy efficiency and computational productivity.
2Use of energy by moving object
If photonic integrated circuits are used for computing, then energy efficiency is improved, but computational scale is limited due to physical constraints and device errors
Solution Approach 1:
The patent merges photonic and electronic circuits into a unified photoelectric integrated system. The photonic circuit provides energy-efficient data handling, while the electronic circuit extends computational scale and precision. This merging allows the system to overcome the physical constraints of pure photonic systems while maintaining energy efficiency.
Solution Approach 2:
The electronic circuit acts as an intermediary that bridges the gap between the photonic circuit's energy efficiency and the requirements for large-scale computation. It converts optical signals to electrical signals for processing, enabling scalable computation while preserving the energy advantages of photonic transmission.
3Loss of energy
If photonic circuits are used for neural network computing, then energy consumption is reduced, but device errors and redundancy increase making advanced AGI tasks difficult
Solution Approach 1:
The patent implements feedback mechanisms where the electronic circuit monitors and corrects errors from the photonic circuit. The system uses the electronic subsystem to detect and compensate for photonic device errors, maintaining computational accuracy while preserving the low power consumption advantages of the photonic approach.
Solution Approach 2:
The system incorporates error correction capabilities in advance through the electronic circuit subsystem. By preparing error detection and correction mechanisms beforehand, the system cushions against the inherent device errors in photonic circuits, ensuring reliable computation for advanced AGI tasks while maintaining energy efficiency.
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 high energy efficiency and computational capability, supporting complex AI tasks with reduced redundancy and error accumulation, achieving accuracy comparable to electrical computing while significantly lowering power consumption.
Implementation Method 1
a parallel diffractive compression encoder and a diffractive decoder are configured to collect two-dimensional light field information, convert the two-dimensional light field information into one-dimensional information through waveguide transmission, and perform computing by using a diffractive coding weight
Implementation Method 2
collect two-dimensional light field information, convert the two-dimensional light field information into one-dimensional information through waveguide transmission
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A large-scale distributed photoelectric intelligent computing system, including: a diffractive compression encoder , configured to collect two-dimensional light field information through a grating array, convert the two-dimensional light field information into one-dimensional information through waveguide transmission and transmit the one-dimensional information to a first diffractive region, and perform computing using a diffractive coding weight to output a one-dimensional vector through an output end of the first diffractive region; a reconfigurable interference feature embedder consisting of an interferometer array composed of a plurality of thermo-optical phase modulators, in which the interferometer array is configured to perform a multiplication operation on the one-dimensional vector, to output a computed result in a one-dimensional vector form; and a diffractive decoder, configured for diffractive decoding the computed result in the one-dimensional vector form through a second diffractive region, to compute and output final light field information through a diffractive decoding weight.