Photonic Circuit Temporal Multiplexing for Large-Scale MAC Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Photonic circuits for optical computing face limitations in performing multiply-and-accumulate (MAC) operations due to the concurrent need for a large number of modulator and source pixels, which restricts the size of the input vector K, thereby limiting accuracy and power efficiency.
Innovation Solution
Employing temporal multiplexing to perform MAC operations, where an input row vector is temporally encoded via light-beam intensity and column vectors of the weight matrix are encoded via modulator-pixel transmissivity, decoupling the number of source and modulator pixels from K, allowing K to be effectively unlimited.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of modulator and source pixels are used concurrently to perform MAC operations, then computational accuracy is improved, but device complexity and power consumption increase
Solution Approach 1:
The patent applies temporal multiplexing where modulator pixels and source pixels are activated in periodic sequences rather than simultaneously. Each pixel is reused across multiple time slots to perform different multiplication operations, reducing the total number of pixels needed while maintaining computational accuracy through sequential processing of the input vector elements.
Solution Approach 2:
The system transitions from a static concurrent architecture to a dynamic temporal multiplexed architecture. The modulator pixels and source pixels are dynamically switched between different operational states across time slots, allowing the same physical hardware to perform multiple computational functions that would otherwise require separate static components.
2Measurement precision
If a large number of modulator and source pixels are used concurrently, then computational accuracy is improved, but power consumption increases
Solution Approach 1:
By using periodic temporal multiplexing, the system activates modulator and source pixels only when needed for current computational operations rather than keeping all pixels continuously active. This periodic activation significantly reduces overall power consumption while maintaining the ability to perform high-accuracy MAC operations through the time-multiplexed sequence of operations.
3Device complexity
If the number of source and modulator pixels is reduced, then device complexity is decreased, but the size of input vector K is limited
Solution Approach 1:
The patent resolves the contradiction by adding a temporal dimension to the computation. Instead of increasing the spatial number of pixels to handle larger input vectors, the system uses time slots to sequentially process each element of the input vector. This dimensional transition from space to time allows the system to handle arbitrarily large K values with a fixed, reduced number of physical pixels.
Solution Approach 2:
The system dynamically reconfigures which pixels are active and what operations they perform based on the current time slot and input vector index. This dynamic reconfiguration enables the same small set of pixels to adaptively handle different input vector sizes and computational requirements without requiring proportional increases in hardware complexity.
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 approach enables high accuracy modeling and high power efficiency by allowing large values of K, reducing power consumption to low levels while maintaining accuracy, suitable for deep neural network algorithms and other AI/ML operations.
Implementation Method 1
the source pixel is controlled to modulate an intensity of the light beam to temporally encode the input row vector
Implementation Method 2
the modulator pixels are controlled to modulate individual transmissivities of the modulator pixels to temporally encode column vectors of the weight matrix
Implementation Method 3
charge is accumulated at the detector pixels respectively in response to the transmitted light beams
Data Source
AI summary
Various embodiments of the present disclosure are directed to a photonic circuit for a vector-matrix operation. A source pixel is configured to generate a light beam. An optical fan-out structure is configured to generate a plurality of copies of the light beam. A plurality of modulator pixels are configured to respectively transmit the plurality of copies with individual transmissivities to generate a plurality of transmitted light beams. A plurality of detector pixels are configured to accumulate charge respectively in response to the plurality of transmitted light beams. A controller is configured to control the source pixel and the plurality of modulator pixels to modulate an intensity of the light beam and the individual transmissivities to perform the vector-matrix multiplication operation. The intensity is modulated to temporally encode an input row vector, and the individual transmissivities are modulated to temporally encode corresponding column vectors of a weight matrix.