Accelerate AI Processing with Photonic Speed and Precision

Introduction

In the race to build more powerful and efficient artificial intelligence systems, the processing speed of convolutional neural networks (CNNs) remains a critical factor. As AI applications grow in complexity, from autonomous vehicles to real-time data analytics, the need for fast, energy-efficient processing becomes even more urgent. Enter our optical convolutional neural network accelerator—an innovation that merges the power of photonics with advanced AI processing.

The Problem

Traditional electronic-based CNN accelerators, while effective, are approaching their physical and energy limits. They require vast amounts of power and often struggle to meet the demands of real-time AI applications that need both speed and accuracy. Data centers and autonomous systems are constantly challenged to balance the growing demand for computational power with the increasing costs and environmental impact of energy consumption.

The Solution

Our optical CNN accelerator takes a groundbreaking approach by leveraging light to perform data processing tasks typically reserved for electronics. This photonic-based system significantly reduces power consumption while dramatically increasing processing speed. Optical components can transmit data at the speed of light, allowing AI models to run faster, more efficiently, and with far lower energy requirements than traditional electronic accelerators.

Key Benefits

Unmatched Speed: By using optical pathways for data transmission and processing, this accelerator can handle massive AI workloads in a fraction of the time it takes traditional systems. For industries relying on fast decision-making, such as autonomous driving or real-time analytics, this speed boost is game-changing.
Energy Efficiency: Optical systems consume far less power than electronic counterparts, which not only lowers operational costs but also significantly reduces the carbon footprint of data centers and large-scale AI systems.
Scalability: This technology is scalable, making it ideal for both small, AI-powered devices and large data centers. The optical components can be integrated into existing AI infrastructure, enhancing performance without requiring a complete overhaul.
Versatile Applications: While primarily targeted for CNNs, this accelerator can be adapted to a wide range of AI applications, from image recognition and natural language processing to real-time video analytics and beyond.

Why License This Technology?

Licensing this optical CNN accelerator gives your company a competitive edge in AI processing. It offers a unique combination of speed, energy efficiency, and scalability that traditional electronic accelerators simply cannot match. As AI applications continue to expand, this technology positions your business to meet the growing demands of the future, while significantly lowering costs and energy consumption.

Conclusion

Optical technology is the future of AI acceleration. By licensing this innovation, you’ll be able to offer cutting-edge solutions that meet the speed and efficiency requirements of modern AI, setting your company apart in a rapidly evolving landscape.

Please see terms and conditions

Abstract
Claims

An accelerator for modern convolutional neural networks applies the Winograd filtering algorithm in a wavelength division multiplexing integrated photonics circuit modulated by a memristor-based analog memory unit.

The invention claimed is:

1. A neural network, the neural network including a system on chip, wherein the system on chip comprises:

a coherent light source configured to receive one or more analog signals and output one or more light signals based, at least in part, on the one or more analog signals;

an analog memory configured to generate one or more voltage signals;

a photonic element-wise matrix multiplication circuit coupled to the coherent light source and the analog memory, wherein the photonic element-wise matrix multiplication circuit configured to:

receive the one or more voltage signals from the analog memory and modulate the one or more light signals responsive to the one or more voltage signals; and

generate an analog electrical signal based, at least in part, on the modulated light signals,

wherein the one or more voltage signals correspond to one or more filter signals.

2. The neural network of claim 1, further comprising a photonic inverse-Winograd transform circuit, wherein:

the photonic element-wise matrix multiplication circuit is further configured to transform the one or more light signals into a Winograd domain;

the one or more filter signals are analog Winograd domain filter signals; and

the photonic inverse-Winograd transform circuit is configured to remove the one or more light signals from the Winograd domain.

3. The neural network of claim 2, wherein the system on chip further comprises:

a filter buffer configured to receive a plurality of filters;

a Winograd transform circuit configured to transform the plurality of filters into digital-Winograd domain filter signals; and

a digital-to-analog converter coupled to the analog memory and configured to convert the digital-Winograd domain filter signals into the analog Winograd domain filter signals,

wherein the analog memory is configured to detect the analog Winograd domain filter signals and generate the voltage signals in response thereto.

4. The neural network of claim 3, wherein the analog memory comprises a memristor.

5. The neural network of claim 3, wherein a timing of the digital-to-analog converter is controlled by a clock.

6. A method for accelerating computations for a convolutional neural network, comprising:

receiving, at a coherent light source, one or more analog signals;

outputting one or more light signals based, at least in part, on the one or more analog signals;

generating, using an analog memory, one or more voltage signals;

receiving, at a photonic element-wise matrix multiplication circuit, the one or more voltage signals from the analog memory;

modulating, using the photonic element-wise matrix multiplication circuit, the one or more light signals responsive to the one or more voltage signals, wherein the one or more light signals corresponding to one or more wavelength-values;

integrally summing, using the photonic element-wise matrix multiplication circuit, the one or more wavelength-values of the one or more light signals; and

generating, using the photonic element-wise matrix multiplication circuit, an analog electrical signal corresponding to the one or more light signals so integrated; and

wherein the one or more voltage signals correspond to one or more analog Winograd domain filter signals.

7. The method for accelerating computations for a convolutional neural network of claim 6, further comprising:

transforming, using the photonic element-wise matrix multiplication circuit, the one or more light signals into a Winograd domain; and

removing, using a photonic inverse-Winograd transform circuit, the one or more light signals from the Winograd domain.

8. A photonic accelerator for a convolutional neural network, comprising:

a coherent light source configured to receive one or more analog signals and output one or more light signals based, at least in part, on the one or more analog signals;

an analog memory configured to generate one or more voltage signals;

a photonic element-wise matrix multiplication circuit comprising a plurality of microring resonators and a photosensitive balanced detector summation circuit and being coupled to the coherent light source and the analog memory, wherein the photonic element-wise matrix multiplication circuit configured to:

receive the one or more voltage signals from the analog memory;

modulate the one or more light signals responsive to the one or more voltage signals, the modulated light signals having one or more wavelengths having one or more wavelength-values;

perform integration on the one or more wavelength-values of the modulated light signals; and

generate an analog electrical signal corresponding to the integrated one or more wavelength-values of the modulated light signals;

wherein the one or more voltage signals correspond to one or more filter signals.

9. The photonic accelerator of claim 8, further comprising a photonic inverse-Winograd transform circuit, wherein:

the photonic element-wise matrix multiplication circuit is further configured to transform the one or more light signals into a Winograd domain;

the one or more filter signals are analog Winograd domain filter signals; and

the photonic inverse-Winograd transform circuit is configured to remove the one or more light signals from the Winograd domain.

10. The photonic accelerator of claim 9, wherein the analog memory comprises a memristor.

11. The photonic accelerator of claim 10, wherein the photonic element-wise matrix multiplication circuit is further configured to:

receive the one or more voltage signals from the analog memory and modulate the one or more light signals responsive to the one or more voltage signals using the plurality of microring resonators;

and integrate the one or more light signals so modulated using the photosensitive balanced detector summation circuit.

Title

Optical convolutional neural network accelerator

Inventor(s)

Armin Mehrabian, Volker J. Sorger, Tarek El-Ghazawi, Mario Miscuglio

Assignee(s)

George Washington University

Patent #

11704550

Patent Date

July 18, 2023