Authors:

Ruben Canora Alvarez

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

The intrinsic wave properties of light provide a powerful mechanism for achieving unparalleled data transmission rates at minimal energy cost. Consequently, in recent years the idea of developing All-Optical Neural Networks (AONNs) has raised as the solution to the computational limitations of the current Neural Network architectures. Despite their potential, the development of deep AONNs is limited by the high optical power demands of conventional nonlinear optical processes, which limits scalability. In our work, we experimentally tested our novel nonlinear optical activation function scheme based on quantum interference, consisting of an atomic ensemble driven by two laser fields. In addition, the nonlinear responses we achieved are formally equivalent to the two fundamental activation functions in digital neural networks: Rectified Linear Unit (ReLU) and Sigmoid functions. Not only we proved our fully tunable multi-input, multi-output network scheme, but also demonstrated the feasibility of constructing large-scale, deep AONNs with millions of neurons powered by less than 100W of optical power and at ultra-fast speeds. Furthermore, with the aid of Spatial Light Modulators (SLMs), we performed all the linear computations at the speed of light, and combined with the nonlinearity, we brought into reality a fully functional one-layer AONN with a high level of accuracy for image classification. By demonstrating minimal propagation losses and high scalability, our results inaugurate a new era toward high-speed, scalable and energy- efficient optical neural networks for next-generation AI hardware. Keywords: Neural Networks; Optics; Ultra-Fast Computations; Innovation; AI

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Ruben Canora Alvarez