Information Processing and Sensing with Photonic Neurons
Abstract
The ultimate goal of this project is to develop a new solution for energy-efficient information processing. In these days when artificial intelligence (AI) systems consume enormous amounts of energy and computing power is reaching the limits of digital electronics, new disruptive approaches are urgently needed. Neurons can, with extremely low power consumption, perform tasks that go far beyond the state of the art of AI systems. Efficient neural coding mechanisms are responsible for this amazing performance. Photonic neurons are optical systems with ultra-fast neuron-like spiking outputs. This project aims to implement in photonic neurons the mechanisms by which neurons encode information. We propose the development of laser systems and algorithms that accurately mimic neural information processing. At the hardware level, using commercial diode lasers (that are highly efficient, fast and low cost) we will build photonic neurons capable of using neural coding mechanisms for computation and sensing. To develop specialized photonic neurons that selectively respond to external inputs, we will conduct experiments and simulations to find conditions in which diode lasers respond to weak inputs, as neurons do. At the software level, we will develop new data analysis tools to decipher the information encoded in the emitted sequences of optical spikes. Our overall goal is a proof-of-concept demonstration of information processing based on neural coding, performed with diode lasers. Our solution for developing photonic neurons will have large impacts, because using neural coding mechanisms can lead to a new generation of neuromorphic photonic computing systems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA86552417022
Entities
People
- Cristina Masoller
Organizations
- Air Force Office of Scientific Research
- Polytechnic University of Catalonia
- United States Air Force