Development of trainable all-optical spiking neural networks in integrated photonics and their applications

01 October 2014 → 30 September 2017
Regional and community funding: Special Research Fund
Research disciplines
  • Engineering and technology
    • Other engineering and technology
neural networks fotonica
Project description

The aim of this project is to build a small-scale, miniaturized ‘brain’ on a silicon chip. Although fabricated by the same infrastructure as digital chips, it will handle optical instead of electronic signals. Similar to its biological counterpart, it will consist of ‘neurons’, sending out pulses to each other. Information in such networks is represented in the pulse-timing.
I will show that this system is able to solve non-trivial tasks in which, e.g., many time-varying input signals have to be processed simultaneously. Similarly to our brain, which faces an equivalent computational challenge every day, it will do this in an energy-efficient way. By processing optical signals instead of electrical ones, it will also be faster than electronic solutions. Indeed, as illustrated by the recent development of optical interconnections to increase the information transfer in future generation digital chips, the transfer of large amounts of information can happen at higher speeds and using less power. Additionally, optical phenomena have much richer intrinsic dynamics, making it easier to mimic neural behaviour.
During my PhD research, I have studied a class of tiny optical components that indeed behave phenomenologically identical to biological neurons. I will now integrate them into a larger network, and incorporate an electronic layer on top of the optical layer that will allow the system to learn from examples.