September 2020
Spotlight Summary by Daniel Brunner
Reconfigurable all-optical nonlinear activation functions for neuromorphic photonics
Is a neural network really a neural network if there are no neurons? And for that matter, when we talk about neural networks in photonics, what specific action should a photonic component actually implement? Neither the human brain nor the different neural network concepts make due with a single neuron type, and in this context Jha et al. have integrated a reconfigurable all-optical neuron on a standard silicon photonics platform that accommodates such diversity. The authors leverage the interplay between only three standard photonic components to facilitate the most relevant neuronal activation functions in a single package. Such a re-programmable universal photonic neuron might be key towards more general-purpose photonic neural network architectures, and the demonstrated results are an important step for future neural network processors leveraging integrated photonics.
You must log in to add comments.
Add Comment
You must log in to add comments.
Article Information
Reconfigurable all-optical nonlinear activation functions for neuromorphic photonics
Aashu Jha, Chaoran Huang, and Paul R. Prucnal
Opt. Lett. 45(17) 4819-4822 (2020) View: Abstract | HTML | PDF