Accession Number:

AD1061410

Title:

Scalable Photonic Machine for Neuromorphic Computation Computational Cognition

Descriptive Note:

Technical Report,30 Jun 2015,29 Jun 2017

Corporate Author:

Centralesupelec Yvette France

Personal Author(s):

Report Date:

2018-08-16

Pagination or Media Count:

15.0

Abstract:

The main objectives of the project have evolved since the original proposal, and will be detailed below, but they have kept their original motivation of providing novel paradigms of implementations for optical machine-learning architectures and more specifically reservoir computing. The first objective is to design a large-scale spatiotemporal photonic network with several thousands of nonlinear nodes to realize the physical embodiment of a reservoir computer. We have demonstrated in WP2 that the proposed integrated reservoir can perform well on a typical Boolean task, with very low power consumption. The power budget could even be further reduced in future work by reducing the number of nodes optically injected, for instance by injecting the data only on the four central nodes, as suggested in previous work by some of the authors Katumba17 with a passive chip. Moreover, from an experimental point of view, it is simpler to inject the data on less nodes, as it reduces the routing density on the chip. Finally, current simulation results are encouraging enough to justify a hardware implementation. This could be envisioned in a near future providing additional funding using the PICs4All European technological platform Pics4all16, where designs of photonics chips can be realized on demand for research teams at a reasonable cost. The second objective is to study the possibility of photonics integration of reservoir computer for optical telecom applications. The third objective is to design online training procedure for photonics reservoir computing with full analog output as most design relies on batch training and digitalization of systems states and output. We have currently programmed the training scheme through a simple gradient descent.

Subject Categories:

  • Human Factors Engineering and Man Machine Systems

Distribution Statement:

APPROVED FOR PUBLIC RELEASE