DeepLight has the following objectives:

  1. Design, deploy and experimentally demonstrate a photonic neuromorphic hardware
    1. Demonstrate experimentally all-optical neurons by using integrated optical memory devices as sigmoid and tanh activation elements and to establish the theoretical and experimental transfer function of its activation process.
    2. Design an interferometric, CMOS compatible integrated photonic linear algebra matrix layout employing CMOS plasmonic filtering and phase shifting elements.
  2. Accelerate photonic neuromorphic architectures via WDM and IQ modulation
    1. Employ WDM-based parallelization and IQ multiplexing as performance acceleration factors, designing a ground-breaking performance, accelerated photonic DNN architectures.
    2. Design an integrated WDM photonic neuromorphic hardware layout and to demonstrate its operation via circuit-level simulations when performing in real DL algorithm execution tasks.
  3. Establish theoretical framework and develop deep learning algorithms over photonic neuro-hardware
    1. Establish the theoretical foundations for optically-enabled DL models.
    2. Develop photonic-compatible alternatives for well-known DL layers through weight regularization schemes and noise-aware training process.
    3. Develop non-negative DL models circumventing current accuracy-related barriers.