NEoteRIC, an EU-funded project in the framework of Horizon 2020, proudly announces the completion of a high-performance photonic processor based on programmable photonics being capable of revolutionizing microwave photonics and artificial intelligence applications.
NEoteRIC’s primary objective is the implementation of holistic photonic machine learning paradigms addressing imaging applications in an unconventional approach to provide paramount frame rate increase and classification performance enhancement. The technological cornerstone of NEoteRIC relies on the development and upscaling of a high speed reconfigurable photonic FPGA-like circuit that will incorporate highly-dense and fully reconfigurable components.
Key Features of NEoteRIC’s Photonic Processor Include:
Multiple processing units: The processor contains 140 tunable basic units
Multiple inputs for more complex AI tasks: Over 30 inputs and outputs each offering advanced capabilities of programmability and complex neuromorphic processing compared to the state of the art.
Superb Controllability: The processor is equipped with multiple photodiodes that monitor the power at different points of the programmable mesh so as to precisely adjust the weights for demanding AI applications and a dense low frequency control unit that regulates multiple tunable basic units in parallel.
Low loss optical packaging: The NEoteRIC Photonic Processor uses advanced 2.5D optical and electrical packaging including 64 single-mode fibers co-packaged with 450 electrical I/Os. The package incorporates silicon-photonic die with photonic bumps and electrical bumps packaged on an optical/electrical interposer. The Interposer hosts two Photonic-Plug fiber connectors with high fiber count of 32 fibers per connector. The package coupling losses remain below 3 dB at the input and the output.
NEoteRIC’s “unconventional” chips are being assessed as a proliferating neuromorphic computational platform that allows optical implementation of powerful non-von Neumann architectures such as Reservoir Computing, Recurrent Neural Networks, Deep Neural Networks and Convolutional Neural Networks simultaneously by the same photonic chip.
Currently, NEoteRIC’s disruptive photonic platform is co-integrated with advanced medical imaging modalities such as single-pixel time-stretched cytometry and event-based cytometry; paving the way for the infiltration of analogue, high-speed optical processing to state-of-the-art, industry-ready applications. Results up to now are really promising, offering unprecedented low power consumption and data throughput, thus validating NEoteRIC’s far-sight technological vision.