| Project Detail |
Plasmonics potential for more energy-efficient neuromorphic computing
Neuromorphic computing that encompasses devices that can mimic the natural biological structures of the human nervous system presents a promising energy-efficient alternative over conventional computing architectures. The EU-funded PlasmoniAC project will invest in best-in-class material and technology based on plasmonics, to further optimise the computational power, size and energy of neuromorphic chips. If successful, the project will demonstrate a powerful artificial plasmonic neuron suite. It could boast up to three orders of magnitude higher computational efficiency per neuron, and up to six orders of magnitude lower energy consumption, compared to top state-of-the-art neuromorphic machines.
PlasmoniAC invests in neuromorphic computing towards sustaining processing power and energy efficiency scaling, adopting the best-in-class material and technology platforms for optimizing computational power, size and energy at every of its constituent functions. It employs the proven high-bandwidth and low-loss credentials of photonic interconnects together with the nm-size memory function of memristor nanoelectronics, bridging them by introducing plasmonics as the ideal technology for offering photonic-level bandwidths and electronic-level footprint computations within ultra-low energy consumption envelopes. Following a holistic hardware/software co-design approach, PlasmoniAC targets the following objectives: i) to elevate plasmonics into a computationally-credible platform with Nx100Gb/s bandwidth, um2-scale size and >1014 MAC/s/W computational energy efficiency, using CMOS compatible BTO and SiOC materials for electro- and thermo-optic computational functions, ii) to blend them via a powerful 3D co-integration platform with SixNy-based photonic interconnects and with non-volatile memristor-based weight control, iii) to fabricate two different sets of 100Gb/s 16- and 8-fan-in linear plasmonic neurons, iv) to deploy a whole new class of plasmo-electronic and nanophotonic activation modules, v) to demonstrate a full-set of sin2(x), ReLU, sigmoid and tanh plasmonic neurons for feed-forward and recurrent neurons, v) to embrace them into a properly adapted Deep Learning training model suite, ultimately delivering a neuromorphic plasmonic software design library, and vi) to apply them on IT security-oriented applications for threat and malware detection. Succeeding in its targets will release a powerful artificial plasmonic neuron suite with up to 3 orders of magnitude higher computational efficiencies per neuron and 1 and 6 orders of magnitude higher energy and footprint efficiencies, respectively, compared to the top state-of-the-art neuromorphic machines. |
