Oscillatory Neural Networks on FPGA for Robot Obstacle Avoidance Application

Today AI relies on high-performance hardware systems (CPUs, GPUs, and specialized AI accelerators). But such hardware platforms also consume a lot of power and cannot cope with the ever-expanding AI workload and complexity. New thinking is needed to improve AI performance, lower power consumption and match the needs of various novel applications, which cannot merely exist with conventional hardware.
EU H2020 NEURONN contributes to this with its ultra-low-power capability and high energy-efficiency system architecture. We are developing oscillatory neural networks (ONN) with novel nanoelectronics devices to perform ultra-low-power computations. We have developed a first-ever digital implementation of an ONN on FPGA, comprising an AI-on-ONN platform. ONN-on-FPGA controls a mobile robot’s direction based on the obstacles it sees via eight proximity sensors. Depending on the obstacles, ONN issues a signal as “go right,” “go left,” or “go straight.”


Project Acronym and Title: NeurONN: Two-Dimensional Oscillatory Neural Networks for Energy Efficient Neuromorphic Computing

Funding source: H2020

Grant Agreement Number or Funding reference: 871501

Owner/Project contact: CNRS/LIRMM/Aida Todri-Sanial’s team

Country: France

Address: 161 Rue Ada, 34095 Montpellier

Type of organisation: RTO – Research Technology Organisation