Open source deep learning platform dedicated to Embedded hardware and Europe

The NEUROKIT2E project aims at proposing a Deep Learning Platform for Embedded Hardware around an established European value chain (providing AI hardware and software).

This open-source platform will provide the necessary tools for Europe to play on the same level with its American and Chinese competitors, and take the lead on a competitive aspect (still underdeveloped but which will quickly prove to be essential): embedded AI.

The NEUROKIT2Es offer targets to serve European industry needs as a priority, enabling easy and fast Neural Network design, optimization and implementation on constrained hardware targets.

This challenging development will be powered using N2D2 (for Neural Network Design & Deployment) platform, developed by CEA, Coordinator of NEUROKIT2E, for the exploration and development of AI applications, and embed specific hardware components from European players such as NANOXPLORE, STMICRO, , all members of the NEUROKIT2E consortium.NEUROKIT2E will provide essential outcomes such as the end of the dependence of EU stakeholders to American and Chinese tools, the valorisation of EU hardware technologies, the benchmark of hardware and architectures before implementation to ensure the success of embedded AI in devices from energy consumption to sparse resources exploitation, the capacity to overcome the heterogeneity of components through a single, open, sovereign platform while ensuring reliability, safety, transparency.

The NEUROKIT2E framework will impact the AI landscape as the first one dedicated to embedded AI and will position EU stakeholders in this key market.The NEUROKIT2E consortium relies on 5 EU countries active in the hardware and software activities, France, Netherlands, Austria, Germany and Italy, with a balance between private and public research: 11 private companies (5 large industries: THALES, IFAG, TTTECH AUTO AG, STMICRO, DOLPHIN; 5 SMEs: NANOXPLORE, Almende, Spiki, Deepsensing, HTS) ; 4 RTOs (CEA, IMEC-NL, SAL, FBK).