Skip to content

NEUROKIT2E Ecosystem

NEUROKIT2E is a Horizon Europe-funded collaborative project aimed at developing an open-source deep learning platform for embedded hardware, enabling Europe to lead in the embedded AI domain. The project focuses on providing tools and infrastructure for edge AI and on-device machine learning.

Products and Services

Core Offerings

  • AIDGE Platform*: A deep learning framework designed for embedded AI, offering:
  • Neural Network design, optimization, and deployment.
  • Integration of hardware models with neural network models.
  • Advanced compression and pruning methods to reduce model size while maintaining performance.
  • Support for synchronous (tensor-based) and event-driven (spiking) coding in a single development chain.
  • N2D2 Platform: Developed by CEA (the project coordinator), this tool facilitates neural network design and deployment for embedded applications.

Use Cases

  • Applications include automotive systems, industrial IoT, healthcare devices, and building automation. For example:
  • Codasip's "Occupancy Monitoring for Efficient Building Control" use case leverages radar sensors with RISC-V hardware to process data locally.

Developer Resources

NEUROKIT2E provides extensive resources tailored to developers working on edge AI/ML: - Open-Source Frameworks: AIDGE and N2D2 platforms simplify neural network design and deployment on constrained hardware. - Documentation: Comprehensive guides and tutorials for using the platform. - Code Generation Tools: Automates application-specific optimizations for embedded targets. - Academic Flexibility: Allows researchers to prototype systems while aligning with industrial needs.

Partnerships

NEUROKIT2E is supported by a consortium of 25 partners across five European countries, including: - Large Industries: Thales, Infineon, STMicroelectronics, TTTech, Dolphin Design. - SMEs: NX, GrAI Matter Labs, Almende, Spiki, Deep Vision Systems. - Research Organizations: CEA (Coordinator), IMEC, Silicon Austria Labs (SAL), FBK. - EU Initiatives: Part of the dAIEDGE cluster promoting distributed AI at the edge.

Infrastructure

The project leverages a robust European value chain: - Hardware Integration: Utilizes components from European manufacturers like STMicroelectronics and Infineon. - Benchmarking Tools: Ensures optimal performance for energy efficiency, reliability, and resource utilization across heterogeneous hardware platforms.

Company Strategy

NEUROKIT2E’s strategy focuses on: 1. European Sovereignty in Embedded AI: Reducing dependence on American and Chinese tools by creating a sovereign platform. 2. Real-Time Edge AI Solutions: Meeting critical requirements like energy efficiency, data confidentiality, and real-time processing. 3. End-to-End Development Chain: Simplifying the process from neural network design to hardware implementation. 4. Sustainability: Optimizing energy consumption in embedded devices.

Stakeholders and Shareholders

The project brings together a balanced mix of private companies (12 large industries/SMEs) and public research organizations (4 RTOs). It is funded under the European Union Horizon Europe program via CHIPS JU.

Recent Developments

  1. Workshops & Conferences:
  2. Presented at HiPEAC 2024 and the European Conference on Edge AI Technologies.

  3. Showcased the AIDGE framework's capabilities in real-world applications.

  4. Funding & Progress:

  5. Received €19.75 million in funding from Horizon Europe.

  6. Achieved significant milestones in developing compression techniques and hybrid coding frameworks.

  7. Collaborations:

  8. Codasip’s RISC-V cores are being tailored for spiking neural networks (SNNs) as part of use case implementations.

NEUROKIT2E represents a significant step forward in empowering Europe’s edge AI ecosystem. By providing an open-source platform that integrates hardware-software co-design with advanced neural network optimization techniques, it positions European stakeholders as leaders in the competitive embedded AI market.

References

  • [1] https://www.neurokit2e.eu
  • [2] https://silicon-austria-labs.elsevierpure.com/en/publications/neurokit2e-deep-learning-for-edge-ai-applications-targeting-embed
  • [3] https://rpanderson-neurokit2.readthedocs.io/en/latest/
  • [4] https://github.com/neuropsychology/NeuroKit
  • [5] https://edge-ai-tech.eu/conference-homepage-2024/
  • [6] https://archivio.unime.it/sites/default/files/XXVII%20CdA%2031.01.2023%20ESR%20NEUROKIT2E.pdf
  • [7] https://codasip.com/2024/11/04/neurokit2e-a-deep-learning-platform-dedicated-to-embedded-hw-and-europe/
  • [8] https://edge-ai-tech.eu/conference-homepage-2023/
  • [9] https://neuropsychology.github.io/NeuroKit/functions/index.html
  • [10] https://codasip.com/company/europe-support/