【CEA-Leti】2020边缘AI软硬件协同创新与多领域应用报告_17页_8mb

Edge Artificial Intelligence Summary

Core Content

Edge Artificial Intelligence (Edge AI) refers to the deployment of AI algorithms directly on hardware devices, rather than relying solely on cloud-based processing. This approach is essential for applications requiring real-time decision-making, privacy, and energy efficiency. CEA-Leti, a leading technology research institute in France, is at the forefront of developing Edge AI solutions that are both high-performance and low-power.

Main Points

1. Why Edge AI is Necessary

• Responsiveness: Applications like braking systems in autonomous vehicles require immediate processing without cloud latency.
• Privacy: With GDPR regulations, data processing at the edge ensures that sensitive information remains local and secure.
• Efficiency: Reducing data transfers and storage costs by processing data on-site, where it is generated.
• Autonomy: Enables devices to make decisions independently, which is critical for medical applications like diabetes management.

2. CEA-Leti's Role

• Mission: Combine high-performance computing with low energy consumption in ultra-miniaturized systems.
• Program: Launched an Edge AI program to develop semiconductor solutions that support AI at the edge, from design to manufacturing.
• Collaboration: Works with CEA-List and other partners to create integrated hardware/software solutions.

3. Key Technologies

• In-Memory Computing: Reduces energy consumption by performing computations directly in memory.
• Neuromorphic Architectures: Inspired by biological systems, these architectures allow for energy-efficient, parallel processing.
• 3D Stacking: CEA-Leti and Intel collaborate on 3D IC technologies to enhance performance and integration.
• Resistive Memories (OxRAM, PCM): These non-volatile memories are key for low-power, high-performance Edge AI systems.

4. Applications and Success Stories

• Diabeloop: A French company using Edge AI to automate diabetes treatment via a system that integrates AI with CGM and insulin pumps.
• SPIRIT: The world's first fully integrated neural network on-chip using non-volatile resistive memory, enabling low-latency, low-power AI.
• Retine: A programmable vision chip for high-speed image analysis.
• Intact: A 3D chiplet-based system for advanced computing.
• Sigma Fusion: A low-power sensor fusion solution for safe perception in AI applications.

5. Ecosystem and Collaboration

• Grenoble as a Center of Excellence: Recognized as one of four French AI centers, with a strong multidisciplinary ecosystem.
• MIAI @ Grenoble-Alpes Institute: A key partner in AI research, focusing on health, environment, and energy.
• European Projects: CEA-Leti collaborates with IMEC (Belgium), Fraunhofer/FMD (Germany), and others to build a pan-European AI platform.

Key Information

• Data Processing: 80% of data will need to be processed at the edge by 2025, compared to 20% currently.
• Energy Efficiency: Edge AI systems aim to perform thousands of billions of operations per second with less than 1 Watt of power.
• Research Focus: CEA-Leti is developing in-memory computing, neuromorphic systems, and 3D stacking technologies to support Edge AI.
• Budget and Impact: The program has a €54 million budget, supporting 28 chairs of excellence across various AI domains and fostering public-private collaboration.
• Leadership: Emmanuel Sabonnadière, CEO of CEA-Leti, has a background in semiconductor and industrial leadership, and is actively involved in various AI and technology initiatives.

Appendix Highlights

• AI Tutorials: CEA-Leti offers educational resources to help understand the technologies behind AI.
• CEA-Leti Overview: Founded in 1967, CEA-Leti is a global leader in micro- and nanotechnologies, with over 1,900 employees, 3,100 patents, and cleanroom facilities in Grenoble, France.
• Location and Partnerships: CEA-Leti has offices in Silicon Valley and Tokyo, and collaborates with major companies like STMicroelectronics and Schneider Electric.

Conclusion

Edge AI is becoming essential for real-time, secure, and efficient AI applications. CEA-Leti is a key player in this field, combining cutting-edge research with industrial collaboration to develop next-generation AI solutions. Their focus on hardware-software integration, energy efficiency, and neuromorphic computing positions them as a global leader in the advancement of Edge AI technologies.