Keynote: Beyond Moore for the AI Factory Era: Heterogeneous Integration, Chiplets and Photonic Interconnects as the Foundations of Future Intelligent Systems
The emergence of AI factories, hyperscale data centers and next-generation autonomous robotic systems is driving an unprecedented demand for computing performance, memory bandwidth and energy-efficient connectivity. As conventional monolithic scaling reaches its physical and economic limits, heterogeneous integration has become a fundamental enabler for the next generation of intelligent computing platforms.
Future architectures will increasingly rely on the seamless integration of heterogeneous chiplets combining advanced logic, memory, RF functions, sensing, and optical communication within highly integrated 2.5D and 3D systems. In this context, silicon photonics and co-packaged optics are emerging as key technologies to overcome the limitations of electrical interconnects, enabling the massive bandwidth density and energy efficiency required by AI infrastructures. CEA-Leti has developed a unique portfolio of technologies spanning advanced packaging, active interposers, 3D integration, RF heterogeneous integration, silicon photonics, III-V integration on silicon, and photonic packaging, bridging the gap from device innovation to system-level demonstrations.
This presentation will discuss the technological roadmap towards future AI computing systems and illustrate how heterogeneous integration will become the cornerstone for building sustainable, high-performance and scalable AI infrastructures.
Key Technologies Covered
- Advanced Chiplet Architectures for AI Systems
- 2.5D/3D Integration and High-Density Interconnects
- Co-Packaged Optics (CPO) for AI Factory Connectivity
- Silicon Photonics and Heterogeneous III-V Integration
- RF and Millimeter-Wave Heterogeneous Integration
- Next-Generation Memory and Compute Integration
- Advanced Thermal Management and Power Delivery
- Design-Technology Co-Optimization (DTCO) and System Co-Design