CEA-Leti, CEA-List, and PSMC Forge Partnership to Develop Advanced 3D-Stacked AI Chiplet Platform
French research institutes CEA-Leti and CEA-List have entered a strategic collaboration with Taiwanese semiconductor foundry Powerchip Semiconductor Manufacturing Corporation (PSMC) to co-develop a next-generation AI chiplet platform. The project, announced for April 2026, aims to integrate RISC-V processors, microLEDs, and silicon photonics within advanced 3D stacking and interposer architectures.
The core objective is to create a modular, high-performance, and energy-efficient system-in-package (SiP) solution tailored for artificial intelligence and high-performance computing (HPC) workloads. The platform will leverage PSMC’s 2.5D and 3D integration foundry services to combine multiple heterogeneous chiplets. Key technological components include:
* RISC-V Cores: Open-standard processors designed for flexibility and efficiency.
* Silicon Photonics: Optical interconnects using microLEDs for high-speed, low-power data transmission between chiplets and potentially off-package.
* Advanced Packaging: Utilization of silicon interposers and 3D stacking (likely using hybrid bonding) to achieve high-density integration and short interconnects.
This collaboration directly addresses critical bottlenecks in AI hardware, such as memory bandwidth limitations and the power consumption of electrical data movement over distance. By integrating optical I/O directly into the package via microLED-based silicon photonics, the consortium targets a significant reduction in communication latency and power draw compared to traditional electrical interfaces.
The partnership combines distinct expertise: CEA-Leti contributes its R&D in silicon photonics, 3D integration, and microLEDs; CEA-List brings its knowledge in embedded AI and RISC-V computing architectures; and PSMC provides the essential industrial-scale advanced packaging and foundry manufacturing capabilities. The resulting platform is envisioned to enable more scalable and powerful AI accelerators, potentially influencing future designs for data centers, edge AI, and autonomous systems.