The demands for higher bandwidth, lower latency and higher energy efficiency keep growing at a rapid pace, with the unprecedented expansion of AI workloads in modern data centers. Silicon Photonics offers a scalable solution to keep pace with these demands by integrating advanced silicon fabrication and 3D packaging techniques to enable high density optical integration across AI clusters. Key challenges exist around large package sizes, optical coupling of dense arrays of optical fibers, thermal stack design and package reliability, that need to be resolved for continuous evolution of silicon photonics to higher future bandwidth regimes. This will require step changes in currently available electrical and optical integration processes, including materials development, high precision tooling, thermal designs and highly integrated electro-optical test platforms, thus delivering the performance and scalability required for next-generation AI-driven data centers.
Sandeep Razdan is Director of Process Development Engineering at NVIDIA, where he leads the development of advanced silicon photonics packaging and test technology for next-generation GPU & Switch optical interconnects. At Cisco, he spearheaded the development of a cutting-edge Co-Packaged Optics (CPO) system that demonstrated industry-first power savings for AI/ML clusters. His experience also includes building a manufacturing supply chain ecosystem and leading process technology for silicon photonics platform as a Senior Principal Engineer at Broadcom. During his time at Intel, he led pivotal research and development for Silicon Photonics System-in-Package (SiP) architectures and drove various IC package architectures. Sandeep holds a PhD in Materials Science & Engineering from RPI in Troy, NY and has 20+ patents and papers in the fields of Silicon Photonics & Semiconductor processing.