Lumentum Holdings, Qualcomm Technologies, and Corning Incorporated announced a collaboration to demonstrate a high-density optical die-to-die interconnect for artificial intelligence systems. The companies will showcase the technology at ECOC 2026 in Málaga, Spain, combining Qualcomm's die-to-die interface intellectual property with Lumentum's vertical-cavity surface-emitting laser platform and Corning's multimode fiber connectivity.
How the technology works
Die-to-die, or D2D, refers to direct communication between semiconductor chips. The demonstration uses optical channels operating at 32 gigabits per second using non-return-to-zero signaling, a simple digital encoding scheme. The architecture is designed to support emerging D2D standards such as UCIe, which stands for Universal Chiplet Interconnect Express, an open standard for connecting chiplets within a package or across short distances. By using optical rather than electrical links, the system can extend high-density connectivity while maintaining low latency and power consumption.
The proof-of-concept configuration demonstrates approximately one terabit per second per millimeter of shoreline bandwidth density. Shoreline bandwidth density measures how much data can flow across the edge of a chip or package. The companies project this could scale to roughly four terabits per second per millimeter through higher lane rates and denser channel arrangements.
Lumentum's optical engine is able to achieve approximately 10 terabits per second of aggregate transmit and receive capacity. VCSELs are semiconductor lasers that emit light vertically from the chip surface, making them suitable for dense two-dimensional arrays. The platform uses backside lens technology and draws on Lumentum's manufacturing experience from three-dimensional sensing applications. Corning contributes multimode fiber glass ferrules for high-density fiber connectivity, reflecting the growing importance of fiber density as optical links move closer to compute devices.
AI systems are demanding more computational power than ever, but traditional copper-based electrical interconnects are hitting physical limits as data volumes surge and chip designs become more complex. Optical connectivity offers a way to break through these barriers by carrying vastly more data with lower energy consumption and heat generation. This development from Lumentum, Qualcomm, and Corning points toward a future where AI infrastructure can scale more efficiently, enabling larger models and faster training without the prohibitive power and thermal costs that currently constrain data center growth.