Samsung unites HBM, logic, optics in advanced AI packaging

Samsung is moving to address a critical constraint facing AI infrastructure operators. As AI models grow larger and data centers expand to support them, two physical limits become acute: the power required to run and cool the systems, and the bandwidth—the speed at which data can move between memory, processors, and other components. Traditional data center designs separate these functions across different chips and modules, forcing data to travel longer distances and consuming more energy with each transfer.

By packaging high-bandwidth memory, logic chips, and silicon photonics together, Samsung is attempting to shrink these distances and optimize the pathways data takes. The use of silicon photonics—which transmits data using light rather than electrical signals—offers the potential for faster, more energy-efficient data movement. This integrated approach aligns with the industry's shift toward chiplets and advanced packaging as a way to improve performance when the limits of traditional semiconductor scaling grow tighter.