- Marvell Technology (MRVL, Financial) launches an innovative multi-die packaging platform, enabling AI accelerators that are 2.8 times larger than traditional designs. - The platform features a modular RDL interposer, improving die-to-die interconnect efficiency and reducing power consumption and manufacturing costs. - Marvell is collaborating with three of the top four hyperscalers, with the solution now in production ramp.
Marvell Technology, Inc. (MRVL, Financial) has announced the introduction of its advanced multi-die packaging platform, aimed at enhancing AI accelerator performance. The platform facilitates multi-die accelerator designs that are 2.8 times larger than conventional single-die implementations, offering significant improvements in compute density.
A key feature of Marvell's new platform is its modular re-distribution layer (RDL) interposer, which integrates 1390 mm² of silicon and four HBM3/3E memory stacks. This novel approach allows more efficient die-to-die interconnects, which result in lower power consumption and better thermal management, along with increased chiplet yields and reduced total cost of ownership.
The platform is already production-qualified with a major hyperscaler and is currently entering production. Marvell's collaboration extends to key industry partners, including ASE, Amkor, SEMCO, and SPIL, underscoring the ecosystem's alignment with this innovative approach. These collaborations are anticipated to support the development of custom XPUs and CPUs for cloud and AI clusters.
This packaging solution not only addresses supply chain flexibility concerns but also aligns with Marvell's broader strategy to transform AI infrastructure performance and value. With the semiconductor industry increasingly moving towards chiplet-based architectures, Marvell's solution is positioned to play a pivotal role in the next generation of data center infrastructure development.
