Data center workloads are quickly evolving, demanding high compute density with varying mixes of compute, memory and IO capability. This is driving architectures that are moving away from a one-size-fits-all monolithic solution to disaggregated functions that can be independently scaled for specific applications.

It is imperative to adopt the latest process nodes to deliver the needed compute density. However, doing so with traditional monolithic SoCs presents an inherent disadvantage due to escalating costs and time to market challenges resulting in unfavorable economics. To address this dilemma, chiplet-based integration strategies are emerging where compute can benefit from the most advanced process nodes, while application-specific memory and IO integrations can reside on mature trailing process nodes.