Chiplet architecture is attracting interest across semiconductor design as high-performance computing, artificial intelligence and data center workloads place new demands on hardware. The concept offers a way to build systems from multiple chip components rather than relying only on a single, increasingly large die. For industrial automation and technology leaders evaluating compute platforms, the discussion raises practical questions about scaling, cost, integration and how quickly hardware can evolve.
A 2019 presentation from Samsung Foundry’s Device Solutions Americas organization welcomed OCP’s ODSA initiative and outlined the company’s foundry context and interest in chiplets. It did not set out a detailed Samsung contribution to ODSA; that portion of the presentation was marked “TBD.” Its value is instead in framing the technical and business pressures that were motivating interest in chiplet-based approaches.
Why chiplet architecture is gaining attention
One attraction of chiplets is the possibility of expanding a design in more than one dimension. The presentation points to increasing chip size in the X and Y directions, as well as adding components in the Z direction through stacking. These approaches are relevant to compute-intensive applications, including AI, where growing models can create demand for more hardware resources.
Modular designs may also help a system adapt over time. The presentation gives the example of SerDes chiplets that could be replaced with photonics, illustrating how a component might be changed as an application or technology requirement evolves. This is a potential design direction, not a claim that such a substitution is universally available or straightforward.
Another motivation is separating the development timelines of different components. A chiplet-based system could, in principle, allow parts of a design to follow different process-node or product schedules. That decoupling may be useful when the pace of change differs across functions, although the overall system still depends on effective integration among its components.
Chiplets, scaling and semiconductor economics
As semiconductor process development advances, the economics of adopting each new node become important to both chip designers and foundries. The 2019 presentation posed a central question: as costs rise, can fewer customers justify designing for a given node, and could lower-cost approaches bring more customers into the market? It raised chiplets as one possible part of that discussion, rather than presenting them as a guaranteed way to reduce costs.
For engineering teams, the economic case cannot be considered in isolation. A design must be assessed as a complete system, including its components, packaging and design flow. A modular architecture may offer flexibility, but the presentation does not quantify its cost or establish that it will be less expensive in every application. The relevant question is whether a particular partitioning and integration strategy fits the workload and business requirements.
Hardware and software innovation move at different speeds
The presentation also highlighted a possible mismatch between software and hardware evolution in data center computing. It contrasted rapid changes in Google search software from 2004 to 2013 with a smaller number of network hardware architecture changes over the same period, citing published material. The example illustrates a broader challenge: software capabilities may develop quickly, while the underlying hardware platform changes on a different schedule.
Chiplets may be of interest in this context because modular components could provide another way to refresh or adapt parts of a platform. Whether that shortens innovation cycles depends on design choices, validation and the ability to integrate components. It is not a substitute for system-level planning, but it highlights why hardware flexibility is part of the conversation around future data center infrastructure.
The role of foundry and packaging services
Chiplet-based products depend on more than the design of individual dies. Foundry capabilities, packaging, test infrastructure, design services and intellectual property all contribute to turning a concept into a manufacturable system. In its presentation, Samsung described its foundry offering in terms of full-service support, capacity and quality assurance, flexible design services, fast turnaround time, test infrastructure and compatibility with established design tools.
The presentation also emphasized the connection between process technology and packaging, including 3D through-silicon via production for high-bandwidth memory and image sensors. These capabilities are relevant context for discussions of multi-component integration, but the source did not specify a particular ODSA implementation or make claims about the performance of a chiplet-based system.
More broadly, the company described its semiconductor activities as spanning memory, foundry, system LSI, LED and display, with design services and IP among the supporting areas. For system integrators and technology decision-makers, this illustrates why evaluating a foundry partner can involve the surrounding ecosystem as well as process technology alone.
Why collaboration and business models matter
The presentation identified a need for new business models alongside technical progress. Modular design can involve multiple components, development organizations and integration responsibilities. That makes collaboration important: participants need clarity on how designs, interfaces and supporting services fit together. ODSA was presented as part of the wider industry conversation, but the presentation did not define Samsung’s future role in the initiative.
For plant managers and automation leaders, the immediate takeaway is not that chiplets should replace conventional designs. Rather, chiplets represent one approach to addressing changing compute requirements. When assessing a platform, consider the workload, expected product evolution, integration needs and the available foundry and packaging support. The architecture should be judged against the application—not on modularity alone.
Conclusion
Chiplet architecture sits at the intersection of computing demand, semiconductor economics and the pace of hardware innovation. Its appeal includes potential flexibility in scaling and component evolution, while its practical value depends on system integration, manufacturing support and viable business arrangements. The 2019 Samsung presentation framed these issues as reasons for interest and industry discussion, not as settled outcomes. For technical decision-makers, careful evaluation of the complete design and its lifecycle remains essential.