From award to architecture
How GlobalFoundries is rethinking design for manufacturing
GlobalFoundries is integrating design for manufacturing right from the concept phase of new technologies. This is giving rise in Dresden to a data-driven, hybrid DfM approach that shortens time-to-market, secures automotive robustness and makes first-silicon success more likely.
In GlobalFoundries’ Dresden Fab 1, millions of process data points are generated every day. They are used not only to monitor manufacturing, but are also systematically fed back into chip design. Design for manufacturing here is not a downstream optimisation step, but part of the technology architecture – with direct effects on ramp-up stability and time to market.
Oliver Aubel, corporate lead automotive solutions at GlobalFoundries Fab 1 in Dresden, describes DfM as a three-stage system: design learning, lithography optimisation using OPC and retargeting, and yield learning in the process. “DfM for us consists of design learning, targeted lithography adjustments such as OPC and retargeting, as well as yield learning in the process,” he says.
DfM starts before the mask set
At GlobalFoundries, DfM already begins in the concept phase of new technologies. Real manufacturing data, process windows and yield experience from comparable technologies are fed into new designs at an early stage. Customers can use a DfM scoring tool to check the robustness of their still unfinished layouts. Weak points are identified and a DfM kit proposes optimisations. A dry run is possible even before mask data preparation. “DfM for us already starts in the concept phase of new technologies,” says Aubel. “This shortens time to market, and first-silicon prototypes often work straight away on the first attempt using this flow.”
Friction points arise particularly in highly sensitive design configurations that react strongly to manufacturing variations. Early analysis makes it possible to address such risks in a targeted way before they trigger costly iterations.
This has shortened time to market, and first‑silicon prototypes often work first time using this flow.
Data becomes design rules
The digital manufacturing architecture automatically translates statistical patterns and recurring hotspots into concrete design rules. These flow directly into process design kits and layout constraints. Metal lines in the back-end-of-line can, for example, be given “dogbone” structures through OPC adjustments to make them more robust in lithography. In certain cases, retrofits are possible – targeted design adaptations without a completely new mask set.
“DfM at GlobalFoundries is strongly data-driven, but is increasingly evolving into a learning system,” explains Aubel. At the same time, the approach remains deliberately hybrid. Complex interactions between process windows, equipment parameters and layout structures cannot be fully evaluated algorithmically. “Especially in the case of subtle interactions, the practical experience of our process and yield engineers is crucial to avoid false correlations.”
Robustness is more important than performance
As the automotive share grows, priorities in DfM are shifting. In safety‑critical vehicle architectures, robustness, temperature stability and long life cycles take priority. “Robustness and reliability are valued more highly than maximum performance,” says Aubel. For technologies such as FD‑SOI or highly integrated automotive SoCs, design spaces are defined more conservatively and process windows are specified more broadly. Layout rules are defined more strictly in order to safeguard against variability and ageing effects over the long term.
This is where design for manufacturing and design for reliability become closely interlinked. Design restrictions arise from reliability data – performance is not maximised if it could endanger long‑term stability.
Integrated organisation instead of interface problems
DfM is firmly embedded organisationally in Dresden. Yield, DfM and process integration teams work closely with product and technology development. Shared digital platforms and transparent dashboards ensure end‑to‑end visibility of design and manufacturing parameters. As a result, stable structures were already in place before the ramp‑up of larger automotive volumes. During ramp‑up, the focus is on OPC, re‑targeting and comprehensive evaluation as part of the product safe launch. Standardised business processes ensure a secure transition from development to series production.
For us, DfM is not a corrective mechanism but an integral part of technology development.
Connecting innovation speed and stability
In semiconductor manufacturing with around 1,500 process steps and cycle times of up to 90 days, there is no possibility for rework. Errors must be avoided before wafer start. This is exactly where the Dresden approach comes in. Design learning improves fundamental elements across technologies. OPC and re-targeting increase the manufacturability of existing designs. Yield learning optimises process windows without mask adjustments and thus increases yield and reliability in a cost-efficient way. DfM therefore links data, design and process into a closed control loop – and at GlobalFoundries it is becoming a strategic instrument for combining innovation speed with industrial stability.
