GlobalFoundries has unveiled FDX Fusion, a new semiconductor manufacturing platform aimed at robots, vehicles, industrial machines and other devices that need to run AI close to the sensors producing their data. The company is promising “7nm-class” digital performance and more than twice the density of its first-generation FDX technology.
The important limitation is timing: FDX Fusion is a roadmap, not a chip consumers can buy today. GlobalFoundries targets early demonstrator silicon in 2027, a preliminary design kit in mid-2027 and manufacturing in Dresden, Germany, beginning in 2028.
What GlobalFoundries announced
According to the company’s October 9 announcement, FDX Fusion is the next extension of its fully depleted silicon-on-insulator, or FD-SOI, process family. GlobalFoundries is positioning the platform for “physical AI”—systems that sense their surroundings, make decisions and act in the real world.
That includes automotive radar, sensor fusion, factory automation, connected devices, autonomous machines and robotics. Unlike a data-center accelerator built primarily to run very large models, these products often need a mixture of modest AI compute, analog sensing, radio connectivity, embedded memory and real-time control in a tightly constrained power envelope.
GlobalFoundries says FDX Fusion is intended to combine those functions within one process technology rather than forcing a designer to use multiple chips optimized for different jobs.
“7nm-class” does not mean a conventional 7nm chip
The phrase 7nm-class digital performance needs context. GlobalFoundries is not claiming that FDX Fusion is the same manufacturing process as the 7nm FinFET nodes commonly associated with high-end CPUs and graphics processors.
FD-SOI uses a thin insulating layer beneath the transistor channel to reduce leakage and allow engineers to adjust transistor behavior. It remains a planar approach, while leading high-performance processors generally use FinFET or newer gate-all-around transistor structures.
GlobalFoundries’ claim is that its new process can deliver digital performance in the 7nm class while retaining strengths in analog, mixed-signal, radio-frequency and low-leakage design. That combination may be more valuable for a radar controller or battery-powered robot than chasing the highest possible CPU clock speed.
The confirmed technical targets
GlobalFoundries lists several goals for the first FDX Fusion generation:
- More than 2× density compared with first-generation FDX technology
- 7nm-class digital performance
- RF performance up to 0.5THz fT/fmax
- Integrated analog and mixed-signal functions
- Low-leakage operation
- Embedded non-volatile memory
- In-memory computing support
- Dense logic and secure connectivity options
These are company targets, not independently measured production results. GlobalFoundries has not published final power curves, yields, transistor counts, wafer pricing or customer product benchmarks. The first demonstrator chips are still months away.
Why this process could matter for robots and cars
Many AI devices need more than raw neural-network performance. A robot must read cameras and other sensors, communicate over wired or wireless links, control motors with predictable timing and make safety-related decisions without waiting for a cloud server.
A vehicle has similar requirements. Radar and camera data must be processed with low latency, while networking, security and control functions need to remain reliable across temperature and power changes. Integrating more of that work into fewer chips can reduce board space, communication delays and system complexity.
BMW, Bosch, Infineon, NXP, Siemens and Tesla provided supportive statements in the announcement. Their comments indicate industry interest, but they are not product commitments. None of those companies announced a retail device or vehicle using FDX Fusion.
The development schedule
The current roadmap has three practical milestones:
- Early 2027: Demonstrator silicon targeted for initial customer evaluation
- Mid-2027: Initial process design kit, or PDK, planned for chip designers
- 2028: Manufacturing scheduled to begin at GlobalFoundries’ Dresden facility
A PDK gives chip designers the models, rules and building blocks needed to create a design for a particular manufacturing process. Releasing one does not mean consumer products ship immediately. Designing, validating and qualifying an automotive or industrial chip can take years, so the first widely available products may arrive well after manufacturing begins.
GlobalFoundries also says the start of development work in Dresden is contingent on German government approval under the IPCEI Advanced Semiconductor Technologies program. That dependency makes the 2028 timeline a target rather than a guarantee.
Dresden is getting a €1.1 billion expansion
FDX Fusion will be tied to GlobalFoundries’ SPRINT expansion in Dresden. The company says it is investing €1.1 billion to raise the site’s capacity to more than one million 300mm wafers annually by the end of 2028, up from a current maximum of roughly 950,000 wafers depending on the product mix.
The expanded plant is intended to support automotive, aerospace, defense and industrial customers that value European manufacturing and long product lifecycles. GlobalFoundries says it will also be able to offer Europe-only processes and data flows for customers with strict security requirements.
The investment fits a broader strategy of supplying specialized chips and packaging rather than competing directly with the most advanced smartphone and server processors. A day before the FDX Fusion announcement, GlobalFoundries disclosed a $2 billion agreement with TSMC to manufacture silicon interposers in New York for advanced AI chip packages.
What this means for consumers
There is no FDX Fusion processor, price or release date for consumers. The platform is sold to chip designers and equipment manufacturers, not directly to buyers.
If the roadmap succeeds, its effect will be indirect: future cars, robots, smart-home equipment, industrial sensors and wearables could combine more local intelligence with lower power use and less dependence on a constant cloud connection. Local processing can also reduce latency and keep some sensor data on the device, although privacy and security ultimately depend on the finished product’s design.
Consumers should not expect the “7nm-class” label to translate automatically into faster laptops or phones. FDX Fusion is aimed at a different balance of compute, sensing, connectivity and control.
What remains unknown
GlobalFoundries has not identified launch customers, final chip costs, manufacturing yields or the exact products expected to use the platform. It has also not provided independent performance-per-watt comparisons against modern FinFET processes or competing FD-SOI roadmaps.
The most meaningful proof points will come when demonstrator silicon reaches customers in 2027. Until then, FDX Fusion is an ambitious manufacturing plan with credible industrial partners—but still a plan.
Sources: GlobalFoundries FDX Fusion announcement, GlobalFoundries Dresden expansion announcement, ComputerBase, and Reuters.