A multi-project wafer that fills continuously
A multi-project wafer, or MPW, is a simple idea with important economics: several independent chip designs share one physical wafer.
Each customer receives dies containing their own design. The cost of the wafer and the manufacturing setup can be distributed across multiple projects instead of being carried by one customer alone. Europractice describes the effect plainly: by combining designs into one mask set, an MPW prototype can reduce fabrication cost to around 5% to 10% of a dedicated prototyping wafer run REF1.
That is a large reduction, but it is specifically a reduction in fabrication NRE. The customer still has to design, verify, package and test the IC. An imec IC-Link discussion of low-volume ASICs makes the same distinction: MPW lowers the initial mask and wafer commitment, while design expertise and requirements definition remain part of the project REF2.
Conventional MPW services make the manufacturing side work through scheduled shuttle runs. A process is assigned a tape-out date, projects are collected until the deadline, the wafer is assembled, and manufacturing begins as a batch.
That model reduces cost, but it also introduces a calendar: miss the deadline and you wait for the next shuttle.
FlexMPW replaces that fixed shuttle calendar with continuous matching.
Continuous intake instead of a fixed tape-out date
Projects can enter the FlexMPW pool continuously. xSilica keeps track of which designs can share a manufacturing run and how they can be allocated on the wafer. When a compatible combination becomes viable, the wafer closes and manufacturing starts.
That changes the scheduling problem. A traditional shuttle asks which projects can make the next fixed date; FlexMPW asks which projects can be combined now into a viable wafer. The customer provides a small number of parameters that tell the system where flexibility exists.
Target quantity
The target quantity is the number of dies you would ideally like to receive.
It describes the preferred allocation, not necessarily a hard requirement. If the wafer can be closed efficiently at that quantity, there is no reason to allocate more.
Maximum quantity
The maximum quantity defines how far the target allocation may expand.
Unused wafer area does not always need to remain empty. If another project cannot use that space and you are willing to receive more dies, the scheduler may increase your allocation up to the agreed maximum to close the wafer sooner. You can enforce a maximum fixed outcome by setting the maximum quantity equal to the target quantity.
Minimum quantity
The minimum quantity is the smallest useful allocation for the project.
A prototype team may prefer twenty dies but still be able to complete its measurements with ten. Making that distinction explicit gives the scheduler more freedom to fit compatible projects together. Dicing constraints can otherwise leave awkward strips of wafer area that are difficult to use. If a slightly smaller allocation creates cleaner dicing lanes and a better overall layout, the scheduler can reduce the quantity, but never below the customer's stated minimum. A little flexibility across several projects can be enough to close the wafer sooner.
Expansion clock
The expansion clock defines when xSilica may allocate more than the target quantity. Early in the waiting period, the scheduler first looks for an efficient fit close to each project's requested quantity rather than immediately filling everyone to their maximum. Once the expansion clock expires, additional allocation can be used to help close the wafer sooner.
This makes time an explicit scheduling variable. If schedule is relaxed, set a longer clock; if getting onto a wafer quickly matters more, allow expansion sooner.
Process compatibility
Not every design can share every wafer.
Projects must use compatible Rapid CMOS configurations. Where a design can tolerate more than one supported configuration, the customer can allow those alternatives to increase the number of possible matches.
The design remains constrained to validated process options. Flexibility never means combining projects that require incompatible manufacturing steps.
Dynamic tiling
Dicing introduces a geometric constraint that normal two-dimensional packing does not have: the saw streets have to form continuous lanes across the wafer. A combination of die sizes that looks efficient on screen can therefore leave strips or blocks that are difficult to use once the dicing plan is taken into account.
Customers who choose to accept a larger physical die can give the scheduler another degree of freedom. A smaller design may then occupy the space between wider dicing lanes, even though the delivered die is physically larger than the layout itself requires. This option is intended for bare-die and evaluation-carrier deliveries, where that extra outline can be tolerated.
If a larger outline helps close the shared wafer sooner, xSilica absorbs the cost of the unused die area rather than charging the customer for silicon they did not request.
Dynamic physical layout
A FlexMPW project does not need to reserve a permanent coordinate on a future wafer when it is submitted.
Placement remains computational while the wafer is open. The scheduler can rearrange compatible projects as new submissions arrive, searching for a better combination of area, quantity and process requirements. Your dies therefore do not have to occupy one contiguous block. If you prefer adjacent placement to reduce across-wafer variation between samples, you can request it, but that may reduce the number of dies that fit efficiently. Any allocation still remains within the minimum and maximum quantities you specified.
Only when the run closes does that computational layout become the physical wafer plan.
What FlexMPW does not solve
FlexMPW is a manufacturing and scheduling model. It does not, by itself, remove the engineering NRE of creating a custom IC.
That is why it sits inside the wider xSilica platform rather than being the whole proposition. The constrained process and design environment address the design side; FlexMPW addresses how compatible designs can share physical manufacturing without inheriting a fixed shuttle calendar.
What FlexMPW does not promise
FlexMPW does not eliminate the basic economics of sharing.
If there are not enough compatible projects to make a shared run viable, a project may still have to wait. FlexMPW changes the scheduling mechanism; it does not remove the basic economics of sharing.
The difference is that waiting is no longer tied to an arbitrary date months in the future. Every new compatible submission can improve the current solution and potentially close the run.
For projects where schedule certainty matters more than sharing cost, a Dedicated Wafer remains the more direct option.
Flexibility becomes a scheduling resource
The principle behind FlexMPW is to turn real customer flexibility in quantity, placement and compatible process options into shorter waiting time. Instead of asking every customer to conform to the same shuttle date, the manufacturing system uses the flexibility each project actually has to decide when a shared wafer is ready to run.
References
[1] EUROPRACTICE, “MPW Fabrication.” [Online]. Available: https://europractice-ic.com/services/fabrication/. [Accessed: Sep. 8, 2026].
[2] imec, “Innovate with hardware in today’s substream markets,” Dec. 23, 2016. [Online]. Available: https://www.imec-int.com/en/articles/the-new-hardware-hipsters. [Accessed: Sep. 8, 2026].