Static device performance in a gate-all-around (GAA) nanosheet transistor lives or dies on the inner spacer - the small dielectric region that isolates the gate from the source/drain epitaxy between stacked channels. Get it wrong and parasitic capacitance balloons; get it right and the device beats the FinFET it replaces. International Business Machines Corporation filed on this problem years before any foundry shipped GAA, and its early 2020 grant is a clean example of method-claim drafting in a contested area.
US10553679B2, "Formation of self-limited inner spacer for gate-all-around nanosheet FET" (issued 2020-02-04), is classified in H01L 29/66545 (manufacture of insulated-gate FETs) with H01L 21/02532 and related fabrication subclasses. Read the title as the claim's spine: the novelty is not the spacer as a structure but that its formation is self-limited - the recess and fill are bounded by the process chemistry rather than by a timed etch. That distinction is the whole game in claim construction here.
Why does a process limitation matter more than a structural one? Because a competitor can almost always reach a similar final geometry by a different route. A claim that reads on 'an inner spacer of width X' is broad but vulnerable to prior art; a claim that reads on a specific self-limiting mechanism is narrower but harder to design around without giving up the yield benefit that made the method worth patenting.
The design-around space, then, is in the formation chemistry. A foundry that achieves a comparable spacer profile using a timed, non-self-limited etch, or a fundamentally different selective-deposition scheme, operates outside the literal scope of the independent claim - even if the finished transistor looks identical in cross-section. That is the kind of distinction that decides freedom-to-operate analyses.
The assignee story is also instructive. IBM Research generated a deep stack of GAA process IP in the 2018-2020 window - inner spacers, threshold-voltage control, bottom isolation - well ahead of volume production at Samsung or TSMC. Much of that portfolio became licensing and cross-license leverage rather than product, which is the characteristic pattern of a research-led IDM-turned-licensor.
For an IP strategist, the lesson is to read GAA grants from this era as method claims first. The structures they describe became table stakes; the enforceable value sits in the specific fabrication sequences, and that is where the priority dates and claim language actually bite.