Replacement Metal Gate Void Free Fill

Void-free fill of the replacement-gate trench is the aspect-ratio problem that RMG integration hits last and worst: once the dummy polysilicon is pulled and the high-k/work-function liners are deposited, what's left is a narrowing slot that the fill metal has to close completely from the bottom up, or the transistor ships with a hollow seam sealed inside its own gate. This is a distinct failure mode from the thermal-sequencing story that makes replacement metal gate work in the first place, and it's distinct from contact-plug fill too — a contact hole is a roughly uniform cylinder etched through dielectric, while a scaled FinFET or gate-all-around gate trench is a tall, re-entrant channel whose open width keeps shrinking as liner after liner gets added inside it, so the geometry gets harder to fill exactly where the process has the least room left to fix it. At the 14nm FinFET node the gate trench opening after all liners was still wide enough for conventional CVD tungsten to seal cleanly. By 7nm and 5nm, with fin pitch under 30nm, the same trench — after a ~1nm interfacial oxide, ~2nm HfO2 high-k, and multiple ~1-2nm TiN/TaN work-function layers for NMOS/PMOS tuning — can leave an open channel of only a few nanometers for the actual fill metal, and that channel is taller (deeper) relative to its width than almost any other fill step in the flow.

Dummy poly pulled
high-k + WF liners already deposited on trench walls
Trench opening narrows
each liner consumes width from both sidewalls
Fill metal deposition — the pinch-off risk windowthis is where a seam or void gets locked in permanently
Nucleation layer
thin CVD or ALD seed, conformal on all walls
Bulk fill (fluorine-free W or Co)
must grow bottom-up faster than sidewall-to-sidewall
Pinch check
if sidewalls meet before bottom fills: seam/void locked in
Gate CMP
polish exposes top — a buried void is invisible here
Electrical test / burn-in
void shows up as gate resistance variation or early fail

The industry's answer has been to change the fill chemistry rather than fight the geometry head-on. Conventional WF6-based CVD tungsten reduces with hydrogen and leaves fluorine residue that attacks the thin high-k/work-function liners in a scaled trench, so advanced nodes increasingly use fluorine-free tungsten (FFW, deposited from WCl5 or similar chlorine-based precursors) or switch the fill metal entirely to cobalt, which wets TiN liners better and nucleates more uniformly at sub-10nm openings. Applied Materials' Volta and Endura platforms and Lam Research's ALTUS line are the tool families most cited for this transition, both pushing toward near-ALD-like layer-by-layer growth so the deposition front advances evenly across the whole trench cross-section instead of growing faster on the upper sidewalls — which is exactly the condition that pinches off a trench from the top before the bottom has filled, sealing a void no CMP step can ever reveal or repair.

Gate-Trench Fill: Conformal vs. Bottom-Up Growth same starting trench, two growth regimes, two outcomes high-k + WF liner stack fill metal (W / Co) void / seam Panel A — Sidewall-dominant growth: pinch-off VOID sealed, unreachable pinch point H2-reduced WF6 CVD, high sidewall rate Panel B — Bottom-up growth: void-free seal flat growth front fluorine-free W or Co ALD-like nucleation Fill-front position vs. deposition time t0 fill complete open sealed sidewall closure (Panel A) bottom fill, lagging bottom-up growth (Panel B)

Fab metrology treats this as a destructive-inspection problem, not an in-line one: because the void is sealed beneath the CMP-polished gate top, it's invisible to overlay or CD-SEM, so fabs rely on cross-section TEM sampling and focused-ion-beam (FIB) slicing of monitor wafers to catch pinch-off trends before they show up as yield loss, often correlating against gate-trench width measured earlier in the flow by critical-dimension SEM. A gate that seals with a buried void doesn't necessarily fail at wafer test — the surrounding metal still conducts — but it carries a resistance penalty and a reliability risk under electromigration-like current stress at the thinned fill cross-section, and that risk scales directly with how close the nominal trench width sits to the fill chemistry's pinch-off threshold, which is why foundries treat gate-trench aspect ratio as a hard scaling limit rather than a tunable margin.

Node generationTypical gate trench opening after linersDominant fill riskCommon mitigation
14nm FinFET~15-20nmMinor, manageable with standard CVD WConventional WF6/H2 tungsten
10-7nm FinFET~8-10nmSidewall pinch-off before bottom fillsFluorine-free W (WCl5-based), thinner liners
5nm FinFET / early GAA~5-6nmHigh void probability with CVD growthCobalt fill, ALD-like nucleation, Applied Volta / Lam ALTUS
3nm GAA nanosheet and beyondSub-5nm, multiple stacked sheetsFill must thread between stacked channels, not just one trenchSelective bottom-up nucleation, further liner thinning

This fill-aspect-ratio ceiling is the quiet reason gate-all-around nanosheet stacks keep pushing liner thickness down rather than up, even though thinner liners complicate the work-function tuning that the replacement-metal-gate module exists to enable — every nanometer saved on the high-k/metal liner stack covered in replacement-metal-gate-rmg is a nanometer handed back to the fill metal's pinch-off margin, and the same bottom-up-versus-conformal growth tradeoff this row describes is the direct geometric cousin of the seam-avoidance problem already covered for contact holes in tungsten-plug-contact-fill-process — different trench, same physics, and together the two rows cover tungsten's complete role in the transistor, from the gate itself down to the contact that reaches it.

Take replacement metal gate void free fill further

Ask the copilot about this term, or have our engineers assess it against your process.