14nm FinFETPreview

CG CSOH Spin Coat

263/ 362

PMD Surface Recess Simulation

M2 ESL Cap Deposition
263PMD Surface Recess Simulation
+49 steps

Process Cross-Section

Gate Cut (sense W)Fin Cut (sense L)M1 · M01 · PMD Surface Recess Simulation (CG28 end-state, planar)S/D ContactGate ContactWSiSiN capSiNTiWCESLPOPPMDPolySiO2eSiGeEpiTiNTiAlSiO2 ProtectSiOCNHfO2SiSTIWSiSiN capSiNTiCESLPOPPMDWPolyeSiGeTiNTiAlSiO2 ProtectSiOCNHfO2SiSTI

Step highlight

In surface‑limited regimes, reactive species modify only the outermost atomic layers of the dielectric, weakening network bonds and enabling controlled removal without deep penetration or subsurface damage .

In depth

Device Context and Integration Logic

The PMD Surface Recess Simulation is positioned after tungsten contact fill and CMP to intentionally reshape the post‑CMP topography of the pre‑metal dielectric (PMD) surface, compensating for CMP‑induced erosion and pattern‑density‑dependent heigh

t nonuniformity . CMP removes conductive overburden by combining chemical reactions and mechanical abrasion, and this inevitably exposes and partially erodes the surrounding dielectric, leading to local height loss that propagates into subsequent BEOL layers . If uncorrected, these surface variations translate into non‑uniform dielectric coverage and variable metal line thickness in the first interconnect layer, directly affecting resistance and reliability . By performing a controlled PMD surface recess at this point, the integration flow resets the effective reference plane for the upcoming M1 etch‑stop and ILD depositions, improving thickness uniformity and lithographic focus margin for M1 patterning .

Physical and Chemical Mechanisms

The physical basis of PMD surface recessing is selective dielectric material removal governed by surface‑reaction‑limited chemistry rather than bulk transport, analogous to the principles described for atomic‑scale etching processes . In surface‑limited regimes, reactive species modify only the outermost atomic layers of the dielectric, weakening network bonds and enabling controlled removal without deep penetration or subsurface damage . This contrasts with conventional high‑rate plasma etching, where energetic ions and radicals can induce charge trapping and structural damage in porous or low‑k dielectrics . The self‑limiting nature of surface reactions ensures that local topography differences translate into proportional removal, which is essential for correcting CMP‑induced pattern‑dependent effects rather than amplifying them . From a device‑physics perspective, maintaining a uniform dielectric environment above contacts minimizes parasitic capacitance variation and preserves predictable electric‑field distributions in the first metal level, which is increasingly interface‑dominated at advanced nodes .

Material and Method Selection Logic

The PMD material system in a 14 nm FinFET BEOL stack typically prioritizes mechanical compliance and low permittivity to reduce RC delay, but these same properties make the dielectric sensitive to plasma damage and non‑uniform etching . Therefore, surface recess methodologies favor chemistries and energy regimes that decouple chemical bond breaking from physical sputtering, following the same rationale that motivates ALE‑like approaches for fragile interlayer dielectrics . Process parameters interact directionally: increasing chemical reactivity enhances removal efficiency but risks loss of selectivity to adjacent barrier or cap materials, while increasing physical energy improves uniformity at the cost of dielectric damage . The simulation aspect of this step is critical because dielectric removal interacts nonlinearly with pattern density, similar to how deposition non‑uniformity complicates CMP modeling . Accurate modeling allows prediction of post‑recess topography and ensures that the resulting surface is optimized for conformal ESL cap deposition, which relies on uniform surface chemistry and height to function as an effective diffusion and etch barrier in the next module .

Node‑Specific Considerations at 14 nm

At the 14 nm FinFET node, BEOL integration margins are constrained by tighter metal pitch and increased sensitivity of device performance to interconnect parasitics, making surface topography control more critical than in earlier planar nodes . The three‑dimensional nature of FinFETs increases layout‑dependent effects that propagate into BEOL layers, so even small dielectric height variations can translate into significant resistance and capacitance variability . Compared with later M1 surface recess steps, the PMD surface recess is distinct because it acts on a contact‑level dielectric that directly interfaces with tungsten plugs rather than patterned metal lines, requiring higher selectivity to metal and barrier materials . This early correction of dielectric topography establishes a stable foundation for all subsequent BEOL layers, reducing cumulative variation and supporting manufacturability at the 14 nm technology node .

Risks & Challenges

  • [High] Dielectric Over‑Recess: Excessive surface‑reaction progression can remove more PMD material than intended, reducing dielectric thickness above contacts and increasing parasitic capacitance and leakage risk due to altered electric‑field distribution .
  • [High] Pattern‑Dependent Non‑Uniformity: If surface reactions are not fully self‑limiting, local pattern density can drive differential removal rates, re‑introducing topography variations similar to CMP erosion effects rather than correcting them .
  • [Medium] Plasma‑Induced Dielectric Damage: Residual ion energy or radical penetration can break dielectric network bonds or create charge traps, degrading mechanical integrity and reliability of low‑k PMD materials (Engineering Practice).
  • [Medium] Loss of Selectivity to Contact Metals or Barriers: Insufficient chemical discrimination can partially attack tungsten or barrier interfaces, increasing contact resistance and variability in the subsequent M1 interconnect .
  • [Low] Simulation‑Model Mismatch: Inaccurate representation of pattern‑dependent removal physics in the simulation can lead to systematic bias between predicted and actual post‑recess surfaces, undermining downstream process windows .

A free account opens the full rationale, risk analysis, and the paper and patent citations behind them.

Sign Up Free

Or

Need every step? Pay once for lifetime access.

Pay by card, PayPal, Apple Pay, or Google Pay — exact options shown at checkout · 14-day money-back guarantee