interconnect scaling resistance
**Interconnect Scaling and RC Challenges** is the **BEOL engineering problem where shrinking metal line dimensions causes resistivity to increase super-linearly (due to electron surface and grain boundary scattering) while the narrowing line-to-line spacing increases capacitance — compounding the RC delay that has, since the 90 nm node, exceeded gate delay as the dominant performance limiter in digital ICs, forcing the semiconductor industry to pursue new metals, dielectrics, and architectural solutions to prevent interconnects from strangling the performance gains of transistor scaling**.
**The Resistivity Problem**
Bulk copper resistivity: 1.7 μΩ·cm. But at narrow line widths, effective resistivity increases dramatically:
- **Grain Boundary Scattering**: Electrons scatter at Cu crystal grain boundaries. At line widths comparable to grain size (10-30 nm), more boundaries per unit length → higher resistivity.
- **Surface Scattering**: Electrons scatter at the Cu/barrier interface. The ratio of surface to volume increases as lines narrow. The Fuchs-Sondheimer model: ρ_eff = ρ_bulk × (1 + 3λ/(8w) × (1-p)), where λ = electron mean free path (39 nm for Cu), w = line width, p = specularity parameter.
- **Barrier/Liner Volume**: TaN/Ta barrier (2-3 nm) + Cu seed occupies an increasing fraction of the narrow trench. At 12 nm line width: barrier + liner consume 30-50% of the cross-section.
**Effective Resistivity by Line Width**
| Line Width | Cu ρ_eff | vs. Bulk |
|-----------|---------|----------|
| 100 nm | 2.0 μΩ·cm | 1.2× |
| 50 nm | 2.5 μΩ·cm | 1.5× |
| 20 nm | 4.5 μΩ·cm | 2.6× |
| 12 nm | 7-10 μΩ·cm | 4-6× |
**The Capacitance Problem**
As line-to-line spacing shrinks:
- Inter-line capacitance increases (C ∝ k × length / spacing).
- Even with low-k dielectric (k=2.5), the capacitance per unit length increases at each node.
- Coupling capacitance causes: RC delay increase, dynamic power increase (P ∝ CV²f), crosstalk noise between adjacent signals.
**RC Delay Impact**
For a metal line: delay ∝ R × C ∝ (ρ_eff / A) × (k × ε₀ × L² / spacing).
- At 7 nm node: M1 RC delay (~2-5 ps/mm) exceeds gate delay (~1 ps).
- At 3 nm node: M1 RC ~5-10 ps/mm. Interconnect dominates total path delay for all but the shortest wires.
**Industry Solutions**
**New Metals (Lower ρ at Narrow Width)**
| Metal | Bulk ρ (μΩ·cm) | Electron MFP (nm) | Advantage at <20 nm |
|-------|----------------|-------------------|---------------------|
| Cu | 1.7 | 39 | Standard, best bulk ρ |
| Co | 5.8 | 11 | Less size effect below 15 nm |
| Ru | 7.1 | 6.6 | Barrierless (Ru self-barriers), less size effect |
| Mo | 5.5 | 14 | Good scaling, Intel 18A candidate |
| W | 5.3 | 15 | Established CVD process |
- **Co**: Adopted for M0/M1 at 7 nm (Intel). Higher bulk ρ but less severe size effect than Cu at <15 nm.
- **Ru**: Barrierless integration (no TaN/Ta barrier needed), saving cross-section for conducting metal.
- **Mo**: Intel 18A intercept reportedly uses Mo for local interconnect.
**Dielectric Solutions**
- Porous low-k (k=2.0-2.5), air gaps (k_eff ~1.5-2.0): reduce C.
- 3D integration (chiplets, BSPDN): shorten wire lengths, reducing total R×C.
Interconnect RC Scaling is **the fundamental physical limit that governs chip performance at advanced nodes** — the inescapable reality that as wires shrink to nanometer dimensions, their resistance rises and their capacitance increases, creating a signal propagation bottleneck that no amount of transistor improvement can overcome without concurrent interconnect innovation.