interconnect rc delay

**Interconnect RC Delay** is the **signal propagation delay through on-chip metal wires caused by wire resistance (R) and parasitic capacitance (C)** — which has surpassed transistor gate delay as the dominant performance limiter at advanced nodes, with the RC time constant increasing as metal cross-sections shrink despite improvements in conductor and dielectric materials. **The RC Delay Problem** - **RC delay**: $\tau = R \cdot C = \rho \frac{L}{A} \cdot \epsilon \frac{A_{cap}}{d}$ - As metal pitch scales: wire cross-section shrinks → R increases. Wire spacing shrinks → C increases. - **Double penalty**: Both R and C get worse simultaneously. - At 28nm: gate delay ~5 ps, interconnect delay ~20 ps — wires are 4x slower than transistors. - At 3nm: gate delay ~1 ps, interconnect delay ~50+ ps — wires are 50x slower. **Resistance Scaling** - Copper resistivity increases dramatically at nanoscale due to: - **Grain boundary scattering**: More grain boundaries per unit length in narrow wires. - **Surface scattering**: Electrons scatter off wire surfaces (Fuchs-Sondheimer effect). - **Barrier/liner thickness**: 2-3 nm TaN/Ta liner occupies 20-40% of wire cross section at M1 pitch < 30 nm. - Cu bulk: 1.7 μΩ·cm → Cu at 20 nm width: ~5-8 μΩ·cm (3-5x increase). **Capacitance Scaling** - Wire-to-wire capacitance: $C \propto \epsilon_r \frac{H}{S}$ (H = wire height, S = spacing). - Low-k dielectrics: SiO2 (k=4.0) → SiCOH (k=2.5-3.0) → Air gap (k=1.0). - Further k reduction limited by mechanical and thermal requirements. **Solutions Being Deployed** | Approach | Target | Benefit | |----------|--------|---------| | Alternative metals (Co, Ru, Mo) | Lowest metal levels | Thinner barriers → more conductor area | | Air gap dielectrics | Tightest pitch layers | k=1.0 between wires | | Backside power delivery (BSPDN) | Power/ground routing | Frees front-side for signal routing | | Subtractive patterning (Ru, Mo) | Tightest pitches | Avoids damascene barrier limitations | | Repeater insertion | Long signal paths | Break long RC lines into shorter segments | **Impact on Chip Architecture** - **Chiplets**: Avoid longest on-chip wires by splitting into smaller dies. - **3D stacking**: Vertical connections (TSV, hybrid bonding) shorter than horizontal wires. - **Near-memory compute**: Minimize data movement distance to reduce interconnect bottleneck. Interconnect RC delay is **the fundamental performance bottleneck of modern semiconductor technology** — solving it requires simultaneous innovation in conductor materials, dielectric materials, patterning approaches, and chip architecture, making BEOL engineering as critical as transistor design.

Go deeper with CFSGPT

Get AI-powered deep-dives, save terms, and run advanced simulations — free account.

Create Free Account