local interconnect

**Local Interconnect (M0 / LI)** is the **lowest metal layer that provides short-range wiring within a standard cell** — connecting transistor contacts to each other and to Via-0 without using the first global metal layer (M1), reducing M1 congestion and enabling more compact cell layouts at advanced nodes. **What Local Interconnect Does** - **Connects**: Source/drain contacts to gate contacts within the same cell. - **Routes**: Simple intra-cell connections (e.g., connect NMOS drain to PMOS drain in an inverter). - **Decouples**: Cell-internal routing from inter-cell global routing on M1. **Why Local Interconnect Was Introduced** - At older nodes (28nm+): M1 handled both intra-cell and inter-cell routing — manageable. - At 14nm and below: Cell pin density increases, M1 becomes severely congested. - Adding M0 as a dedicated intra-cell routing layer frees M1 for inter-cell signal routing. **Local Interconnect Implementation** | Node | LI Material | Pitch | Process | |------|------------|-------|---------| | Intel 10nm | Cobalt (Co) | ~36 nm | Subtractive | | TSMC 7nm | Tungsten (W) | ~40 nm | Damascene | | Samsung 5nm | Ruthenium (Ru) | ~28 nm | Subtractive | | Intel 18A | Ruthenium (Ru) | ~22 nm | Subtractive | **Subtractive vs. Damascene M0** - **Damascene** (TSMC): Trench etch in dielectric → barrier + seed → Cu/W fill → CMP. Standard but limited by barrier thickness at tight pitches. - **Subtractive** (Intel, Samsung): Deposit metal blanket → etch metal into lines. No barrier needed inside features — maximum conductor volume. Works best with etch-friendly metals (Ru, Mo). **Impact on Cell Design** - **Bidirectional M0**: Some implementations allow both horizontal and vertical M0 routing within the cell — more flexible cell architecture. - **Unidirectional M0**: Simpler design rules but fewer routing options. - **Dual-M0 (M0A + M0B)**: Two local interconnect sub-layers at orthogonal angles — maximum intra-cell connectivity. **Routing Resources per Cell** - M0: 2-4 tracks per cell (intra-cell only). - M1: 4-6 tracks per cell (inter-cell signal routing). - M2: Power rail (Vdd, Vss) + signal routing. Local interconnect is **essential infrastructure for dense standard cells at advanced nodes** — by handling intra-cell wiring at the lowest metal level, it relieves the routing pressure on M1 and enables the continued cell height scaling that drives logic density improvements.

Go deeper with CFSGPT

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

Create Free Account