Deep-Tech Challenges in 2nm Physical Design Stage

1. Physical Implementation

Standard Cell Design & Placement

GAA introduces discrete nanosheet widths (e.g., 2/3/4 stacks), forcing standard cells to be quantized, which reduces flexibility in sizing and optimization.
Cell height (track-based design) is no longer easily scalable due to BEOL and BSPDN constraints – placement becomes non-uniform and harder to optimize.
GAA-based cells need orientation-aware placement to avoid systematic variation or stress misalignment.

Density & Routing Congestion

The routing tracks per cell have not scaled proportionally with gate density, increasing congestion in lower metal layers.
Additional vias for backside power TSVs restrict floorplan freedom – tools must support power-aware early planning and 3D-aware placement algorithms.

2. Clock Tree Synthesis (CTS)

Skew and Variation Resilience

With ultra-tight skew margins (<20ps), skew and latency optimization in CTS must now consider:
- Process variation-aware buffering
- Temperature-aware timing corners
- Power supply fluctuation due to IR drops and BSPDN routing

CTS Architecture

H-tree and multi-point CTS must coexist in large SoCs, especially for 3D stacked die or chiplets.
Dynamic clock gating and multiple clock domains need integrated CTS + logic optimization.

New Materials, More Delay Uncertainty

Metal resistivity at 2nm increases due to quantum scattering and grain boundary effects, especially in narrow wires (BEOL M0-M2).
Clock nets, with long lengths and high loads, suffer more non-linear delay behavior, requiring enhanced corner modeling.

3. Static Timing Analysis (STA)

Variation-Aware Timing

Traditional PVT corners are insufficient. STA must now include:
- Statistical STA (SSTA)
- Layout-dependent effects: WPE (Well Proximity Effect), DFM-aware variations
- Timing-aware IR drop coupling
Voltage drop + temperature feedback loops must be embedded into timing models.

Clock-Data Path Interactions

With tighter setup/hold margins, cross-domain clock-data interactions (e.g., async FIFOs, CDCs) need path-aware robustness verification.
Tools must support multi-cycle path analysis across chiplets and TSV-connected dies.

4. Design for Testability (DFT)

Scan Insertion in 3D Context

Scan chain planning must be 3D-aware, as TSVs and backside vias limit accessibility.
GAA structures are more prone to subtle defects, so additional controllability/observability points are needed.

Fault Models

Classical stuck-at and transition faults are not enough. At 2nm, we need:
- Cell-aware ATPG
- Path delay faults
- Bridge/open faults from manufacturing defects in nanosheets or BSPDN
Integration with process-aware fault simulation is essential.

Power-Aware Testing

Power droop during scan shift or capture can cause false failures. BSPDN complicates this by distributing power differently – test power models must adapt.

5. Place & Route (P&R)

Placement Challenges

Macro-aware placement needs to consider backside TSV access and alignment.
With tighter tracks, pin access and metal blockage become critical – EDA tools require pin accessibility estimation during early placement.

Routing and DRC Explosion

Advanced DRC rules: double-patterning, EUV rules, via spacing, metal width variation now running into thousands of constraints.
BSPDN requires orthogonal power routing, and routers must avoid power TSVs while preserving signal integrity.
Pin density and congestion need smart track assignment – standard routers often fail without manual guidance or AI-assisted route prediction.

Summary

At 2nm, the traditional design flow breaks down without tight co-optimization across:

Device layout (GAA-aware standard cells)
IR/thermal-aware P&R
3D-aware scan/DFT logic
Variation-aware STA
Skew-minimized, low-jitter CTS
Multiphysics-aware signoff

All this demands new generation EDA algorithms, cross-domain modelling, and AI-augmented design methodologies to even get a functional and testable chip to tape out.

Back to Blog