Standard cell pin access and physical design in advanced lithography

Xiaoqing Xu, Brian Cline, Greg Yeric, David Z. PanProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2016 · DOI

TL;DR

As chip manufacturing pushes to smaller nodes (sub-14nm), the way signals access standard cells (pin access) becomes a major bottleneck due to complex lithography rules. This paper proposes a holistic approach, PICO, that optimizes pin access across multiple design stages (cell, placement, routing) to improve overall chip design efficiency and manufacturability.

What Problem Does This Paper Address?

Modern chip manufacturing is rapidly shrinking transistors to sub-14nm nodes. This miniaturization, while enabling more powerful and efficient chips, introduces significant challenges in the physical design process, particularly concerning "pin access." Pin access refers to how electrical signals connect to the fundamental building blocks of a chip, known as standard cells.

The core problem is that as components get smaller and denser, the available space for routing these connections becomes extremely limited. Furthermore, advanced lithography techniques like multiple patterning (MPL), which are necessary to print these tiny features, impose complex geometric and coloring constraints. These constraints make it difficult to route signals without them interfering with each other or violating design rules. This interference can lead to routing failures, impacting the chip's functionality and yield. Existing solutions often address pin access in isolation at a single design stage, failing to capture the full picture and leading to suboptimal results.

Methods

The paper introduces a holistic approach that tackles pin access challenges across three key physical design stages:

  • Standard Cell Pin Access Optimization (PICO): This engine aims to maximize "pin access flexibility" at the individual standard cell level. It defines metrics like "Valid Hit Points" (VHPs) and "Valid Hit Point Combinations" (VHPCs) to quantify this flexibility. PICO incorporates two main techniques:
  • LELE-Aware Via-1 Assignment: This focuses on placing vias (connections between metal layers) in a way that is compatible with LELE (Litho-Etch-Litho-Etch) multiple patterning, avoiding odd cycles in conflict graphs.
  • SADP-Aware Pin Access: This addresses Self-Aligned Double Patterning (SADP) constraints by optimizing line-end extensions for metal wires to ensure manufacturability. This is formulated as a mixed-integer linear programming (MILP) problem.
  • Placement Mitigation: At the placement stage, the paper acknowledges that placing cells too close together exacerbates pin congestion. It suggests using placement mitigation techniques to relax space between cells, improving pin accessibility.
  • Routing and Placement Planning: This stage bridges the gap between cell-level optimization and detailed routing.
  • Intra-cell and Inter-cell Pin Access Pre-computation: A look-up table (LUT) is built to store compatible pin access configurations for individual cells and pairs of adjacent cells.
  • Single Row Pin Access Graph: A graph is constructed to represent the pin access compatibility within a row of standard cells.
  • Local and Global Pin Access Planning: These schemes use the pre-computed information to guide the detailed router. This includes dynamic hit point scoring and a net deferring technique to prioritize nets with fewer access options.
  • Pin Access Driven Rip-up and Reroute: This final step is used to improve overall routability by re-routing nets that couldn't be placed initially, guided by pin access scores.

Datasets

  • ARM 10nm PTM library: Used for standard cell pin access optimization (PICO). This library consists of approximately 700 cells.
  • OpenSparc T1 benchmarks: Used for evaluating detailed routing with pin access planning schemes.
  • Nangate 45nm library (modified and scaled): Used for detailed routing comparisons.

Key Insights and Ideas

  • Holistic Approach is Crucial: Pin access is not an isolated problem. Optimizing it at the standard cell level, then considering its impact on placement and routing, leads to significantly better results than addressing each stage independently.
  • MPL Constraints are Game-Changers: The complex rules imposed by multiple patterning lithography (LELE and SADP) fundamentally change how pin access must be designed, requiring specific awareness in each design stage.
  • Quantifying Pin Access Flexibility: Defining metrics like VHPs and VHPCs provides a concrete way to measure and optimize pin access quality, which directly translates to improved routability.
  • Trade-offs Exist: Achieving better routability through pin access planning often comes with a slight increase in wirelength and via count, highlighting the need for careful balancing.
  • Graph-based Planning for Routing: Representing pin access compatibility as graphs allows for systematic analysis and planning, enabling more intelligent routing decisions.

Key Results

  • Standard Cell Pin Access Optimization (PICO):
    • Achieved a 10x or more improvement in the number of Valid Hit Point Combinations (VHPCs) for most standard cells compared to a basic Design Rule Check (DRC) approach.
  • Over 25% of cells showed a 30% or more improvement in Valid Hit Points (VHPs).
  • The "PICO + MILP" scheme consistently outperformed "PICO + LP" in terms of solution quality, with acceptable runtimes for cell-level optimization.
  • Detailed Routing with Pin Access Planning:
    • Achieved approximately 10% average improvement in routability compared to a state-of-the-art 2D SADP-aware detailed router ("DAC'14").
  • The proposed schemes achieved competitive or faster runtimes than "DAC'14" while delivering better solution qualities.
  • The net deferring technique in pin access planning was shown to be controllable, with a deferring cycle of 3 providing a good balance between routability and runtime.

Limitations and Future Work

  • Limitations:
    • The paper notes that it is difficult to achieve 100% routability in a reasonable runtime due to the complexity of modern designs and SADP constraints.
  • The current placement mitigation techniques may become obsolete with advanced lithography, as simple pin-area cost might not capture detailed routing scenarios accurately.
  • The paper acknowledges that incorporating local M2/via-1 routing congestion into cell cost evaluation at the placement level remains an open question.
  • Future Work:
    • Focus on advanced placement mitigation techniques to further reduce pin access interference.
  • Investigate paradigm-shifting concurrent routing algorithms to overcome the limitations of sequential routing in future technology nodes.
  • Explore how to better incorporate local M2/via-1 routing congestion into placement cost functions.

Summary updated

Abstract

Standard cell pin access has become one of the most challenging issues for the back-end physical design in sub-14nm technology nodes due to increased pin density, limited number of routing tracks, and complex DFM rules/constraints from multiple patterning lithography. The standard cell I/O pin access problem is very difficult also because the access points of each pin are limited and they interfere with each other. There have been several studies across various standard cell and physical design stages, including standard cell pin access optimization, placement mitigation and routing planning, to achieve overall pin access optimization. In this paper, we will introduce a holistic approach across different design stages to deal with the pin access issue while accommodating the complex DFM constraints in advanced lithography.
BibTeX
@inproceedings{xu2016standard,
  author = {Xiaoqing Xu and Brian Cline and Greg Yeric and David Z. Pan},
  title = {{Standard cell pin access and physical design in advanced lithography}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2016},
  doi = {10.1117/12.2222289},
}
Figure