Effective decap insertion in area-array SoC floorplan design

Chang-Tien Lu, Ieee Fellow, Chao-Hung Lu, Hung-Ming Chen, Chien-Nan Jimmy LiuACM Trans. Design Autom. Electr. Syst 2008 · DOI

TL;DR

This paper tackles the challenge of power supply noise in modern chip designs by proposing a novel two-step approach that integrates decoupling capacitor (decap) insertion directly into the floorplanning stage. By optimizing decap placement and sizing earlier in the design flow, the authors achieve significant reductions in decap budget and overall chip area while still meeting power integrity requirements.

What Problem Does This Paper Address?

As VLSI technology shrinks, operating voltages decrease, leading to tighter noise margins and increased susceptibility to power supply noise. This noise, primarily from IR-drop and Ldi/dt effects, can cause erroneous chip functions and reduce yield. While decoupling capacitors (decaps) are a common solution, traditional methods insert them late in the design flow (after routing), often leading to overestimation of the required decap budget and wasted chip area. Furthermore, existing approaches don't fully exploit the available empty space within a floorplan, which could be used for decap insertion without increasing the overall chip size. This paper specifically addresses these inefficiencies in the context of area-array architectures, which are increasingly popular for high-performance chips due to their uniform bump distribution.

Methods

The paper introduces a two-step methodology:

  1. Noise-Driven Floorplanning Algorithm: This algorithm aims to reduce potential noise by intelligently placing high-current consumption modules further apart. It utilizes an O-tree representation, a powerful data structure for floorplans, and introduces modified "Delete" and "Insert" operations. These new operations are designed to prevent high-current modules from being placed adjacently, thereby mitigating noise hotspots. The algorithm employs a simulated annealing approach to explore the design space and find an optimized floorplan.
  2. Noise-driven Decap Planning with Minimum Area Insertion (NDP_MAI): After the floorplan is generated, this approach focuses on inserting the minimal required decap budget. It first calculates the optimal decap size for each module using a refined model that considers both peak and average current consumption, and then decomposes this budget into smaller decaps. Crucially, it identifies "feasible regions" for these smaller decaps, prioritizing "empty space" (areas not occupied by modules) over "available space" or "extensive space" to minimize area overhead. The algorithm then selects the optimal location within these feasible regions to insert the decaps, aiming to reduce both area cost and noise. Finally, a compensation equation is used to account for the impact of wire resistance on decap effectiveness.

Datasets

The paper evaluates its methodology using five MCNC benchmark circuits: apte, hp, xerox, ami33, and ami49. These are standard benchmarks used in the field for evaluating floorplanning and related algorithms. The authors note that these benchmarks do not inherently include noise constraints, so they set specific noise constraints (0.13V and 0.25V) for their experiments.

Key Insights and Ideas

  • Early Decap Integration is Key: The most significant insight is that integrating decap planning into the floorplanning stage, rather than treating it as a post-routing step, offers substantial benefits in terms of area reduction and decap budget optimization.
  • Exploiting Empty Space: The paper highlights that floorplans often contain unused "empty space" that can be effectively utilized for decap insertion without increasing the overall chip area. Prioritizing this space is a critical aspect of their approach.
  • Decomposition of Decaps: Breaking down the total required decap budget into smaller, strategically placed decaps not only helps fit them into available space but also reduces the charge/discharge time, improving effectiveness.
  • Noise-Aware Floorplanning as a Precursor: The noise-driven floorplanning step is crucial because it proactively reduces noise hotspots, thereby lowering the overall decap requirement before the insertion phase even begins.
  • Refined Decap Budgeting: The authors propose a more accurate model for calculating decap budgets that considers the difference between peak and average current consumption, leading to less overestimation compared to previous methods.

Key Results

  • Reduced Decap Budget: The proposed methodology reduces the decap budget by an average of 38% compared to the approach by Zhao et al. [2002]. For specific benchmarks like ami33 and ami49, reductions of 52% and 46% were observed, respectively.
  • Minimized Area Overhead: The approach significantly reduces the increased area due to decap insertion. For example, on the ami49 benchmark, their method resulted in an increased area of 154,990 µm², while an integrated approach (floorplan + decap computation + insertion) led to a much larger 112,652 µm² increase.
  • Effective Noise Suppression: Despite the reduced decap budget and area, the proposed method effectively suppresses power supply noise, meeting the specified noise constraints (e.g., 0.13V).
  • Improved Floorplan Quality: The noise-driven floorplanning algorithm leads to better initial floorplans by separating high-current modules, which in turn contributes to the overall effectiveness of the decap insertion.

Limitations and Future Work

  • Run-time Complexity: The paper acknowledges that the new "Insert" operation in their floorplanning algorithm has a higher time complexity (O(m²)) compared to some existing methods, which can lead to longer run-times. While their integrated approach shows better area reduction, the run-time penalty is noted as "very huge."
  • Future Work: The authors suggest exploring the integration of power gating technology with decap insertion to further improve decap budgeting and the overall approach. They also mention that their cost function for decap insertion can be extended to consider other objectives like wire length.

Summary updated

BibTeX
@article{fellow2008effective,
  author = {Chang-Tien Lu, IEEE Fellow and Chao-Hung Lu and Hung-Ming Chen and Chien-Nan Jimmy Liu},
  title = {{Effective decap insertion in area-array SoC floorplan design}},
  journal = {ACM Trans. Design Autom. Electr. Syst},
  year = {2008},
  doi = {10.1145/1391962.1391974},
}
Figure