Numerical Delay Models for Buffering Candidate Net Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional buffering approaches in electronic design automation (EDA) are inefficient as they spend significant computational time on nets that do not benefit from repeater insertion, and current methods fail to accurately identify candidate nets for buffering, leading to suboptimal results and wasted effort.
Innovation Solution
The use of numerical delay models to quickly determine whether buffering a net will improve delay, allowing for optimized buffering only on nets that benefit, thereby skipping unnecessary computations and improving overall performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional buffering approaches are used to determine optimal buffering solutions for all nets, then comprehensive buffering optimization is achieved, but significant computational time is wasted on nets that do not benefit from buffering
Solution Approach 1:
The patent applies preliminary action by using a fast analytical model to pre-screen nets and identify only those that are likely to benefit from buffering before applying the computationally intensive numerical optimization. This preliminary filtering step prevents wasted computation on nets that would not benefit from buffering, while ensuring that promising candidates undergo full optimization.
Solution Approach 2:
The patent segments the buffering optimization process into two distinct phases: (1) a fast screening phase using analytical models with logical effort and parasitic delay values to identify candidate nets, and (2) a detailed optimization phase using numerical methods only on the identified candidates. This segmentation allows the system to efficiently handle large numbers of nets while maintaining optimization accuracy where needed.
2Manufacturing precision
If numerical optimization techniques are applied to all nets, then accurate optimal cell sizes are determined, but the computational complexity increases significantly
Solution Approach 1:
The patent applies partial action by using the computationally intensive numerical optimization technique only partially - specifically, only on the subset of nets identified as buffering candidates by the fast analytical model. This avoids the excessive computational complexity that would result from applying numerical optimization to all nets, while still achieving high accuracy for the nets that benefit most from buffering.
3Ease of manufacture
If two-phase buffering approaches are used with buffer topology construction followed by buffer sizing, then systematic optimization is achieved, but most runtime is spent in the buffer sizing phase
Solution Approach 1:
The patent reverses the conventional two-phase approach by performing preliminary buffer candidate identification using fast analytical models before committing to detailed buffer sizing. This preliminary action filters out non-candidate nets early, so that the subsequent buffer sizing phase operates on a much smaller set of nets, dramatically improving overall computational efficiency while maintaining systematic optimization.
Data Source
AI summary
Systems and techniques are described for efficiently and accurately identifying candidate nets that would benefit from buffering. A buffering process can then be performed only on the identified candidate nets. Embodiments described herein can quickly and accurately identify nets for which performing buffering optimization would most likely waste computational time (so they can be skipped for the buffering transformation), thereby improving the overall performance of buffering optimization and overall physical synthesis optimization. Some embodiments use a buffer topology generating process to generate a buffer topology for a net and then use a numerical sizing process to size the buffers in the buffer topology and the driver gate.