Statistical Timing Analysis for Latch Circuits with Feedback Loops
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current statistical timing analysis methods are inefficient in handling level-sensitive latches and feedback loops in high-end VLSI circuits, leading to computational complexity issues and impractical application in modern circuits, especially when dealing with self-dependent timing variables.
Innovation Solution
The method involves iterative calculations to predict circuit timing yield and critical cycle mean using a clock scheme and reduced timing graph, with iterative processes to update signal arrival and departure times and cumulative delays, addressing self-dependence and computational complexity through linear combinations of Gaussian random variables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Monte Carlo simulation is used for statistical timing analysis, then timing distribution accuracy is improved, but computational time increases exponentially
Solution Approach 1:
The circuit is divided into independent paths for separate analysis. Each path's timing characteristics are computed independently and then combined, avoiding the need to simulate all possible scenarios simultaneously. This segmentation reduces computational complexity while maintaining accuracy for circuits without feedback loops.
Solution Approach 2:
The patent transforms the computational approach by changing parameters from full probabilistic simulation to using mean and variance propagation. By assuming Gaussian distributions and using analytical solutions for path-based STA, the method achieves acceptable accuracy with significantly reduced computational time compared to Monte Carlo simulation.
2Productivity
If path-based STA is used to identify critical paths, then computational complexity is reduced, but accuracy deteriorates due to exponential growth with circuit size
Solution Approach 1:
The patent extracts and removes feedback loops from the circuit graph, transforming it into a Directed Acyclic Graph (DAG). This extraction eliminates the exponential complexity issue by converting the problem into a form that can be solved efficiently using standard path-based STA algorithms on acyclic structures.
Solution Approach 2:
The patent introduces an iterative dynamic approach where feedback loops are temporarily broken and analyzed in iterations. In each iteration, the timing analysis is performed on the current graph structure, and results are used to update the analysis in subsequent iterations until convergence is achieved, enabling handling of complex circuits with feedback.
3Productivity
If block-based STA is used for progressive computation, then computational complexity becomes linear, but accuracy deteriorates due to inability to handle feedback loops and latches
Solution Approach 1:
The patent introduces an intermediary transformation step that converts circuits with feedback loops and latches into equivalent DAGs by breaking feedback paths. This intermediary representation allows the use of efficient block-based STA algorithms while maintaining the ability to handle complex circuit structures through iterative analysis of the transformed graph.
Solution Approach 2:
The patent employs periodic iterative analysis where the timing computation is performed in repeated cycles. Each iteration analyzes the circuit graph, updates timing parameters, and checks for convergence. This periodic action enables the handling of feedback structures by progressively refining results until stable values are achieved.
4Ease of operation
If classical worst-case timing analysis is used, then simplicity is maintained, but timing predictions become overly pessimistic and conservative
Solution Approach 1:
The patent replaces the mechanical worst-case analysis approach with a statistical methodology. Instead of assuming all delay variations occur simultaneously in the worst direction, the method uses probability distributions (Gaussian assumptions) to model timing variations and computes statistical metrics like mean and variance, providing more realistic predictions while maintaining computational tractability.
Data Source
AI summary
Statistical timing analysis methods for circuits having latches and feedback loops are described wherein the circuit yield, and/or the critical cycle mean (the largest cycle mean among all loops in the circuit), may be iteratively calculated with high speed and accuracy, thereby allowing their ready usage in the analysis and validation of proposed circuit designs.