ML-Based Clustering for Clock Tree Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional clock tree synthesis techniques for integrated circuits are computationally expensive and time-consuming due to the need for multiple cluster evaluations to ensure design rule constraints are met, especially when dealing with a large number of sinks.
Innovation Solution
A machine-learning model is trained to predict whether clusters satisfy design rule constraints, reducing the need for extensive timing analysis by evaluating clusters and iteratively adjusting the number of clusters until all constraints are met, thereby improving computational efficiency and reducing the time required to achieve a valid clustering solution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional timing analysis is performed at each clustering iteration to evaluate design rule constraints, then clustering solution accuracy is improved, but computational time and resource consumption increase significantly
Solution Approach 1:
A machine learning model is trained in advance on historical clustering data to learn the relationship between cluster characteristics and design rule constraint satisfaction. During iterative clustering, this pre-trained model quickly predicts whether clusters satisfy constraints without performing full timing analysis, thereby maintaining accuracy while dramatically reducing computational time at each iteration
Solution Approach 2:
Instead of performing actual timing analysis on each cluster at every iteration, the patent uses a machine learning model that has learned from previous timing analysis results. The model creates a simplified copy or approximation of the timing analysis function, enabling fast predictions that preserve the essential accuracy needed for constraint verification
2Reliability
If the number of clusters is increased to satisfy design rule constraints, then clustering solution reliability is improved, but computational complexity and resource utilization increase
Solution Approach 1:
The machine learning model provides immediate feedback on whether current clusters satisfy design rule constraints by predicting constraint satisfaction based on cluster characteristics. This feedback mechanism allows the iterative clustering process to efficiently determine when to stop increasing the number of clusters, avoiding unnecessary computational complexity while ensuring reliability through constraint satisfaction
Solution Approach 2:
The patent changes the approach from performing complex timing analysis to using a machine learning model that evaluates clusters based on learned parameters and features. This parameter-based evaluation method reduces computational complexity while maintaining the ability to reliably assess whether clusters meet design constraints
Data Source
AI summary
Aspects of the present disclosure address systems and methods for performing a machine-learning based clustering of dock sinks during clock tree synthesis. An integrated circuit design comprising a clock net that includes a plurality of clock sinks is accessed. A set of clusters are generated by clustering the set of clock objects of the clock net. A machine-learning model is used to assess whether each cluster satisfies one or more design rule constraints. Based on determining each cluster in the set of dusters is assessed by the machine-learning model to satisfy the one or more design rule constraints, a timing analysis is performed to determine whether each cluster in the set of clusters satisfies the target timing constraints. A clustering solution for the clock net is generated based on the set of clusters in response to determining each cluster satisfies the one or more design rule constraints.