ML Predictor Cell Spreader for 7nm DRC Violation Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
At advanced process nodes, the correlation between global-route congestion maps and detailed route design rule check (DRC) violations is weak, leading to ineffective routability optimization and increased risk of design tapeout failures due to unresolved DRC violations.
Innovation Solution
A machine learning-based predictor is used to identify potential DRC hotspots, and a cell spreader engine redistributes white space around these hotspots to minimize DRC violations without affecting timing or area, improving routability through accurate prediction and targeted redistribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If global-route congestion maps are used for routability optimization, then placement optimization can be performed, but the correlation with actual DRC violations is weak leading to ineffective routability optimization
Solution Approach 1:
A machine learning predictor is introduced as an intermediary between global routing and detailed routing to translate congestion map data into accurate DRC violation predictions. The predictor learns the complex mapping relationship between congestion patterns and actual DRC violations, serving as a mediator that bridges the gap between the two stages and provides reliable guidance for placement optimization.
Solution Approach 2:
The system transforms the congestion map parameters into predicted DRC violation parameters through machine learning. By changing the representation from raw congestion data to predicted violation locations and types, the system achieves accurate correlation with actual DRC violations, enabling effective routability optimization at advanced process nodes.
2Productivity
If traditional placer-based routability optimization is used, then placement can be optimized based on congestion maps, but numerous detailed route DRC violations remain that require manual fixing
Solution Approach 1:
The system performs preliminary identification of DRC violation hotspots and applies targeted placement adjustments before detailed routing begins. By predicting potential DRC violations in advance and optimizing placement proactively, the system resolves most DRC issues automatically without requiring manual intervention, ensuring design tapeout success.
Solution Approach 2:
The machine learning predictor provides feedback about predicted DRC violations to the placer, which then adjusts placement to eliminate these violations. This closed-loop feedback mechanism enables the system to iteratively improve placement quality and resolve DRC issues automatically, significantly reducing manual fixing requirements.
3Loss of information
If global routing is performed to generate congestion maps, then routing information is obtained, but the maps mislead routability optimization engines due to poor correlation with actual DRC hotspots
Solution Approach 1:
The system replaces the traditional mechanical interpretation of congestion maps with a machine learning-based prediction system. Instead of directly using congestion map data to guide optimization, the ML model learns the underlying patterns and predicts actual DRC violations, substituting the flawed mechanical approach with an intelligent system that accurately captures the complex relationships at advanced process nodes.
Data Source
AI summary
A routability optimization engine comprising a hotspot prediction engine to predict locations of a plurality of hotspots in a circuit layout based on a machine learning system, a white space calculator to calculate white space around each of the plurality of hotspots, and a cell spreader engine to redistribute white space around each of the plurality of hotspots to improve routability of the circuit layout.