P/G Via Removal Using ML for IR Drop and Routing Congestion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In integrated circuit (IC) design, the IR voltage drop due to conductor resistance leads to reduced supply voltage, causing performance degradation and requiring over-design of the power/ground (P/G) network, which decreases routing resources and increases redesign time.
Innovation Solution
A machine-learning (ML) model is trained using images of P/G vias and wires to estimate IR drop distribution and routing congestion, optimizing the P/G network by modifying vias and wires to balance routing resources and IR drop, employing a multitask learning approach for efficient design.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the P/G network is over-designed to compensate for IR voltage drop, then supply voltage stability is improved, but routing resources are decreased and design complexity increases
Solution Approach 1:
The ML model performs preliminary analysis of IR drop and routing congestion during the design planning stage, enabling proactive optimization of the P/G network before finalization. This allows the design to achieve adequate voltage stability without excessive over-design, thereby reducing overall network complexity
Solution Approach 2:
Traditional manual or rule-based P/G network optimization methods are replaced with a machine-learning-based system that automatically analyzes IR drop and routing congestion patterns. This substitution enables more efficient optimization that simultaneously considers multiple constraints, reducing the need for conservative over-design
2Reliability
If the P/G network is over-designed to compensate for IR voltage drop, then supply voltage stability is improved, but routing resources are decreased
Solution Approach 1:
The ML model performs preliminary analysis of IR drop and routing congestion during the design planning stage, enabling proactive optimization of the P/G network before finalization. This allows the design to achieve adequate voltage stability without excessive over-design, thereby preserving routing resources for other uses
Solution Approach 2:
The system dynamically adjusts P/G network parameters (via locations, wire routes, conductor widths) based on ML-predicted IR drop and congestion patterns. This enables precise optimization that maintains voltage stability only where needed, freeing up routing resources in less critical areas
3Measurement precision
If traditional methods are used to analyze and modify P/G networks, then design accuracy is maintained, but runtime is increased and productivity is decreased
Solution Approach 1:
Traditional iterative simulation and manual analysis methods are replaced with a pre-trained ML model that rapidly predicts IR drop and routing congestion. This substitution maintains sufficient accuracy for design decisions while reducing analysis runtime by orders of magnitude, thereby improving overall design productivity
Solution Approach 2:
The ML model is pre-trained on comprehensive P/G network data beforehand, enabling it to perform rapid predictions during the actual design process. This preliminary preparation allows the system to maintain high accuracy in IR drop analysis while operating at speeds that dramatically improve design productivity
4Manufacturing precision
If multiple redesign iterations are performed to optimize P/G network, then design quality is improved, but time consumption is increased
Solution Approach 1:
The ML model performs preliminary optimization predictions early in the design process, identifying the most effective P/G network modifications before detailed implementation. This upfront guidance reduces the number of iterative redesign cycles needed, maintaining high optimization quality while minimizing time consumption
Solution Approach 2:
The system uses ML predictions of IR drop and routing congestion as feedback to guide P/G network modifications. This feedback mechanism enables more accurate first-pass optimizations, reducing the need for multiple redesign iterations and the associated time losses
Data Source
AI summary
A method, a system, and non-transitory computer readable medium for power and ground (P/G) routing for an integrated circuit (IC) design are provided. The method includes generating input features for a machine-learning (ML) model based on IR drop and routing congestion analysis for a P/G network for the IC design, and modifying a set of P/G vias or a set of P/G wires in the P/G network according to modifications identified by the ML model. The ML model comprises a feature extractor pre-trained using a plurality of images of P/G vias and P/G wires.