Differentiable Global Router for Concurrent Circuit Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional global routers either operate sequentially, leading to suboptimal routing solutions for all nets in a circuit, or incur high computational costs when optimizing multiple nets concurrently.
Innovation Solution
A differentiable global router that utilizes a routing DAG forest structure to make coordinated selections of DAGs and edges, enabling concurrent routing for millions of nets by transforming the search space into a continuous and scalable domain using gradient algorithms and deep learning tools like PyTorch.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional global routers operate sequentially on each net, then the routing process is simple and fast, but the routing solution quality becomes suboptimal
Solution Approach 1:
The patent merges multiple sequential routing operations into a single concurrent optimization process. By formulating the routing problem as a differentiable optimization task that simultaneously considers all nets, the system achieves both high speed and high quality routing solutions, resolving the trade-off between sequential simplicity and concurrent quality.
Solution Approach 2:
The patent transforms the discrete routing selection problem into a continuous differentiable optimization problem. By using differentiable relaxation techniques and continuous parameters to represent routing decisions, the system enables gradient-based optimization that simultaneously optimizes all nets while maintaining computational efficiency.
2Manufacturing precision
If combinatorial optimization-based global routers concurrently optimize multiple nets, then routing solution quality improves, but computational costs become high
Solution Approach 1:
The patent replaces traditional combinatorial optimization methods with differentiable optimization techniques. By substituting discrete combinatorial search with continuous differentiable functions and gradient-based optimization, the system achieves high routing quality while significantly reducing computational cost and energy consumption.
Solution Approach 2:
The patent changes the optimization approach from discrete combinatorial parameters to continuous differentiable parameters. This transformation enables the use of efficient gradient-based optimizers that can handle large-scale routing problems with millions of nets, reducing both computational cost and time requirements.
3Measurement precision
If the routing search space is discrete and combinatorial, then routing decisions are precise, but scalability to millions of nets becomes difficult
Solution Approach 1:
The patent transforms the discrete routing decision space into a continuous differentiable space. By representing routing decisions through continuous parameters and differentiable functions, the system achieves both precise routing decisions and scalability to millions of nets, as continuous optimization is inherently more efficient than combinatorial search.
Solution Approach 2:
The patent adds a differentiable dimension to the routing problem formulation. By introducing continuous relaxation layers and differentiable optimization objectives, the system operates in an extended parameter space that enables both precise discrete routing decisions and efficient large-scale optimization.
Data Source
AI summary
Mechanisms for generating metal routing guides in a circuit involve forming a plurality of directed acyclic graphs (DAGs) embodying routing trees for nets in the circuit, generating 2-pin subnets and 2-pin path candidates from the routing trees, and forming a DAG forest from the routing trees, the 2-pin subnets, and the 2-pin path candidates.