Deep Learning Routing Prediction for EDA Timing Closure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In electronic design automation, there is a conflict between power, performance, and area (PPA) and design turnaround time (TAT) at advanced nodes, where existing tools often require trade-offs and struggle to predict post-route timing and design-rule-check (DRC) accurately, making it costly to estimate detail route timing and DRC at the pre-route stage.
Innovation Solution
A computer-implemented method using a deep neural network to transform and predict routing information, including wire length and via number, by receiving global route data, generating detail route data, and training the neural network with both types of data to generate routing information for each routing grid, addressing the challenge of predicting post-route RC parasitic parameters and congestion from the pre-route stage to the detail route stage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional routing tools are used to ensure accurate post-route timing and DRC prediction, then manufacturing precision is improved, but productivity deteriorates due to time-consuming detail route calculation
Solution Approach 1:
The patent applies preliminary action by performing detail route calculations and training the neural network model in advance during an offline training phase. The model learns from pre-computed routing data including wire lengths, via counts, and timing/DRC outcomes. During actual routing operations, the pre-trained model quickly predicts results without performing full detail route calculations, thus achieving both accuracy and speed.
2Productivity
If deep neural network prediction is used to reduce runtime, then productivity is improved, but measurement precision deteriorates without accurate prediction of post-route timing and DRC
Solution Approach 1:
The patent implements feedback by using actual post-route timing and DRC results from traditionally routed designs to train and refine the neural network model. The model learns from the feedback loop where predicted values are compared against ground truth measurements, and the training process adjusts weights to minimize prediction errors, thereby improving accuracy while maintaining fast inference speed.
Solution Approach 2:
The patent substitutes the mechanical/computational routing calculation system with a neural network-based prediction system. Instead of performing detailed physical routing calculations to determine wire lengths and via counts, the system uses a trained neural network that has learned the underlying patterns from training data, replacing the traditional mechanical routing process with an intelligent prediction mechanism.
3Reliability
If detailed route data is calculated to improve prediction accuracy, then reliability is improved, but loss of time increases due to extensive computation required
Solution Approach 1:
The patent applies preliminary action by pre-calculating detailed route data during an offline training phase and storing it as training samples. The neural network model learns from this pre-computed data, so during actual routing operations, the system only needs to perform lightweight prediction operations rather than full detail route calculations, significantly reducing computation time while maintaining reliability.
Data Source
AI summary
The present disclosure relates to a computer-implemented method for routing in an electronic design. Embodiments may include receiving, using at least one processor, global route data associated with an electronic design as an input and generating detail route data, based upon, at least in part, the global route data. Embodiments may further include transforming one or more of the detail route data and the global route data into at least one input feature and at least one output result of a deep neural network. Embodiments may also include training the deep neural network with the global route data and the detail route data and predicting an output associated with a detail route based upon, at least in part, a trained deep neural network model. Embodiments may also include generating routing information for each routing grid.