IC DTCO Transfer Learning for Faster PPA Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The DTCO process for integrated circuits (ICs) is time-consuming and costly due to the difficulty in reflecting design information for optimizing performance, power, and area (PPA), and producing process design kits (PDKs based on input parameters.
Innovation Solution
An IC design system using Gaussian processes and Bayesian optimization for transfer learning to predict and optimize PPA by learning correlations between source and target designs, facilitating the construction of optimized input parameter combinations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional DTCO process is used to optimize PPA of IC, then optimization accuracy is improved, but process time increases considerably
Solution Approach 1:
The system performs preliminary actions by pre-processing source design data and constructing a Gaussian process model in advance. Transfer learning is performed beforehand to establish correlations between source and target designs, so that when actual PPA optimization is needed, the model is already prepared and can make predictions quickly without requiring extensive computation during the optimization process itself.
Solution Approach 2:
The system creates a virtual copy of the design space through Gaussian process modeling. Instead of performing exhaustive simulations on actual target designs, the system uses transfer learning to copy knowledge from source designs and creates a predictive model that replicates the behavior of expensive simulations, enabling fast evaluation of multiple design points.
2Reliability
If comprehensive design information is reflected in DTCO process, then optimization quality is improved, but computational complexity increases
Solution Approach 1:
The system changes parameters by transforming the complex multi-dimensional design space into a simplified probabilistic model. The Gaussian process model uses kernel functions to capture correlations between designs, transforming complex simulation data into manageable probability distributions that can be efficiently queried for optimization without requiring exhaustive analysis of all design parameters.
Solution Approach 2:
The Gaussian process model serves as an intermediary between source design data and target design optimization. Transfer learning acts as a mediator that transfers knowledge from source to target designs, enabling the system to handle comprehensive design information indirectly through the probabilistic model rather than directly through complex simulations.
3Measurement precision
If multiple source designs are used for transfer learning, then prediction accuracy is improved, but data processing time increases
Solution Approach 1:
The system merges multiple source design datasets into a unified Gaussian process model. The transfer learning framework combines information from multiple source designs by learning their correlations with the target design simultaneously, consolidating multiple data sources into a single predictive model that leverages all available information without requiring separate processing for each source design.
Data Source
AI summary
An example integrated circuit (IC) design system includes a processor, a storage device, and a design technology co-optimization (DTCO) framework. The storage device is configured to store input parameters and performance, power, and area (PPA) of a plurality of source designs and a target design corresponding to the input parameters as a dataset. The DTCO framework, implemented as software and performed by the processor, is configured to perform a first transfer learning that learns first correlations between the target design and each of the plurality of source designs based on the dataset, and to perform a second transfer learning that learns second correlations between the target design and the plurality of source designs based on the first transfer learning.