Continuous PVT Timing Prediction via Distance-Based Corner Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PVT corner analysis for integrated circuits is limited by its discrete nature, failing to provide accurate predictive coverage and accuracy for continuous operating conditions, as it relies on a finite set of trained corners.
Innovation Solution
A corner prediction system that selects a subset of trained PVT corners based on a distance-based selection criterion and uses a prediction algorithm to generate performance metric values for target corners, with a validation component modifying the algorithms to refine predictions and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If discrete PVT corner analysis is used, then the analysis is simpler and requires fewer trained corners, but the predictive coverage and accuracy for continuous operating conditions deteriorates
Solution Approach 1:
The patent transforms the discrete PVT corner parameters into a continuous parameter space by introducing distance metrics and interpolation variables. The target corner parameters are continuously adjusted based on weighted combinations of reference corners, enabling continuous prediction across the PVT space rather than discrete corner-by-corner analysis.
Solution Approach 2:
The patent introduces an intermediary prediction model that acts as a mediator between discrete reference corners and continuous target corners. This model uses distance-based weighting and interpolation algorithms to bridge the gap between discrete training data and continuous prediction requirements, achieving both accuracy and efficiency.
2Measurement precision
If more trained PVT corners are used, then the predictive coverage and accuracy improves, but the computational complexity and data requirements increase
Solution Approach 1:
The patent extracts only the essential features from the full set of PVT corners by selecting reference corners based on distance criteria. Instead of using all available trained corners, the system extracts and utilizes only those corners that are most relevant to the target prediction, reducing computational overhead while maintaining accuracy.
Solution Approach 2:
The patent segments the PVT corner space into regions of influence around each reference corner. By dividing the continuous parameter space into discrete zones and assigning target corners to the nearest reference corners, the system simplifies the prediction process while maintaining local accuracy through targeted analysis.
3Measurement precision
If distance-based selection criterion is applied, then the prediction accuracy for target corners improves, but the computational overhead for selecting corners increases
Solution Approach 1:
The patent performs preliminary actions by pre-calculating and storing distance metrics between PVT corners before the actual prediction process. Reference corners and their distances are pre-processed and organized, so that during prediction, the system only needs to retrieve and apply pre-computed values rather than calculating distances in real-time, significantly reducing computational overhead.
Data Source
AI summary
A corner prediction system applies data generated through discrete process, voltage, and temperature (PVT) corner prediction to achieve highly accurate continuous corner prediction coverage. Embodiments of the corner prediction system can be trained to generate accurate performance metric prediction for a continuous range of PVT corners within a design space given a set of available pre-trained PVT corners. The corner prediction system can address the need to provide accurate continuous timing prediction coverage of design operating conditions (represented by PVT corners) through the availability of discrete PVT corners.