Hardware Component Modeling via Feature Extraction Logic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for modeling observable properties of hardware components, such as power consumption, are either inaccurate or complex, and statistical models lack suitable input stimulus to generate reliable values for cycle-by-cycle variations.
Innovation Solution
A system comprising a component model and a statistical model, where feature extraction logic extends the component model to output features identifying execution behavior, enabling the statistical model to generate values for observable properties like power consumption by inferring behavior from execution characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a detailed lower level model (e.g., netlist representation) is used to model hardware components, then simulation accuracy is improved, but execution speed deteriorates and development cost increases
Solution Approach 1:
The patent segments the modeling process into two distinct levels: a detailed lower level model for accurate simulation and a higher level model for fast execution. The higher level model captures only the essential behavior needed for software simulation, separating detailed hardware characteristics from functional behavior. This segmentation allows each model to serve its specific purpose without compromising the other.
Solution Approach 2:
The patent extracts only the necessary behavioral characteristics from the detailed lower level model to create the higher level model. By taking out only the essential features required for software simulation (such as instruction cycle timing and major functional blocks), the higher level model achieves fast execution while maintaining sufficient accuracy for its intended purpose.
2Measurement precision
If a detailed lower level model is used to model hardware components, then simulation accuracy is improved, but development cost increases
Solution Approach 1:
The patent segments the modeling effort into two levels with different complexity requirements. The higher level model uses simplified abstractions that are easier and less costly to develop, while the lower level model retains detailed information only where necessary for accurate simulation. This segmentation reduces the overall development cost by avoiding the need to create and maintain a fully detailed model for all simulation purposes.
Solution Approach 2:
The patent extracts only the essential behavioral characteristics from the detailed model, discarding unnecessary details for the higher level model. This extraction process significantly reduces development effort and cost while maintaining sufficient accuracy for software simulation purposes, as only the critical functional behavior needs to be modeled at the higher level.
3Productivity
If statistical models are used to model observable properties like power consumption, then execution speed is improved, but measurement precision deteriorates due to lack of suitable input stimulus
Solution Approach 1:
The patent introduces an intermediary component that extracts relevant features from the lower level model's execution behavior and uses these features as input stimulus for the statistical model. This intermediary layer bridges the gap between the detailed model and the statistical model, providing the statistical model with meaningful input data that improves its prediction accuracy while maintaining fast execution speed.
Solution Approach 2:
The patent changes the parameters fed into the statistical model by extracting specific execution behavior features from the lower level model. Instead of using generic or insufficient input data, the system transforms the detailed execution trace into relevant feature parameters that the statistical model can effectively use to predict observable properties like power consumption with high accuracy.
Data Source
AI summary
The system comprises a component model for modelling aspects of the hardware component, and feature extraction logic for extending the component model to cause the component model, when executing, to output one or more features identifying execution behavior of the component model. A statistical model is then arranged to receive the one or more features output by the component model, and to generate the output dependent on one or more features. The component model may not explicitly model features that can be used to effectively predict values of the observable property, features that a statistical model depends on may still be captured in the underlying logic and implementation of the component model. By extracting features identifying execution behavior of the component model, this can provide a suitable input to the statistical model.