Discriminant Function Module for IC Configuration Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-configurability integrated circuit (IC) designs, such as configurable network-on-chip (NoC) designs, the large number of possible legal configurations makes it difficult for designers to optimize metrics like area, timing, and power consumption, as existing tools often produce inconsistent results with minor changes in constraints, leading to costly rework.
Innovation Solution
A system comprising a constraint solver module and a discriminant function module that automatically identifies optimal legal configurations by enforcing rules and selecting parameters from user-defined requirements, using machine learning to reduce the time needed to generate optimized RTL descriptions and ensure consistent results across varying constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the configurability of the hardware function is increased to address many different needs, then the adaptability improves, but the device complexity increases making it difficult to find optimal configurations
Solution Approach 1:
The system enables self-service by automatically generating and evaluating configurations without manual intervention. The automated evaluation module systematically explores the configuration space, evaluates performance metrics, and identifies optimal configurations, eliminating the need for designers to manually navigate the complex parameter space.
Solution Approach 2:
The system applies parameter changes by systematically varying configuration parameters within defined ranges and evaluating their impact on performance metrics. This structured approach to parameter exploration allows the system to navigate the complex configuration space efficiently and identify optimal settings for adaptability, area, timing, and power consumption.
2Adaptability or versatility
If the number of parameters is increased to provide more configuration options, then the adaptability improves, but the quantity of legal configurations becomes extremely large making optimization difficult
Solution Approach 1:
The system applies segmentation by dividing the large configuration space into manageable segments or regions. The automated evaluation module systematically explores different segments of the parameter space, evaluating configurations in each segment and comparing results, which makes the optimization process tractable despite the large total number of configurations.
Solution Approach 2:
The system applies partial action by evaluating a representative subset of configurations rather than exhaustively analyzing every possible legal configuration. The automated evaluation module strategically samples the configuration space to identify optimal settings without requiring complete enumeration of all configurations, significantly reducing the computational burden.
3Productivity
If existing tools are used to generate RTL descriptions, then the productivity is maintained, but the consistency of results deteriorates with minor changes in constraints leading to costly rework
Solution Approach 1:
The system applies feedback by implementing an automated evaluation module that systematically evaluates generated configurations against specified criteria and constraints. This feedback mechanism ensures that configurations meeting the required standards are identified consistently, and any deviations are detected and corrected, maintaining result stability even when constraints are modified.
Solution Approach 2:
The system applies preliminary action by performing automated evaluation and validation of configurations before final RTL generation. This preliminary assessment ensures that only configurations meeting all criteria are selected, preventing inconsistent results and reducing the need for rework when constraints change.
Data Source
AI summary
A tool is disclosed that includes a discriminant module. The discriminant module finds one configuration, which is selected from many different possible and legal configurations, that is optimal. The optimal configuration is translated into a set of optimized parameters (identified from the library of parameters that the user can select from) and provided to the designer. The designer reviews (and can manually revise or change) the optimized parameters. The optimized parameters are translated into engineering parameters. The engineering parameters are passed, as an input, to the RTL generation module. The RTL generation module produces the RTL description of the hardware function that is optimal and meets the designer's defined requirements.