Visual Exploratory Analysis for Designing Large-Scale Network-on-Chip Architectures: A Domain Expert-Led Design Study

Area of Science:

• Computer Science
• Data Visualization
• Computer Architecture

Background:

• Traditional visualization design studies are led by visualization researchers, potentially limiting domain expert input.
• Domain experts, such as computer architects, face challenges in analyzing complex chip designs and Network-on-Chip (NOC) performance.
• Effective visualization tools are crucial for understanding the relationship between chip design and performance.

Purpose of the Study:

• To explore an alternative design study model where domain experts lead the process, supported by visualization experts.
• To develop Vis4Mesh, a visualization system tailored for computer architects to analyze simulated Network-on-Chip (NOC) behavior.
• To provide spatial, temporal, and architectural context for NOC performance analysis.

Main Methods:

• A domain expert-led design study involving computer architecture experts.
• Development of the Vis4Mesh visualization system.
• Integration of Vis4Mesh with an existing computer architecture visualization tool.
• Validation through a case study and a user study with computer architecture researchers.

Main Results:

• Vis4Mesh effectively provides spatial, temporal, and architectural context for NOC behavior analysis.
• Integration with existing tools allows for deep-dives into specific architectural components.
• The domain expert-led approach facilitated a more synchronized and effective design process.
• User study confirmed the utility and effectiveness of Vis4Mesh for computer architects.

Conclusions:

• Domain expert-led design studies can be a viable and effective model for developing specialized visualization tools.
• Vis4Mesh enhances the analysis of computer chip designs by providing critical context for NOC performance.
• The study offers insights and guidance for future domain expert-led visualization design initiatives.