ICAN: Attention-Based Knowledge Transfer for Microarchitecture Search

Official implementation of "Seeing Through Designs: Attention-Based Knowledge Transfer for Preference-Guided Microarchitecture Search".

Overview

ICAN enables cross-architecture knowledge transfer for microarchitecture design space exploration using:

• Graph Attention Networks for CPU architecture modeling
• Cross-attention fusion for architecture-EDA tool interactions
• Meta-learning for rapid adaptation to new architectures
• Uncertainty-guided optimization with user preferences