This research explores how advanced reasoning-enhanced LLMs can transform hardware design processes by automating High-Level Synthesis (HLS) optimization tasks traditionally requiring expert engineers.

• Demonstrates how models like OpenAI o3-mini and DeepSeek-R1 can understand hardware constraints and suggest optimizations
• Proposes a novel approach where LLMs act as agentic assistants in the iterative HLS process
• Evaluates these models' abilities to balance performance requirements with resource constraints
• Introduces a framework for automating pragma/directive decisions that typically require manual engineering expertise

This research represents a significant step toward automated hardware design processes that could dramatically reduce development time and costs while potentially discovering optimizations human engineers might overlook.

Can Reasoning Models Reason about Hardware? An Agentic HLS Perspective