TDPNavigator-Placer: Thermal- and Wirelength-Aware Chiplet Placement in 2.5D Systems Through Multi-Agent Reinforcement Learning

2026-02-13 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Semiconductor Design · Depth: Expert, quick

Summary

TDPNavigator-Placer is a new multi-agent reinforcement learning framework designed for automated chiplet placement in 2.5D integrated circuits. This system addresses the inherent conflict between minimizing wirelength and managing thermal performance, which traditional methods often struggle with due to their single-objective or weighted-sum optimization approaches. TDPNavigator-Placer assigns these competing objectives to specialized agents, each with distinct reward mechanisms and environmental constraints, operating within a unified placement paradigm. The framework dynamically optimizes placement based on a chiplet's thermal design power (TDP). Experimental results indicate that TDPNavigator-Placer significantly improves the Pareto front compared to existing methods, allowing for more balanced trade-offs between wirelength and thermal performance in complex chiplet assemblies.

Key takeaway

For AI Scientists designing 2.5D integrated circuits, TDPNavigator-Placer offers a robust solution to the long-standing challenge of balancing wirelength and thermal management. You should consider adopting multi-agent reinforcement learning frameworks to handle inherently conflicting design objectives, moving beyond traditional single-objective or weighted-sum approaches. This can lead to superior Pareto fronts and more practical, balanced chiplet placements.

Key insights

Multi-agent reinforcement learning can resolve conflicting objectives in 2.5D chiplet placement.

Principles

• Conflicting objectives require specialized agents.
• Dynamic optimization improves placement trade-offs.

Method

TDPNavigator-Placer uses multi-agent reinforcement learning, assigning wirelength and thermal management to distinct agents with unique reward mechanisms and environmental constraints for dynamic chiplet placement.

In practice

• Optimize 2.5D IC layouts for thermal and wirelength.
• Balance design requirements in heterogeneous chiplets.

Topics

• Chiplet Placement
• 2.5D Integrated Circuits
• Multi-Agent Reinforcement Learning
• Thermal Management
• Wirelength Optimization

Best for: AI Scientist, AI Researcher, AI Engineer, Research Scientist

Related on AIssential

• Trace2Skill: Verifier-Guided Skill Evolution for Long-Context EDA Agents — Evolve natural-language agent skills with verifier feedback to solve complex hardware design problems without model fine-tuning.
• AI Is Designing Radio Chips That Humans Couldn’t Even Imagine — AI-driven inverse design and reinforcement learning can autonomously create high-performance RFICs faster than human experts.
• Order Matters: Unveiling the Hidden Impact of Macro Placement Sequences via Proxy-Guided LLM Evolution — Macro placement sequence is a critical, often overlooked, factor in chip design optimization.
• CADO: From Imitation to Cost Minimization for Heatmap-based Solvers in Combinatorial Optimization — Objective mismatch in heatmap-based CO solvers is resolved by RL fine-tuning that directly optimizes post-decoded solution cost.
• ChipMATE: Multi-Agent Training via Reinforcement Learning for Enhanced RTL Generation — ChipMATE is a self-trained multi-agent framework for RTL generation using mutual verification without a golden oracle.
• Tool-Aware Optimization with Entropy Guidance for Efficient Agentic Reinforcement Learning — TAO-RL stabilizes agentic reinforcement learning by filtering tool-use trajectories and guiding exploration with entropy for efficient LLM policy optimization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.