GAP-LA: GPU-Accelerated Performance-Driven Layer Assignment

Abstract

Layer assignment is critical in global routing because different metal layers have different resistance and capacitance, which directly affects congestion, timing, and power. This paper proposes GAP-LA, a GPU-accelerated performance-driven layer assignment framework for holistic optimization of these objectives. The method achieves improved WNS and TNS with competitive runtime on large-scale designs, and further improves timing when applied to topologies from ISPD 2025 contest-winning flows.