NeuroSteiner: A Graph Transformer for Wirelength Estimation
This addresses wirelength estimation for chip design, offering a faster alternative to optimal solvers, but it is incremental as it builds on existing methods.
The paper tackles the NP-hard problem of estimating wirelength in chip placement by proposing NeuroSteiner, a neural model that distills an optimal solver, achieving 0.2-0.3% error while being 30-60% faster.
A core objective of physical design is to minimize wirelength (WL) when placing chip components on a canvas. Computing the minimal WL of a placement requires finding rectilinear Steiner minimum trees (RSMTs), an NP-hard problem. We propose NeuroSteiner, a neural model that distills GeoSteiner, an optimal RSMT solver, to navigate the cost--accuracy frontier of WL estimation. NeuroSteiner is trained on synthesized nets labeled by GeoSteiner, alleviating the need to train on real chip designs. Moreover, NeuroSteiner's differentiability allows to place by minimizing WL through gradient descent. On ISPD 2005 and 2019, NeuroSteiner can obtain 0.3% WL error while being 60% faster than GeoSteiner, or 0.2% and 30%.