FloorSet

FloorSet是一个针对VLSI布局规划的数据集，由英特尔实验室美国创建，旨在解决系统芯片（SoC）及其子系统的布局规划问题。该数据集包含两部分：FloorSet-Prime和FloorSet-Lite，分别包含100万个训练样本和100个测试样本。FloorSet-Prime模拟了真实SoC的分布，包含完全接触的矩形分区，而FloorSet-Lite则反映了设计早期的简化情况，包含矩形分区，空间占用小于5%。两者都定义了现代设计流程中的硬约束，如形状约束、边缘亲和性、分组约束和预布局约束。FloorSet旨在推动大规模约束优化问题的基本研究，并解决现代ML驱动解决方案中的可重复性问题。

FloorSet is a dataset for VLSI floorplanning, created by Intel Labs USA, aiming to address the floorplanning problems of system-on-chips (SoCs) and their subsystems. This dataset consists of two subsets: FloorSet-Prime and FloorSet-Lite, which contain 1 million training samples and 100 test samples respectively. FloorSet-Prime simulates the distribution of real-world SoCs, featuring fully contacted rectangular partitions, while FloorSet-Lite reflects the simplified scenarios in early design stages, with rectangular partitions occupying less than 5% of the space. Both subsets define hard constraints in modern design flows, including shape constraints, boundary affinity, grouping constraints, and pre-placement constraints. FloorSet aims to promote fundamental research on large-scale constrained optimization problems, and address the reproducibility issues in modern ML-driven solutions.