introvoyz041/analog-circuits-sky130

该数据集包含130,409个基于开源SkyWater SKY130 PDK的ngspice模拟SPICE网表，这些网表统一来自三个上游来源，采用可加载的格式，并具有确定性的拓扑感知训练/验证/测试分割。设计为监督学习的语料库，用于训练模拟电路设计模型，将SPICE网表映射到性能指标。数据集涵盖49个拓扑家族，包括运算放大器、LDO、带隙参考、开关电容采样器、功率转换器、VCO、混频器、LNA、PA等。分割方式确保评估在未见过的拓扑上进行泛化，而不仅仅是参数样本。总行数包括128,226个收敛行和2,183个失败行，使用ngspice 46模拟器，参数扫描基于拉丁超立方采样。This dataset contains 130,409 ngspice-simulated SPICE netlists based on the open-source SkyWater SKY130 PDK. All these netlists are uniformly sourced from three upstream repositories, adopt a loadable format, and feature a deterministic topology-aware train/validation/test split. It is designed as a supervised learning corpus for training analog circuit design models that map SPICE netlists to performance metrics. The dataset covers 49 topology families, including operational amplifiers (op-amps), low-dropout regulators (LDOs), bandgap references, switched-capacitor samplers, power converters, voltage-controlled oscillators (VCOs), mixers, low-noise amplifiers (LNAs), power amplifiers (PAs), and others. The split strategy ensures that evaluation is conducted for generalization to unseen topologies, rather than merely parameter samples. The total number of simulation lines includes 128,226 convergent runs and 2,183 failed runs, using the ngspice 46 simulator, with parameter sweeps based on Latin Hypercube Sampling (LHS).