A Graph-based Benchmark dataset for Printed Circuit Netlist Partitioning

Overview

This work addresses the absence of high-quality benchmark datasets for printed circuit netlist partitioning by introducing BenchPCNP, a labeled dataset constructed from production-verified circuits. The dataset comprises netlist graph data in Protel 2 format partitioned according to IPC-2612 design standards. Netlist partitioning represents a fundamental task in electronic design automation and reverse engineering contexts, with potential applications in graph machine learning methodologies.

Methods and approach

The dataset construction methodology involved five electronic designers partitioning 50 production-verified practical circuits following IPC-2612 design standards. Netlist graphs were extracted in Protel 2 format and annotated with partition labels. The resulting dataset encompasses 54 distinct partition module labels with explicit module partitioning annotations. This structured annotation process establishes a reliable foundation for benchmarking partitioning algorithms and machine learning models.

Key Findings

BenchPCNP contains 50 partitioned circuits derived from production-verified designs with consistent labeling across 54 distinct module categories. The dataset incorporates manual partitioning performed by multiple domain experts, ensuring alignment with established design standards and practical circuit engineering conventions. The labeled graphs enable quantitative performance evaluation for algorithmic approaches to netlist partitioning.

Implications

The availability of BenchPCNP establishes standardized evaluation infrastructure for netlist partitioning research, facilitating comparison across different algorithmic approaches and graph machine learning techniques. This dataset infrastructure promotes research consistency and enables reproducible benchmarking within the domain. The structured annotation with expert partitioning decisions provides empirical grounding for developing and validating netlist partitioning methodologies.