Invited: Toward Sustainable and Transparent Benchmarking for Academic Physical Design Research Conference: Proceedings of the 2026 International Symposium on Physical Design; March 15–18, 2026; Bonn, Germany Proceedings of the 2026 International Symposium on Physical Design (ISPD ’26), March 15–18,
Liwen Jiang, Andrew B. Kahng, Zhiang Wang*, Zhiyu Zheng · 2026
Read Invited: Toward Sustainable and Transparent Benchmarking for Academic Physical Design Research Conference: Proceedings of the 2026 International Symposium on Physical Design; March 15–18, 2026; Bonn, Germany Proceedings of the 2026 International Symposium on Physical Design (ISPD ’26), March 15–18, by Liwen Jiang, Andrew B. Kahng, Zhiang Wang*, Zhiyu Zheng (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-17520.txt.
Introduction
Liwen Jiang and four co-authors present RosettaStone 2.0 for chip design research. Physical design turns logic plans into chip layouts. Fair benchmarks are key to judge new tools. RosettaStone 2.0 gives full RTL-to-GDS reference flows. It covers standard flat designs. It also covers face-to-face hybrid-bonded 3D designs in the Pin-3D style.
Open flows let labs compare tools on the same ground. Without them, each paper's claims stand alone. The framework stresses sustainable and transparent practice. Chip research has long faced a reproducibility gap. Shared flows narrow that gap. They also let small teams test ideas at full scale. The work serves both 2D and 3D design camps.
Catalog record
| CREATOR | Liwen Jiang, Andrew B. Kahng, Zhiang Wang*, Zhiyu Zheng |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-17520
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Invited: Toward Sustainable and Transparent Benchmarking for Academic Physical Design Research Conference: Proceedings of the 2026 International Symposium on Physical Design; March 15–18, 2026; Bonn, Germany Proceedings of the 2026 International Symposium on Physical Design (ISPD ’26), March 15–18, 2026, Bonn, Germany DOI: 10.1145/3764386.3779611 ISBN: 979-8-4007-2314-8/2026/03 CCS: Hardware Electronic design automation Liwen Jiang, Andrew B. Kahng, Zhiang Wang*, Zhiyu Zheng Abstract . This paper presents RosettaStone 2.0, an open benchmark translation and evaluation framework built on OpenROAD-Research [ 1 ] . RosettaStone 2.0 provides complete RTL-to-GDS reference flows for both conventional 2D designs and Pin-3D-style face-to-face (F2F) hybrid-bonded 3D designs, enabling rigorous apples-to-apples comparison across planar and three-dimensional implementation settings. The framework is integrated within OpenROAD-flow-scripts (ORFS)-Research [ 2 ] ; it incorporates continuous integration (CI)-based regression testing and provides a standardized evaluation pipeline based on the METRICS2.1 convention, with structured logs and reports generated by ORFS-Research. To support transparent and reproducible research, RosettaStone 2.0 further provides a community-facing leaderboard, which is governed by verified pull requests and enforced through Developer Certificate of Origin (DCO) compliance. † † cc-license: by 1. Introduction The advancement of 2D and 3D VLSI physical design (PD) research is inherently dependent on rigorous benchmarking frameworks built upon complete flows, realistic benchmarks, and reproducible methodologies. Academic contests have historically provided benchmarks that drive innovation in physical design [ 3 , 4 ] . However, these benchmarks usually lack the full set of inputs required by modern RTL-to-GDS flows; even Liberty, technology LEFs, parasitic models, and signoff constraints are often missing or only partially specified. Moreover, different publications may use different libraries, timing corners, or routing rules under the same testcase name, and scripts are frequently tied to specific tools or versions. As a result, reported quality-of-results (QoR) and runtime improvements are difficult to compare in an apples-to-apples way, especially across independent works. Recent perspectives argue that the next physical design roadmap inflection is driven by fine-grain heterogeneous 3D integration [ 5 ] , where bonding and backside processing make it practical to stack multiple active tiers and tailor per-tier device and interconnect choices to the function being implemented [ 6 ] . Despite strong interest in fine-grain 3D integration, there is still no widely adopted open-source RTL-to-GDS reference flow for Pin-3D-style face-to-face (F2F) hybrid bonding [ 7 , 8 ] , with maintainable 3D enablements and reproducible stage-wise checkpoints. Hence, the integration, fair comparison, and reproducibility of emerging 3D physical design algorithms across different tool versions and Process Design Kit (PDK) variants remain challenging. Reproducible evaluation must increasingly encompass both planar and 3D settings, while enabling more rigorous control not only of algorithmic choices, but of tool enablements and measurement protocols as well. Several 3D integration styles have been explored. Through-silicon via (TSV) based stacking is well-established in commercial products such as high-bandwidth memory, but TSVs with 10 \mu m-scale pitch consume area and add large parasitics [ 9 , 10 ] . Monolithic 3D (M3D) uses nanoscale monolithic inter-tier vias and sequential low-temperature processing [ 11 ] to achieve very high integration density, but remains costly and less mature [ 12 ] . By contrast, face-to-face hybrid bonding bonds two pre-fabricated dies using hybrid bonding terminals (HBTs) in the back-end-of-line (BEOL), with an approximately 1 \mu m-scale pitch and strong manufacturing support [ 13 , 14 ] . Recent products such as AMD’s Zen 4 and Meta’s AR SoC adopt F2F hybrid bonding [ 15 , 16 ] , which makes F2F-based 3D integration a realistic and attractive target for academic 3D PD flows. Recent academic research works have started to address challenges of reproducibility and standardization for 3D physical design. Open-3DBench [ 17 ] provides an open 3D-IC backend benchmarking framework built on OpenROAD-flow-scripts [ 18 ] , covering partitioning, placement, routing, extraction, and thermal analysis across 3D flows. Meanwhile, the IEEE CEDA DATC Robust Design Flow (RDF) [ 19 ] and the ML EDA Commons [ 20 ] seek to advance shared benchmarks, metrics, and evaluation infrastructures based on OpenROAD [ 21 ] and ORFS, with a strong focus on artifact evaluation and sustainable research backplanes. Despite such efforts, a durable and open-source RTL-to-GDS reference flow for Pin-3D-style F2F hybrid bonding – with maintainable enablements and standardized evaluation checkpoints – remains unavailable. To advance the availability of fair and reproducible assessments for 3D PD flows, we make the following contributions. • Sustainable benchmarking backplane. We establish a maintained infrastructure within OpenROAD- and ORFS-Research, featuring co-versioned artifacts, CI-based regression, METRICS2.1-compatible reporting, and community-driven governance. • Pin-3D reference flow. We release an end-to-end F2F 3D RTL-to-GDS reference flow, complete with 3D enablements (e.g., HBT modeling, tier libraries) and standardized checkpoints for reproducible evaluation. • RosettaStone 2.0 roadmap. We outline the extension of this backplane to support systematic benchmark translation, synthetic netlist generation, and unified 2D and 3D validation under explicit evaluation contracts. In the following, Section 2 describes prospects for OpenROAD-Research and ORFS-Research, and introduces RosettaStone 2.0 as a sustainable benchmarking backplane supporting reproducible 2D and Pin-3D physical design evaluation. Section 3 presents our initial Pin-3D enablement and RTL-to-GDS reference flow for F2F hybrid bonding. Section 4 reports example validation results and sensitivity analyses under a consistent evaluation contract. Section 5 outlines the roadmap toward the complete RosettaStone 2.0 framework, and we conclude in Section 6 . 2. Sustainable Benchmarking Backplane Based on OpenROAD-Research and ORFS-Research OpenROAD-Research [ 1 ] and ORFS-Research [ 2 ] are designed as open and collaborative platforms to guide future efforts that will strengthen open EDA infrastructure and community collaboration. They preserve algorithmic diversity by hosting academically validated but potentially non-production code, thus lowering the barrier to experimentation in emerging areas such as ML-assisted optimization and agentic flows; supporting education through reproducible implementations; and fostering global collaboration via contrib-style repositories. In contrast to the mainline repositories [ 21 ] [ 18 ] , which focus on industrial-grade stability, the research backplanes prioritize transparency, reproducibility, and extensibility, making them a natural substrate for sustainable benchmarking. Building on this backplane, we release the Pin-3D RTL-to-GDS reference flow as a set of versioned sub-tools within ORFS-Research. The Pin-3D flow provides an open, reproducible instantiation of fine-grained 3D physical design for face-to-face hybrid bonding, including curated flow scripts, maintainable 3D enablements, and METRICS2.1-compatible structured logging. The structured evaluation kernel enables consistent measurement of wirelength, routing congestion, total negative slack, design-rule violations, and runtime across all stages, allowing direct end-to-end comparison between alternative placement engines and the native OpenROAD-Research flow under identical downstream conditions. While OpenROAD-Research and ORFS-Research provide the execution substrate, systematic benchmarking further requires robust mechanisms for benchmark translation, normalization, and coverage expansion. To this end, RosettaStone 2.0 provides an extended backplane that enables systematic translation of legacy academic benchmarks, integration of synthetic benchmark generation, and unified validation of both 2D and 3D flows under explicit evaluation contracts. These contracts define stage boundaries, exported artifacts, and measurement semantics, ensuring that results remain comparable and reproducible across tool versions, PDKs, and stack variants. Figure 1 . RosettaStone 2.0 roadmap (vision) to sustain and expand the open-source benchmarking backplane. Not all planned components are fully validated in this paper. We frame RosettaStone 2.0 (Figure 1 ) as an advance over RosettaStone 1.0 [ 22 ]