dac-related-work
GitHub指导DAC论文撰写相关工作与定位，强调在六页限制内以“增量声明”形式对比最强基线，覆盖EDA核心会议期刊及基准工具谱系，确保盲审合规并准确引用。
触发场景
安装
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill dac-related-work -g -y
SKILL.md
Frontmatter
{
"name": "dac-related-work",
"description": "Use when writing or auditing the related-work and positioning of an ACM\/IEEE Design Automation Conference (DAC) Research Manuscript, covering the EDA literature lanes (DAC\/ICCAD\/DATE\/ASP-DAC\/TCAD\/TODAES), delta-first positioning against the strongest prior tool, honest benchmark-suite lineage, and third-person self-citation under double-blind review within the six-page budget."
}
DAC Related Work
Position the paper against the EDA literature a DAC reviewer already knows. At six double-column pages, related work is not a survey — it is a delta statement: here is the best prior technique for this design task, here is precisely what it cannot do, here is what we add, measured in QoR. DAC reviewers are drawn from exactly the subcommittee whose prior work you are extending, so an incomplete or unfair positioning is caught immediately.
The EDA literature lanes
Cover the lanes your contribution touches; do not pad with distant ones:
- The top EDA venues — DAC, ICCAD, DATE, ASP-DAC — the conference record of your problem. A physical-design paper that cites no recent DAC/ICCAD placement work reads as disconnected.
- The EDA journals — IEEE TCAD (Transactions on CAD) and ACM TODAES — where the extended and most authoritative versions of techniques live; cite the journal version when it is the canonical reference.
- Adjacent communities when relevant — ISCA/MICRO/HPCA for architecture-facing work, ITC/VTS for test, ISSCC/CICC for circuits, and the ML venues (NeurIPS/ICML/ICLR) only when your ML-for-EDA method genuinely builds on a learning advance.
- The benchmark and tool lineage — the paper introducing the suite you evaluate on (ISPD contests, EPFL benchmarks, ISCAS/ITC circuits, TAU contest, CircuitNet) and the open flow you use (OpenROAD). Miscrediting a benchmark's origin is an easy, avoidable reviewer irritant.
Delta-first positioning
For each closely related technique, write the comparison in one move: what it does → its specific limitation → your advance.
- Weak: "Prior work has studied placement extensively [3-15]." (A citation dump; it positions nothing.)
- Strong: "The analytic placer [8] minimizes wirelength but degrades on macro-heavy designs because it treats macros as movable cells; we handle macros with [mechanism], cutting congestion by X% on the macro-dominated ISPD circuits."
The reviewer's implicit question is always "why is this not just [the obvious prior tool] plus tuning?" Answer it explicitly, early, and in QoR terms.
Name the strongest baseline as the thing to beat
DAC papers are won and lost on the baseline. In related work, identify the current state of the art for your exact task and commit to comparing against it in the evaluation. Comparing only to an old or weak method — while a stronger, well-known tool exists and goes uncited — is the fastest way to a reject, because the assigned reviewer very likely authored or uses that stronger tool.
Honest benchmark and reproducibility lineage
- Credit the benchmark suite correctly and use its standard metrics; reporting a nonstandard metric on a standard suite invites "why not the usual QoR numbers?"
- If you use an open-source flow (OpenROAD, ABC, Yosys) or a public dataset (CircuitNet), cite it and state the version — this both positions you honestly and helps reproducibility.
- Distinguish contest results (ISPD/TAU/CAD-contest rankings) from peer-reviewed papers when you invoke them as prior art.
Third-person self-citation under double-blind
Cite your own prior papers as if written by others: "Prior work [7] proposed..." not "in our previous paper [7] we proposed...". A first-person self-citation de-anonymizes the submission. If your new paper is an extension of your own DAC/ICCAD paper, position the delta in the third person and let the camera-ready restore ownership.
Fitting related work into six pages
- No standalone two-column "Section 2: Related Work" wall if space is tight — weave positioning into the introduction and method, citing at the point of contrast.
- Group distant references (a whole subfield you merely acknowledge) into a single sentence with a bundled citation; spend individual sentences only on the techniques you actually beat or build on.
Output format
[Lanes covered] DAC/ICCAD/DATE/ASP-DAC + TCAD/TODAES + benchmark/tool lineage present?
[Strongest baseline] named and committed to in the evaluation? yes/no
[Delta statements] each close technique -> its limitation -> your QoR advance? yes/no
[Benchmark credit] suite + version + standard metrics correctly attributed? yes/no
[Anonymity] self-citations third-person? yes/no
[Space] positioning woven to fit the six-page budget? yes/no
版本历史
- 9f86f09 当前 2026-07-19 15:12
