brycewang-stanford DAC Writing Style
Use this while writing the body. DAC Research Manuscripts live inside a 6-page double-column budget (plus a references-only page), so the writing discipline is not "explain everything" but "prove one design-automation contribution, in the field's own QoR currency, in six dense pages." DAC reviewers are EDA experts under time pressure; they decide fast whether the paper has a real, measured advance.
The DAC first-page arc
Front-load the contribution. By the end of the first column-and-a-half a reviewer should know:
- The design problem — a concrete EDA/chip-design task (placement, timing closure, functional verification, IR-drop prediction, IP protection) that the community recognizes.
- Why the current state is inadequate — the specific limitation of the best existing tool or technique, ideally with a number (it does not scale past N cells, it leaves X% negative slack).
- The contribution as a measured delta — the new mechanism and the headline QoR result: "we reduce total negative slack by X% at equal area with Y× lower runtime versus [state of the art] on [standard benchmark]."
- Why it generalizes — that the mechanism is not a one-benchmark trick.
Do not open with "AI has transformed chip design." Open with the design problem and the number.
State the contribution as a QoR claim
The EDA community speaks in PPA (power, performance, area), plus wirelength, timing slack (WNS/TNS), routability/DRC, coverage, and runtime. A DAC contribution is strongest when phrased as a delta in those units against a named baseline:
- Weak: "Our method improves quality and is efficient."
- Strong: "On the ISPD placement benchmarks our router reduces total wirelength by X% and DRC violations by Y% while completing in Z× less runtime than [baseline]."
Every headline claim in the abstract and introduction should reappear as a per-benchmark row in a results table, not just an average.
Living in six double-column pages
The budget is the constraint that shapes the paper:
- One contribution, deep. Six pages cannot carry three loosely related ideas; pick the strongest and develop it fully. A sprawling paper reads as thin at DAC.
- Method before menagerie. Spend pages on the mechanism and the evaluation, not on a long taxonomy or a tutorial of standard EDA background the reviewers already know.
- Tables over prose for results. A compact per-benchmark table communicates more per square inch than paragraphs; DAC reviewers read tables first.
- Cut the roadmap. "Section 2 covers... Section 3 covers..." is dead weight at six pages; let section headings do that work.
- References go on the seventh page only. Body content there is a desk-reject; plan the argument to end at page six.
EDA figures that survive double-column at 9-10 pt
- Design figures (floorplans, layouts, congestion/heatmaps, netlist graphs) must be legible at column width — roughly half of a US-letter page wide. Test every figure at final size.
- Label axes with the real QoR units; a curve with no units persuades no one.
- Prefer a clear flow diagram of the technique over a dense algorithm listing when space is scarce; move full pseudocode to the body only if it earns its space.
- Color heatmaps must still read in grayscale and for color-blind reviewers.
Third-person self-citation (double-blind)
Because review is double-blind, refer to your own prior work in the third person: not "in our earlier tool [7]" but "the prior technique [7]." Keep acknowledgements, funding, and grant numbers out of the review version entirely; they return in the camera-ready.
Common DAC prose failures
| Failure | Why it costs you | Fix |
|---|---|---|
| QoR claim only as an average | Reviewers suspect a few benchmarks carry it | Per-benchmark table; report all circuits |
| Baseline unnamed or "a standard method" | Reads as evasive about the real state of the art | Name the strongest prior tool and its tuning |
| Background eats two pages | Leaves no room for the contribution | Assume EDA literacy; cut to the delta |
| Figures illegible at column width | Reviewer cannot verify the claim | Rebuild at final size, real units |
| First-person self-citation | De-anonymizes under double-blind | Third-person throughout |
Output format
[First-page arc] design problem -> inadequacy (with a number) -> QoR delta -> generality? present?
[QoR framing] headline claim stated in PPA/wirelength/slack/runtime vs a named baseline? yes/no
[Budget] body within 6 double-column pages? references-only 7th page clean?
[Figures] legible at column width, real units, grayscale-safe? yes/no
[Anonymity] self-citations third-person; acks/funding removed? yes/no
[Cut list] <lowest-value paragraphs/figures to reclaim space>