| name | isca-related-work |
| description | Use when positioning an ISCA submission against prior art — covering five decades of architecture literature across ACM DL, IEEE Xplore, and dblp, differentiating by mechanism rather than metric, handling self-citations and arXiv preprints under double-blind rules, and exploiting ISCA's unlimited reference pages. |
ISCA Related Work
Architecture is an old field with a long memory. ISCA has run since 1973, its
committee members have often worked an area for twenty years, and the fastest way
to lose a reviewer is to present a mechanism they saw under another name in a
proceedings they attended. Positioning at this venue is therefore archaeology
first, differentiation second.
Search across three catalogs, because ISCA lives in two publishers
ISCA's dual ACM/IEEE sponsorship means its proceedings are split across publisher
libraries by edition (the 52nd, Tokyo 2025, was ACM-published; the 2026 edition
carried IEEE branding — the pattern alternates). Consequences for search:
- dblp (
dblp.org/db/conf/isca/) is the reliable spine: complete, per-edition,
publisher-neutral. Start every sweep there.
- ACM DL and IEEE Xplore each hold only part of the canon; searching one and
concluding "no prior work" is a classic self-inflicted wound.
- Sweep the sibling venues as first-class citizens — MICRO, HPCA, ASPLOS — plus
the journals where architecture ideas resurface (TACO, CAL, TC, IEEE Micro
including its annual Top Picks selections).
- Finish with a recent-window arXiv pass: concurrent preprints will not sink you,
but a reviewer mentioning one you never saw costs credibility in the rebuttal.
The lineage paragraph reviewers actually want
For the two or three mechanisms closest to yours, write lineage, not lists:
Decoupled tracking originates with [x], which [y] extended to
multi-level hierarchies at the cost of . Both retain . We remove that assumption, which is what enables <consequence §4>.
The differentiation must be structural — a different trigger, storage
organization, or contract — and it must be checkable against your design section.
"Unlike prior work, our approach is efficient" differentiates nothing; every
cited author believed the same of theirs.
Where comparison belongs
| Relationship to your work | Treatment |
|---|
| Same problem, mechanism your evaluation can implement | Quantitative baseline in the evaluation (see isca-experiments), plus one lineage sentence here |
| Same problem, mechanism impractical to reimplement | Honest qualitative contrast: what it assumes, what yours assumes, why numbers aren't comparable |
| Same mechanism family, different problem | One sentence acknowledging the family; do not manufacture rivalry |
| Classic paper that named the idea space | Cite it — omitting the acknowledged origin insults exactly the senior reviewers most likely to notice |
| Concurrent preprint (recent arXiv) | Cite neutrally as concurrent; differentiate without claiming priority wars |
Double-blind mechanics, ISCA 2026 wording
The verified 2026 guidelines are specific and slightly unusual:
- References must not be omitted or anonymized — the bibliography stays intact.
Blinding happens in the prose voice, not the reference list.
- Your own prior papers are cited in third person: "extends the tracker of Chen et
al. [7]" — never "our prior design [7]". Audit for possessives near citations.
- Links to your artifacts (repos, project pages) must be fully anonymized;
a github.com/ URL anywhere in the paper is an identity leak.
grep -nEi 'our (prior|previous|earlier|recent) (work|paper|design|study)' *.tex
grep -nEi 'we (previously|earlier) (showed|proposed|presented)' *.tex
grep -nE 'github\.com|gitlab\.com|bitbucket|\.edu/~[a-z]' *.tex *.bib
grep -oE '\\cite[tp]?\{[^}]*\}' *.tex | tr ',' '\n' | sort -u | -l
