isca-topic-selection

ISCA Topic Selection

ISCA reviews reward papers whose central object is a hardware organization: a

pipeline, a cache hierarchy, a memory controller, an accelerator datapath, an

interconnect, a coherence or consistency mechanism. Software may motivate the work

and co-design may strengthen it, but at ISCA the question a program committee member

carries into every discussion is *what does this paper teach us about how to build

machines?* Projects that cannot answer that in one sentence tend to be respected and

rejected in the same review.

The center-of-gravity question

Write down the single artifact your evaluation section studies hardest. Then ask

where its novelty lives:

• In the hardware organization — a new structure, policy, or datapath whose

behavior you measure across workloads. Natural ISCA material.

• In the hardware/software contract — an ISA extension or OS-visible mechanism

where the software side changes too. ISCA takes these, but ASPLOS's reviewer pool

is built for them; decide by which half carries the insight.

• In the software layer alone — a compiler pass, runtime, or scheduler running

on unmodified silicon. Reviewers will ask what the architecture lesson is; if the

answer is "none, but it's fast," route away before writing.

• In circuits or devices — below the microarchitecture line, DAC/ISSCC-adjacent

venues own the expertise ISCA reviewers will say they lack.

A useful stress test: describe the contribution without naming any benchmark. If

what remains is a statement about machine organization ("decoupling X from Y removes

the Z stall class"), the paper is ISCA-shaped. If what remains is only a speedup, it

is a results report, and every top venue struggles to accept those.

ISCA's seat in the architecture year

ISCA's November deadline and June conference define one corner of a four-venue

rotation. Retargeting between them is normal practice; plan it as a calendar

decision, not an act of desperation. Dates below reflect the most recent verified

cycle (ISCA 2026) — reopen each venue's current CFP before committing.

| Venue | Typical deadline seat | Conference seat | Takes best what |

|---|---|---|---|

| ISCA | mid-November (Nov 10/17, 2025 for the 2026 edition) | late June | Broad architecture ideas: cores, memory, accelerators, interconnect, evaluation methodology |

| MICRO | spring (roughly April) | October | Microarchitecture depth: front-end, prediction, scheduling, implementation-level mechanisms |

| HPCA | mid-summer (roughly July/August) | February | Architecture with a performance-systems flavor; a fast second gate after a June ISCA reject |

| ASPLOS | multiple gates per edition | spring | Work whose hardware and software halves are inseparable |

The practical rhythm from ISCA's seat: a paper rejected from ISCA's November gate

can be revised for MICRO in spring; a paper that misses November because evidence

was immature has HPCA's summer gate before the next ISCA November. Retarget on

reviewer-pool fit, never on deadline convenience alone — each venue's committee

notices work dressed for a different room.

Main track or industry track

ISCA 2026 ran a separate industry track with its own later deadlines (abstracts

December 5, papers December 12, 2025). Choose it when the contribution's value is

evidence from real products at scale — measurements, deployment lessons, or

design retrospectives that academia cannot produce — rather than a novel mechanism.

A novel mechanism from an industrial lab still belongs in the main track.

Fit worksheet

Run this before any drafting; keep the answers in the repo.

CLAIM     One sentence, no benchmark names: what about machine organization
          does this paper establish?
LEVEL     Which layer changes: core / memory hierarchy / accelerator /
          interconnect / ISA contract / methodology?
AUDIENCE  Name three recent ISCA papers whose authors are the natural
          reviewers. (Cannot name three -> wrong room.)
DELTA     Against the nearest of those three: what does this design do that
          theirs cannot, stated as behavior not adjectives?
EVIDENCE  Is the evaluation instrument (simulator / RTL / FPGA / silicon
          measurement) credible for THIS claim? (See isca-experiments.)
TIMING    Can the evidence be complete 6 weeks before mid-November?
          If not, which venue's gate matches the real finish date?
ROUTE     ISCA main / ISCA industry / MICRO / HPCA / ASPLOS / defer.

Shapes with a strong ISCA record

The Influential ISCA Paper Award record (see

../../resources/exemplars/library.md) shows the venue's long-run taste. Winning

shapes include a security-motivated cache redesign, a polymorphous core

architecture, 3D-stacked memory organization, a datacenter power-provisioning

measurement study, and a thermal-management design-space classification. Two

lessons: measurement and methodology papers can be first-class ISCA contributions,

and ideas that name a reusable organizing principle outlast ideas that optimize

one metric.

Anti-fits that waste a cycle

• A speedup on an unmodified commercial chip with no architectural mechanism —

systems venues will treat it better.

• A new benchmark suite with no analysis of what it reveals about designs —

IISWC-shaped until the analysis exists.

• An accelerator for workload W whose entire delta is "we specialized for W" —

reviewers ask for the generalizable design insight.

• RTL-quality engineering of a known idea — valuable, but "known idea" is the

review's first line.

• A simulation study whose conclusion depends on a fidelity level the simulator

cannot support (see isca-experiments before deciding fit, not after).

Decide and record

[ISCA fit] strong / plausible / weak
[Center of gravity] <layer + one-sentence organizing claim>
[Nearest reviewers] <3 recent ISCA papers>
[Track] main / industry
[Fallback seat] MICRO <year> / HPCA <year> / ASPLOS <cycle> + trigger condition
[Evidence deadline] <date evidence must freeze for a mid-November submission>

Facts here are the 2026-cycle snapshot verified 2026-07-08

(../../resources/official-source-map.md); the ISCA 2027 CFP was not yet posted at

check time, so treat the November seat as expected, not promised — 待核实. Scope,

track structure, and topics lists are re-decided per edition: reread the live CFP's

topics enumeration before declaring anything out of scope.