Computer Architecture Researcher (Research Coding Tasks)

What You'll Do10

• Select a recent architecture paper that ships working code, ideally one you wrote yourself.
• Remove the portion of the codebase that carries the paper's central contribution and leave the rest intact.
• Write the task statement that tells an agent what to rebuild and how success gets measured.
• Choose held-out workloads that the agent cannot see while solving the task.
• Set a pass threshold for each metric based on what the original paper reports.
• Build a hidden reference solution the agent never gets access to.
• Produce a second solution that clears the threshold, proving the bar is reachable.
• Validate that the packaged task runs and that the baseline results reproduce.
• Evaluate agent submissions against your task and judge whether each attempt clears the metric thresholds.
• Keep the benchmark setup reproducible so results hold across runs.

Requirements7

• Published architecture research at venues such as ISCA, MICRO, HPCA or ASPLOS, or equivalent peer-reviewed work.
• A public code release tied to at least one of your papers, with a GitHub or artifact link you can share.
• PhD in computer science or a related discipline, either finished or in progress.
• First-author papers that report measured gains over a baseline are a strong plus.
• Working knowledge of Docker and reproducible benchmark harnesses.
• Depth in caches, prefetchers, branch prediction or accelerators, evaluated through simulators like gem5 or ChampSim.
• Ability to write clear task specifications and per-metric acceptance criteria.

Who Should Apply

Architecture PhDs and research engineers who publish at top venues and actually ship the code behind their results. You should be comfortable reading a paper, isolating its core mechanism and building a benchmark harness that another party can run without your help.

Salary Insight