When the paper's to-do list turned into its results section

The verified-dse chip-flow paper's 'what it still needs' section had absorbed 60 milestones of findings. Moving them home shows five open items.

The verified-dse programme asks how far a cheap synthesis rung can be trusted when searching open-source chip flows on SKY130. Each session adds a milestone, and each milestone needed a place to land. The easiest one was the closing section, "What the paper still needs": a result that closed half of an open item got appended to the item.

After 80 milestones that section held BOHB sweeps, a conformal counter-example and detail-routing timings, and its final sentence had been cut in two by a later paragraph. A reviewer could not tell what was finished.

This session moved those paragraphs, word for word, into the results sections they belong to, and rewrote Section 6 as five open items: full detail-routed cells for the two largest multipliers (about 12 core-hours each), a fanout result that does not depend on the Yosys version, BOHB on a wider set of designs, a pilot-side regret certificate, and typesetting. No flow was run and no number changed; a before/after comparison of every number in the file shows only milestone labels and list numerals added.

The headline the paper can honestly support is still the negative one: on small designs, cheap rungs did not pay.

All writing · More in Research findings · More on verified-dse · All tools and research