News for hackers. Stuff that matters.

A comment on AlphaChip transformed computer chip design. Parent comment

Re:AlphaChip transformed computer chip design by Drunk_Engineer on Friday September 27, 2024 @06:21PM (#41676108)
From what I saw in the rebuttal papers, the Google cost-function is wirelength based. You can still get good TNS from that if your timing is very simplistic -- or if you choose your benchmark carefully.
[ Parent | on HN ]
Re:AlphaChip transformed computer chip design by negativeonehalf on Friday September 27, 2024 @06:52PM (#41676304)
They optimize using a fast heuristic based on wirelength, congestion, and density, but they evaluate with full P&R. It is definitely interesting that they get good timing without explicitly including it in their reward function!
[ Parent | on HN ]
Re:AlphaChip transformed computer chip design by ithkuil on Saturday September 28, 2024 @03:18AM (#41678524)
Yeah; it means the heuristic they use is a good one
[ Parent | on HN ]
Re:AlphaChip transformed computer chip design by gabegobblegoldi on Thursday October 03, 2024 @01:45AM (#41727676)
Interesting == Suspicious? I think this is a big red flag to those in the know.
[ Parent | on HN ]
Re:AlphaChip transformed computer chip design by clickwiseorange on Friday September 27, 2024 @07:16PM (#41676473)
The odd thing is that they don't compute timing in RL, but claim that somehow TNS and WNS improved. Does anyone believe this? With five circuits and three wins, the results are a coin toss.
[ Parent | on HN ]