What was the target clock period used for MemPool Group to generate the results in the ISPD-2023 paper?

[bodhi91]

In the ISPD-2023 paper "Assessment of Reinforcement Learning for Macro Placement," the mentioned target clock period for the MemPool Group in Nangate45 is 4.0ns, but the script available in the repository [here] uses 3ns. What was the target clock period used to generate the data points for the ISPD-2023 paper?

[sakundu]

Thank you for raising this issue. We have received this question in several emails since early March, and clarified to those correspondents. In our ISPD 2023 experiment, we used a 4ns target clock period for the MemPool Group in Nangate45 (NG45) enablement to evaluate all the macro placement solutions.

You can reproduce the MemPool Group baseline solution in Nangate45 by updating the clk_period to 4 (a one-character edit) and running the flow. Please let us know if you have difficulty in doing this.

[Note: We found that we could further tune the target clock period to 3ns to make the MemPool Group in NG45 more challenging. However, due to the publication schedule constraints, we could not generate solutions for all the macro placements before our ISPD 2023 submission. Therefore, our published ISPD 2023 paper gives results obtained with the target clock period of 4ns for the MemPool Group in NG45 enablement.]