Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer
Authors: Ke Xue, Ruo-Tong Chen, Xi Lin, Yunqi Shi, Shixiong Kai, Siyuan Xu, Chao Qian
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach on the ISPD 2005 and ICCAD 2015 benchmark, comparing the global half-perimeter wirelength and regularity of our proposed method against several competitive approaches. Besides, we test the PPA performance using commercial software, showing that RL as a regulator can achieve significant PPA improvements. |
| Researcher Affiliation | Collaboration | 1National Key Laboratory for Novel Software Technology, Nanjing University 2School of Artificial Intelligence, Nanjing University 3Huawei Noah s Ark Lab |
| Pseudocode | No | The paper includes figures describing the architecture and masks (Figures 2 and 3) but does not provide any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https://github.com/lamda-bbo/macro-regulator. |
| Open Datasets | Yes | We mainly use the ICCAD 2015 benchmark [14] as our test-bed... Besides, we also conduct experiments on ISPD 2005 benchmark [25] |
| Dataset Splits | No | The paper mentions 'pre-train' and 'test' splits (e.g., 'pre-train Mask Regulate and Mask Place on the first four chips... and test on the remaining four chips') but does not specify a separate validation dataset split with exact percentages or sample counts. |
| Hardware Specification | Yes | Device. CPU: Intel(R) Xeon(R) Gold 6430 GPU: 4 Ge Force RTX 4090 |
| Software Dependencies | No | The paper mentions using 'original implementations' for compared methods and lists some software with links in footnotes (e.g., DREAMPlace, Auto DMP, Wire Mask-EA, Mask Place), but it does not specify version numbers for general ancillary software components (e.g., Python, PyTorch, CUDA). |
| Experiment Setup | Yes | Table 4: Hyperparameters Configuration Value. Optimizer Adam, Learning rate 2.5e-3, Total episode 1000, Epoch for update 10, Batch size 64, Buffer capacity 5120, Clip ϵ 0.2, Clip gradient norm 0.5, Reward discount γ 0.95, Mask soft coefficient 1, DREAMPlace evaluation number 3, Trade-off coefficient α 0.7, Grid soft coefficient 4. |