Data and Code for: Enhanced Memetic Algorithm with Adaptive Local Search for Area-Optimized VLSI Standard-Cell Placement
Description
This dataset contains all materials required to reproduce the experimental results reported in the paper "Enhanced Memetic Algorithm with Adaptive Local Search for Area-Optimized VLSI Standard-Cell Placement" (submitted to IETE Journal of Research, 2026). Contents: 1. ema_source_code.zip — Python implementation of the EMA algorithm, benchmark generator, ablation study scripts, MCNC circuit loader, and OpenROAD integration adapter. 2. benchmarks.zip — 16 JSON benchmark files: 10 synthetic instances (3–49 modules), 3 scalability instances (100–500 modules), and 3 MCNC/GSRC soft-module circuits (apte, xerox, hp). 3. results_tables.xlsx — Raw 30-run experimental results for all benchmark instances (Tables 2–7 of the paper), including mean fitness, bounding-box area, HPWL, 95% confidence intervals, and runtime. 4. README.txt — Full documentation covering algorithm parameters, file formats, usage instructions, reproducibility notes, and citation information. All Python code uses only the standard library (Python 3.9+, no third-party packages required). Fixed random seeds are used throughout; all results are reproducible by running the provided scripts. ISPD 2005/2006 benchmark circuits (adaptain, bigblue1, adaptec2, newblue2) are not included due to contest licensing but are freely available from the ISPD contest archive (https://ispd.cc).
Files
Steps to reproduce
Requirements: Python 3.9 or later (no third-party packages). Step 1 — Generate all benchmark instances: python ema_source_code/generate_benchmarks.py --outdir benchmarks/ Step 2 — Run all experiments (30 independent runs each): python ema_source_code/run_all_experiments.py --benchdir benchmarks/ --outdir results/ Step 3 — Run ablation study (Table 4): python ema_source_code/ablation_study.py --benchmark benchmarks/syn_25.json --runs 30 Step 4 — Single benchmark with custom parameters: python ema_source_code/ema_placer.py --benchmark benchmarks/syn_49.json --runs 30 All results use fixed random seeds {0...29} for 30 independent runs. MCNC instances (apte, xerox, hp) are pre-generated by running: python ema_source_code/mcnc_loader.py Full parameter details and file format documentation are in README.txt.