UniCNet: Unified Cycle-Accurate Simulation for Composable Chiplet Network with Modular Design-Integration Workflow
UniCNet is a cycle-accurate simulator supporting effienct simulation for composable chiplet networks.
- Unified and accurate evaluation of entire composable chiplet network aligh with the design flow of chiplet-based architectures.
- Support key design features oriented towards multi-chiplet scenario, including composable topology, modular routing and heterogeneous router.
- Cycle-level chiplet protocal interface model, supporting UCIe protocol and verified it against RTL model.
- Enable multi-thread parallel simulation, achieving up to 4X speedup while maintaining the same accuracy with Booksim2.0.
git clone https://github.com/wangplin/CAL-UniCNet.git UniCNet
cd UniCNetapt install bison
apt install flexcd src
make -j$(nproc)cd ../configs
./../src/booksim UniCNet.cfg// active interposer
integration_file = integration_active_file;
interposer_config = interposer.cfg;// passive interposer
integration_file = integration_passive_file;
routing_function_chiplet = chip_level_dor_mesh;// ucie interface
// more parameters can be found in booksim_config.cpp
interface = ucie;
ucie_flit_size = 64;
ucie_serdes_ratio = 4;
ucie_lane_width = 0.9;
ucie_error_rate = 0.1;enable_parallel = 1;
num_threads = 16;[1] Jiang, N., Becker, D.U., Michelogiannakis, G., Balfour, J., Towles, B., Shaw, D.E., Kim, J. and Dally, W.J., 2013, April. A detailed and flexible cycle-accurate network-on-chip simulator. In 2013 IEEE international symposium on performance analysis of systems and software (ISPASS) (pp. 86-96). IEEE.
[2] https://github.com/booksim/booksim2
If you use UniCNet in your research, we would appreciate the following citation in any publications to which it has contributed:
@ARTICLE{11347548,
author={Wang, Peilin and Wang, Mingyu and Ye, Zhirong and Lu, Tao and Yu, Zhiyi},
journal={IEEE Computer Architecture Letters},
title={UniCNet: Unified Cycle-Accurate Simulation for Composable Chiplet Network With Modular Design-Integration Workflow},
year={2026},
volume={},
number={},
pages={1-4},
doi={10.1109/LCA.2026.3653809}}