文件列表:

CMakeLists.txt (2826, 2023-03-12)
LICENSE (35149, 2023-03-12)
cmake/ (0, 2023-03-12)
cmake/FindTCL.cmake (2344, 2023-03-12)
cmake/GetGitRevisionDescription.cmake (4963, 2023-03-12)
cmake/GetGitRevisionDescription.cmake.in (1283, 2023-03-12)
include/ (0, 2023-03-12)
include/openroad/ (0, 2023-03-12)
include/openroad/Error.hh (2159, 2023-03-12)
include/openroad/InitOpenRoad.hh (1905, 2023-03-12)
include/openroad/OpenRoad.hh (4560, 2023-03-12)
include/openroad/Version.hh (129, 2023-03-12)
include/openroad/Version.hh.cmake (128, 2023-03-12)
src/ (0, 2023-03-12)
src/ArtNetGen/ (0, 2023-03-12)
src/ArtNetGen/CMakeLists.txt (1614, 2023-03-12)
src/ArtNetGen/include/ (0, 2023-03-12)
src/ArtNetGen/include/artnetgen/ (0, 2023-03-12)
src/ArtNetGen/include/artnetgen/MakeArtNetGen.h (275, 2023-03-12)
src/ArtNetGen/include/artnetgen/artNetGen.h (6953, 2023-03-12)
src/ArtNetGen/src/ (0, 2023-03-12)
src/ArtNetGen/src/MakeArtNetGen.cpp (746, 2023-03-12)
src/ArtNetGen/src/artNetGen.cpp (19185, 2023-03-12)
src/ArtNetGen/src/artNetGen.i (2411, 2023-03-12)
src/ArtNetGen/src/artNetGen.tcl (4476, 2023-03-12)
src/ArtNetGen/src/bin.cpp (2549, 2023-03-12)
src/ArtNetGen/src/bin.h (837, 2023-03-12)
src/ArtNetGen/src/distMatching.cpp (8683, 2023-03-12)
src/ArtNetGen/src/distribution.cpp (7681, 2023-03-12)
src/ArtNetGen/src/logger.cpp (4495, 2023-03-12)
src/ArtNetGen/src/logger.h (3134, 2023-03-12)
src/ArtNetGen/src/netlist.cpp (14789, 2023-03-12)
src/ArtNetGen/src/netlist.h (2833, 2023-03-12)
src/ArtNetGen/src/node.cpp (8537, 2023-03-12)
src/ArtNetGen/src/node.h (2768, 2023-03-12)
src/ArtNetGen/src/techMapping.cpp (50053, 2023-03-12)
src/ArtNetGen/src/writeSpec.cpp (30479, 2023-03-12)
src/CMakeLists.txt (7942, 2023-03-12)
... ...

# Artificial Netlist Generator ## Introduction The structural characteristics of a place and route (P&R) benchmark vary widly depending on the logic and synthesis method with target technologies. The structural characteristics affect the solution space of CAD algorithms (e.g., partitioning, placement, routing, etc), so insufficient number of P&R benchmarks may lead to the poor generalizability for unseen circuits. Research communities have tried to publish open-source benchmarks to support CAD research in recent decades, but they are mapped to a specific technology node or written in a data format that is not supported by commercial tools. Furthermore, to protect the intellectual property of the developers, the data must be anonymized (i.e., technical details of the circuits). Researchers somtimes create several new versions of circuits by combining multiple modules or modifying functionality at the RTL level, but these modifications cannot provide a large variation in circuit characteristics compared to the original circuit and consume a lot of human resources for data scalability. One of the solutions to mitigate the lack of circuit benchmarks is to generate artificial circuits that represent realistic circuit characteristics. Artificial circuits can be generated using specific rules and programs so that it is cost-less compared to real-world data in terms of data scalability. This is a strong motivartion why we developed this program called 'artificial netlist generator' (ANG). ANG can create any artificial gate-level netlist from user-specified parameters representing the topological characteristics of the circuit. ANG takes six topological parameters as inputs (see following table) --- * **Topological parameters** |Flag|Description|Recommended Value| |---|---|---| |-num_insts|number of instances|| |-num_primary_ios|number of primary input and outputs|~10% of *-num_insts*| |-avg_net_degree|average net degree(=average number of fanouts)|2.5 ~ 4.0| |-avg_net_bbox|average size of net bounding box (=average hpwl after placement)|0.1 ~ 2.0| |-avg_topo_order|average depth of timing paths (= average depth of leaf nodes)|5.0 ~ 15.0| |-comb_ratio|ratio of combinational logic gates over the number of instances|0.75 ~ 0.95| * **Components in a gate-level netlist** |Graph representation|Decomposition into path groups| |:---:|:---:| ||| --- ANG can be used to perform (a) data augmentation by modifying topolgical parameters extracted from real-world circuits, or (b) data exploration by random sampling in unseen circuit space to be expected. We believe that those approaches will increase the feature coverage of the post-layout database compared to only using real-world circuit benchmarks. |Data generation flow using ANG| |:---:| || |How to sample topological parameters| || ## Usage ./openroad #Read both tech_lef and std_lef read_lef read_lef #Create circuit spec artnetgen_create_spec -num_insts -num_primary_ios -comb_ratio -avg_bbox -avg_net_degree -avg_topo_order -cell_list -out_file #Initialize netlist artnetgen_init -top_module -spec_file #If you want to check MACROs to be mapped, use this following command artnetgen_print_masters #If you want to remove specific MACROs in the list, use this following command artnetgen_set_parameter -dont_use #Run artificial netlist generator artnetgen_run #Write output files artnetgen_write_verilog -out_file artnetgen_write_sdc -out_file ## How to build git clone --recursive https://github.com/daeyeon22/artificial_netlist_generator.git cd artificial_netlist_generator mkdir build & cd build & cmake .. make -j ## Features This code is based on OpenROAD project [(link)](https://github.com/The-OpenROAD-Project/OpenROAD) # Requirements - swig 3.0 - gcc 8.3.1 - cmake 3.20 # Submodules - The-OpenROAD-Project/OpenDB [(link)](https://github.com/The-OpenROAD-Project/OpenDB) - The-OpenROAD-Project/OpenSTA [(link)](https://github.com/The-OpenROAD-Project/OpenSTA) - The-OpenROAD-Project/OpenROAD/src/dbSta [(link)](https://github.com/The-OpenROAD-Project/OpenROAD/tree/master/src/dbSta) # Authors * Paper reference * **Daeyeon Kim**, Sung-Yun Lee, Kyungjun Min, and Seokhyeon Kang, "Construction of Realistic Place-and-route Benchmarks for Machine Learning Applications," *IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems* (TCAD), 2022 [(link)](https://ieeexplore.ieee.org/document/9904912). * **Daeyeon Kim**, Hyunjeong Kwon, Sung-Yun Lee, Seungwon Kim, Mingyu Woo, and Seokhyeong Kang, "Machine Learning Framework for Early Routability Prediction with Artificial Netlist Generator," *Design, Automation & Test in Europe Conference & Exhibition* (DATE), 2021 [(link)](https://ieeexplore.ieee.org/document/9473966).

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：