Hanan Grid Router

This is a rectilinear detail router useful for routing and visualizing signal nets on few blocks (both digital and analog). It uses a modified version of A* search aglorithm at its core with the routes on the Hanan grid.

For now, the following sets of back-end of line design rules are honored:

• Manhattan spacing between metal and via shapes on each layer.
• Metal widths per layer
• Single/mult-cut vias
• User-specified non-default rules at a block level or net level. User can override the following
  • metal spacing and widths in any layer
  • restrict the use of certain routing layers for certain nets
  • net/block specific routing obstacle

Building the router

Requirements:

• A C++ compiler with support for C++14 or higher
• make system

Steps to clone and build:

git clone https://github.com/srini229/hanan_router.git
cd hanan_router
make -j4

The build commands have been verified on: Ubuntu 18.04 (GCC 6), Fedora 36 (GCC 12), Debian 10 on WSL (GCC 8), and Mac OS 10.15 (Clang 11).

Make generates a binary hanan_router on the cloned directory. The repository has a frozen copy of the nlohmann/JSON parser packaged with it to parse the netlist, layer information and user constraints. It has a small version of LEF parser that understands basic LEF syntax for now. The router uses RTree data structure to quickly accessing the geometric structures from memory. It has a copy of the RTree. Migration of the JSON/RTree APIs to Boost libraries is a work in progress. This will eliminate the need for these separate thirdparty libraries.

The JSON file formats reuse the syntax from the ALIGN project. A generic version to use LEF/DEF files of the netlist/design rules is a work in progress.

Interface

The router syntax is:

usage : hanan_router
	-d <layers.json> ; # Abstracted information for each layer metal/via layer used in a technology
	-p <placement file> ; # Placement information in JSON format
	-l <lef file> ; # LEF file containing the pin/blockage information for each leaf cell
	-s <lef scaling> ; # units used in the LEF file
	-uu <user units scaling> ; # user units scaling for placement file (e.g. nm/um)
	-ndr <ndr constraints.json> -o <output dir> ; # user specified non-default rules

The router generates one LEF and one DEF for each hierarchy in the placement. The LEF file has the suffix _interim_hier.lef. There is also a route.log created which has a detailed log of the routers activity.

There is also a GDSII file generator available in the bin/ directory. It is written in python and requires the the following packages numpy, gdspy, json, and argparse. Any missing package can be installed using the command pip install <package>. The syntax to generate the GDSII files using the LEF/DEF files is:

usage: gen_rt_gds.py [-h] [-p PL_FILE] [-g GDS_DIR] [-t TOP_CELL] [-u UNITS] [-s SCALE] [-l LAYERS] [-d DEFF]

options:
  -h, --help            show this help message and exit
  -p PL_FILE, --pl_file PL_FILE
                        <filename.placement_verilog.json>
  -g GDS_DIR, --gds_dir GDS_DIR
                        <dir with all leaf gds files>
  -t TOP_CELL, --top_cell TOP_CELL
                        <top cell>
  -u UNITS, --units UNITS
                        <units in m>
  -s SCALE, --scale SCALE
                        <scale>
  -l LAYERS, --layers LAYERS
                        <layers.json>
  -d DEFF, --deff DEFF  <route def file>

Test directory

There is a small unit test directory test/. To run the test:

./hanan_router -d layers.json -p test.placement_verilog.json -l test.lef

The ndr1.json, ndr2.json and ndr3.json demonstrate different user constraints. To run the router with the user constraints:

./hanan_router -d layers.json -p test.placement_verilog.json -l test.lef.

The LEF/DEF outputs generated by the router can be visualized using klayout.

To generate GDSII file from the LEF/DEF files use:

../bin/gen_rt_gds.py -p test.placement_verilog.json -g . -t TEST_CONC_0  -l layers.json -d TEST_CONC_0.def 

The GDSII files can be visualized in klayout as well. The GDSII file can be imported into commercial PnR tools using the stream in function.