OpenTimer-PI

Code of “Dynamic Supply Noise Aware Timing Analysis with JIT Machine Learning Integration”.

System Requirements

• GNU C++ Compiler v7.3 with -std=c++1z
• Clang C++ Compiler v6.0 with -std=c++17
• PyTorch >= 1.6.0

Basic Usage

~$ ./bin/ot-shell

ot> cd example/simple
ot> read_celllib osu018_stdcells.lib
ot> read_verilog simple.v   
ot> read_sdc simple.sdc
ot> read_noise_models ../sample/models_trace        # The path of MLP models
ot> report_timing

For further usage instructions, please refer to the OpenTimer

Update MLP Models

1. Place the pre-trained MLP (Multi-Layer Perceptron) weights from PyTorch, saved with the ‘.pth’ extension, in the directory ‘OpenTimer-PI/sample.’

2. Execute the script ‘trace_pt_script.py’ located in the aforementioned directory to generate the Intermediate Representations.

3. Recompile OpenTimer.

Compile

~$ cd OpenTimer-PI
~$ mkdir build
~$ cd build
~$ cmake ../
~$ make 

After successful build, you can find binaries and libraries in the folders bin and lib, respectively.

Acknowledgement

We are greatly appreciative of the open-source project OpenTimer , upon which we have built and completed this endeavor