Setup

⚙️ Environment Setup

This project requires Python 3.5, Anaconda/miniconda, and runs in any Linux environment with x86_64 CPU.

📦 1. Install Conda

Please install either Anaconda or Miniconda for Python 3.5 support.

For csh Shell Users

After installation, add the following to your .cshrc (e.g., cshrc.gf55BCDlite_v1090):

# model path
setenv SPECTRE_MODEL_PATH /net/vlsiserver/usr11/library/CSM/gf55BCDlite_v1090_stack7A/55BCDlite/V1.0_9.0/Models/Spectre/55bcdlite/models/

# Conda setup
setenv PATH "${HOME}/anaconda3/bin:${PATH}"
if ( -f "${HOME}/anaconda3/etc/profile.d/conda.csh" ) then
    source "${HOME}/anaconda3/etc/profile.d/conda.csh"
else
    echo "No Conda csh file found in the specified path."
endif

Then run:

source path/to/csh/file
conda activate

You may need to manually install some additional packages via pip later.

📁 2. Clone the Project

git clone https://github.com/lwasoo/AutoCkt.git
cd AutoCkt

🐍 3. Create Conda Environment

From the root directory:

conda env create -f environment.yml #only need to create once
conda activate autockt

If needed, use pip to manually install missing packages.

🏗️ 4. Project Structure

`-- AutoCkt
    |-- autockt \\Core model code
    |   |-- envs
    |   |   `-- ngspice_vanilla_opamp.py \\OpenAI Gym environments
    |   |-- gen_specs
    |   |   `-- ngspice_specs_gen_two_stage_opamp \\generate targets from gen_spec
    |   |-- gen_specs.py \\generate target spec
    |   |-- rollout.py \\validation entry point
    |   `-- val_autobag_ray.py \\Training entry point
    |-- environment.yml \\conda env 
    |-- eval_engines \\interacts with simulator
    |   |-- ngspice
    |   |   |-- TwoStageClass.py \\calculate gain, ibias, ugbw, phm
    |   |   |-- ngspice_inputs
    |   |   |   |-- netlist
    |   |   |   `-- yaml_files
    |   |   |       `-- two_stage_opamp.yaml \\circuit parameters spec
    |   |   `-- ngspice_wrapper.py \\simulator interface

🧪 5. Simulate and Train

Generate Specs

python autockt/gen_specs.py --num_specs 350

Specs will be saved as pickle files under gen_specs/.

Run Training

ipython
run autockt/val_autobag_ray.py

Training results will be stored in ~/ray_results.

To monitor training:

tensorboard --logdir ~/ray_results

✅ 6. Run Validation

To test generalization of the trained model:

1. Generate new specs:

   python autockt/gen_specs.py --num_specs 100
   

2. Run validation:

   ipython
   run autockt/rollout.py /path/to/checkpoint --run PPO --env opamp-v0 --num_val_specs 100 --traj_len 20 --no-render
   

Validation results will be saved as:

• opamp_obs_reached_test
• opamp_obs_nreached_test

The steps for checking the netlist are described in detail in Others.

🔧 7. Customize Circuit (2stage Example)

To modify for cascode topology:

1. Edit line 121 in ngspice_vanilla_opamp.py:

   self.cur_params_idx = np.array([33, 33, 33, 33, 33, 14, 20])
   

2. Modify the initial step vector:

   env.step([2, 2, 2, 2, 2, 2, 2])
   

3. Edit the YAML spec file two_stage_opamp.yaml:
   • Set custom parameter ranges
   • Set target_spec for new design goals

📌 Notes

• Default netlist is located at:
  eval_engines/ngspice/ngspice_inputs/netlist/2opamp
• Training performance is hardware-dependent. For reliable results, use at least 2 cores, ideally 8 cores.
• Results can vary depending on random seed and spec generation.