README.md

---
license: apache-2.0
task_categories:
- reinforcement-learning
language:
- en
tags:
- eda
- analog
- rl
pretty_name: OSIRIS Dataset
---

# OSIRIS: Bridging Analog Layout Circuit Design and Machine Learning with Scalable Dataset Generation 

**OSIRIS** is an end-to-end analog circuit design pipeline capable of producing, validating, and evaluating large volumes of layouts for generic analog circuits.

The [OSIRIS 🤗 HuggingFace repository](https://huggingface.co/datasets/hardware-fab/osiris) hosts the randomly generated dataset discussed in the paper.

This codebse provides the OSIRIS Python script (`osiris.py`) and the Python friendly version of the SkyWater 130nm PDK (`SKY130_PDK`) along with the necessary files to perform place and route operations (in `schematic2layout` folder).

- **Curated by:** hardware-fab
- **License:** Open Data Commons License [cc-by-4.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/cc-by-4.0.md)
- **ICLR 2026 Paper:** [OSIRIS: Bridging Analog Layout Circuit Design and Machine Learning with Scalable Dataset Generation](https://arxiv.org/pdf/2601.19439)

## How to Download
The code is stored in `osiris_code.zip` while the dataset is stored in `Osiris_Dataset.zip`.
```python 
from huggingface_hub import hf_hub_download

file_path = hf_hub_download(
    repo_id="hardware-fab/osiris",
    filename=<filename_to_download>,
    repo_type="dataset",
    local_dir=<download_path>
)
```

Where `<filename_to_download>` is either `osiris_code.zip` or `Osiris_Dataset.zip` and `<download_path>` is the path to your local folder.

## Installation
To run OSIRIS a suite of tools is necessary.

### Netgen
Netgen is used to carry out Layout Versus Schematic (LVS). We use version `1.5.293`.
To install it, please refer to the official GitHub [repository](https://github.com/RTimothyEdwards/netgen).

### Magic
Magic is used to perform Parasitic Extraction (PEX). We use version `8.3.497`.
To install it, please refer to the official GitHub [repository](https://github.com/RTimothyEdwards/magic).

### Ngspice
Ngspice is used to perform simulations. We use version `ngspice-43`.
To install it, please refer to the official [website](https://ngspice.sourceforge.io/).

### SkyWater 130nm PDK
To install the SkyWater 130 nm PDK used throughout the research, please refer to the official [GitHub repository](https://github.com/google/skywater-pdk).
These PDK files are used by Netgen, Magic, and Ngspice to perform LVS, PEX, and simulation respectively. While the `SKY130_PDK` files are 
a Python friendly version of the SkyWater 130nm PDK to perform the place and route operations.

## Usage
You will need to adjust a few paths to your working setup. 
- in each testbench: the path pointing to SkyWater simulation libraries, i.e., `sky130A/libs.tech/ngspice/sky130.lib.spice`.
- in `config.py`: the `WD` path variable.
- in `0_alter_nf.py`: the `base_path` path variable.
- in `pex.sh`: the path pointing to SkyWater root, i.e., `sky130A`.
- in `1_reroute_sch2lay.sh`: the path pointing to `SKY130_PDK`.

Then you will need to compute the fingers permutations. 
To do so, refer to the `2_flow/0_alter/0_alter_nf.py` script.

To experiment with OSIRIS, use the `osiris.py` file.

To perform random exploration of circuit design space, set: 

> EXPLORATION_MODE = 'dataset'

To perform the RL-driven exploration baseline of layout design space reported in the paper, set:

> EXPLORATION_MODE = 'rl'

## Note
This repository is protected by copyright and licensed under the Apache-2.0 license.