Dataset Viewer
The dataset viewer is not available for this dataset.
Job has been terminated due to a temporary spike in resource usage and may be restarted later.
Error code:   JobManagerCrashedError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

SKY130 Analog Layout Dataset

This dataset contains 30,031 GDS layouts built with the SkyWater SKY130 process. It is organized into base circuits, mixed circuits, and isolated single devices.

Dataset structure

dataset/
├── base_circuits/
│   ├── ahuja_ota/
│   │   └── ahuja_ota_<id>/ahuja_ota_<id>.gds
│   ├── gate_driver/
│   │   └── gate_driver_<id>/gate_driver_<id>.gds
│   ├── hpf/
│   │   └── hpf_<id>/hpf_<id>.gds
│   ├── ldo/
│   │   └── ldo_<id>/ldo_<id>.gds
│   ├── lpf/
│   │   └── lpf_<id>/lpf_<id>.gds
│   └── miller_ota/
│       └── miller_ota_<id>/miller_ota_<id>.gds
├── mixed_circuits/
│   └── mixed_<id>/mixed_<id>.gds
├── single_devices/
│   ├── single_cap/
│   │   └── single_cap_<id>/single_cap_<id>.gds
│   ├── single_nmos/
│   │   └── single_nmos_<id>/single_nmos_<id>.gds
│   ├── single_pmos/
│   │   └── single_pmos_<id>/single_pmos_<id>.gds
│   └── single_res/
│       └── single_res_<id>/single_res_<id>.gds
└── generate_labels.py

Base circuits

base_circuits contains complete analog circuit layouts. These GDS files retain their hierarchy and contain named references for individual devices. The device names encode the instance and SKY130 primitive, for example:

M0_sky130_fd_pr__nfet_01v8
M3_sky130_fd_pr__pfet_01v8
C0_sky130_fd_pr__cap_mim_m3_2
XR0_sky130_fd_pr__res_xhigh_po

The current family counts are:

Family Samples
Ahuja OTA 995
Gate driver 1,000
High-pass filter (hpf) 962
Low-dropout regulator (ldo) 989
Low-pass filter (lpf) 971
Miller OTA 977
Total 5,894

Mixed circuits

mixed_circuits contains 4,140 composite layouts. Each sample is stored in its own mixed_<id> directory.

The mixed GDS files are flattened: a sampled file contains one top-level cell without the original device references or device-name labels. Consequently, individual device bounding boxes cannot be recovered from cell names alone.

Single devices

single_devices contains isolated primitive layouts grouped by device class:

Directory Device class Samples
single_cap Capacitor 4,997
single_nmos NMOS 5,000
single_pmos PMOS 5,000
single_res Resistor 5,000
Total 19,997

The device class is encoded by the directory and sample name. The sampled single-device GDS files themselves are flattened and use generic top-cell names such as M0; they do not retain a referenced child cell from which generate_labels.py can infer the device type.

Generating PNG and JSON labels

generate_labels.py reads a hierarchical GDS, identifies device references from their cell names, and writes:

  • A PNG rendering of the GDS geometry with device annotations.
  • A JSON file containing device labels, types, cell names, GDS coordinates, and PNG pixel coordinates.

Requirements

  • Python 3.10 or newer
  • gdstk
  • Pillow
  • The SKY130 KLayout layer-properties file

Install the Python packages if needed:

python3 -m pip install gdstk Pillow

Basic use

Run the script from the dataset root and pass one hierarchical base-circuit GDS:

python3 generate_labels.py \
  base_circuits/hpf/hpf_675/hpf_675.gds

By default, the output files are written beside the input:

base_circuits/hpf/hpf_675/
├── hpf_675.gds
├── hpf_675.json
└── hpf_675.png
Downloads last month
59