PDK - GDSFactory

gdsfactory includes a generic Process Design Kit (PDK), that you can use as an inspiration to create your own.

A process design kit (PDK) includes:

LayerStack: different layers with different thickness, z-position, materials and colors.
Design rule checking deck DRC: Manufacturing rules capturing min feature size, min spacing ... for the process.
A library of Fixed or Parametric cells.

The PDK allows you to register:

cell parametric cells that return Components from a ComponentSpec (string, Component, ComponentFactory or dict). Also known as parametric cell functions.
cross_section functions that return CrossSection from a CrossSection Spec (string, CrossSection, CrossSectionFactory or dict).
layers that return a GDS Layer (gdslayer, gdspurpose) from a string, an int or a Tuple[int, int].

Thanks to activating a PDK you can access components, cross_sections or layers using a string, a function or a dict.

Depending on the active pdk:

get_layer returns a Layer from the registered layers.
get_component returns a Component from the registered cells.
get_cross_section returns a CrossSection from the registered cross_sections.

layers¶

GDS layers are a tuple of two integer number gdslayer/gdspurpose

You can define all the layers from your PDK:

From a Klayout lyp (layer properties file).
From scratch, adding all your layers into a class.

Lets generate the layers definition code from a KLayout lyp file.

import pathlib
from functools import partial

import pytest
from pytest_regressions.data_regression import DataRegressionFixture

import gdsfactory as gf
from gdsfactory.difftest import difftest
from gdsfactory.technology import (
    LayerMap,
)
from gdsfactory.typings import Layer

nm = 1e-3

class LayerMapDemo(LayerMap):
    WG: Layer = (1, 0)
    DEVREC: Layer = (68, 0)
    PORT: Layer = (1, 10)
    PORTE: Layer = (1, 11)
    LABEL_INSTANCES: Layer = (206, 0)
    LABEL_SETTINGS: Layer = (202, 0)
    LUMERICAL: Layer = (733, 0)
    M1: Layer = (41, 0)
    M2: Layer = (45, 0)
    M3: Layer = (49, 0)
    N: Layer = (20, 0)
    NP: Layer = (22, 0)
    NPP: Layer = (24, 0)
    OXIDE_ETCH: Layer = (6, 0)
    P: Layer = (21, 0)
    PDPP: Layer = (27, 0)
    PP: Layer = (23, 0)
    PPP: Layer = (25, 0)
    PinRec: Layer = (1, 10)
    PinRecM: Layer = (1, 11)
    SHALLOWETCH: Layer = (2, 6)
    SILICIDE: Layer = (39, 0)
    SIM_REGION: Layer = (100, 0)
    SITILES: Layer = (190, 0)
    SLAB150: Layer = (2, 0)
    SLAB150CLAD: Layer = (2, 9)
    SLAB90: Layer = (3, 0)
    SLAB90CLAD: Layer = (3, 1)
    SOURCE: Layer = (110, 0)
    TE: Layer = (203, 0)
    TEXT: Layer = (66, 0)
    TM: Layer = (204, 0)
    Text: Layer = (66, 0)
    VIA1: Layer = (44, 0)
    VIA2: Layer = (43, 0)
    VIAC: Layer = (40, 0)
    WGCLAD: Layer = (111, 0)
    WGN: Layer = (34, 0)
    WGclad_material: Layer = (36, 0)


LAYER = LayerMapDemo

some generic components use some

Layer	Purpose
LABEL_INSTANCE	for adding instance labels on `gf.read.from_yaml`
MTOP	for top metal routing

class LayersConvenient(LayerMap):
    LABEL_INSTANCE: Layer = (66, 0)

cross_sections¶

You can create a CrossSection from scratch or you can customize the cross_section functions in gf.cross_section

from gdsfactory.cross_section import CrossSection, cross_section, xsection
from gdsfactory.typings import LayerSpec


@xsection
def strip2(
    width: float = 0.5,
    layer: LayerSpec = (2, 0),
    radius: float = 10.0,
    radius_min: float = 5,
    **kwargs,
) -> CrossSection:
    """Return Strip cross_section."""
    return cross_section(
        width=width,
        layer=layer,
        radius=radius,
        radius_min=radius_min,
        **kwargs,
    )

c = gf.components.straight(cross_section=strip2)
c.plot()

@xsection
def pin(
    width: float = 0.5,
    layer: LayerSpec = "WG",
    radius: float = 10.0,
    radius_min: float = 5,
    layer_p: LayerSpec = (21, 0),
    layer_n: LayerSpec = (20, 0),
    width_p: float = 2,
    width_n: float = 2,
    offset_p: float = 1,
    offset_n: float = -1,
    **kwargs,
) -> CrossSection:
    """Return PIN cross_section."""
    sections = (
        gf.Section(layer=layer_p, width=width_p, offset=offset_p),
        gf.Section(layer=layer_n, width=width_n, offset=offset_n),
    )

    return cross_section(
        width=width,
        layer=layer,
        radius=radius,
        radius_min=radius_min,
        sections=sections,
        **kwargs,
    )


c = gf.components.straight(cross_section=pin)
c.plot()

@xsection
def strip_wide(
    width: float = 3,
    layer: LayerSpec = (2, 0),
    radius: float = 10.0,
    radius_min: float = 5,
    **kwargs,
) -> CrossSection:
    """Return Strip cross_section."""
    return cross_section(
        width=width,
        layer=layer,
        radius=radius,
        radius_min=radius_min,
        **kwargs,
    )

strip = gf.cross_section.strip

cross_sections = dict(strip_wide=strip_wide, pin=pin, strip=strip)

cells¶

Cells are functions that return Components. They are parametrized and accept also cells as parameters, so you can build many levels of complexity. Cells are also known as PCells or parametric cells.

You can customize the function default arguments easily thanks to functools.partial Lets customize the default arguments of a library of cells.

For example, you can make some wide MMIs for a particular technology. Lets say the best MMI width you found it to be 9um.

def mmi1x2(width_mmi: float = 9, **kwargs) -> gf.Component:
    c = gf.components.mmi1x2(width_mmi=width_mmi)
    return c


def mmi2x2(width_mmi: float = 9, **kwargs) -> gf.Component:
    c = gf.components.mmi2x2(width_mmi=width_mmi)
    return c


cells = dict(mmi1x2=mmi1x2, mmi2x2=mmi2x2)

You can define a new PDK by creating function that customize partial parameters of the generic functions.

Lets say that this PDK uses layer (41, 0) for the pads (instead of the layer used in the generic pad function).

pad_custom_layer = partial(gf.components.pad, layer=(41, 0))
c = pad_custom_layer()
c.plot()

PDK¶

You can register Layers, ComponentFactories (Parametric cells) and CrossSectionFactories (cross_sections) into a PDK. Then you can access them by a string after you activate the pdk PDK.activate().

LayerSpec¶

You can access layers from the active Pdk using the layer name or a tuple/list of two numbers.

from gdsfactory.generic_tech import get_generic_pdk

generic_pdk = get_generic_pdk()

pdk1 = gf.Pdk(
    name="fab1",
    layers=LAYER,
    cross_sections=cross_sections,
    cells=cells,
    layer_views=generic_pdk.layer_views,
)
pdk1.activate()

pdk1.get_layer("WG")

<LayerMapDemo.WG: 1>

pdk1.get_layer((1, 0))

1

CrossSectionSpec¶

You can access cross_sections from the pdk from the cross_section name, or using a dict to customize the CrossSection

pdk1.get_cross_section("pin")

CrossSection(sections=(Section(width=0.5, offset=0.0, insets=None, layer='WG', port_names=('o1', 'o2'), port_types=('optical', 'optical'), name='_default', hidden=False, simplify=None, width_function=None, offset_function=None), Section(width=2.0, offset=1.0, insets=None, layer=(21, 0), port_names=(None, None), port_types=('optical', 'optical'), name='s_24eda351', hidden=False, simplify=None, width_function=None, offset_function=None), Section(width=2.0, offset=-1.0, insets=None, layer=(20, 0), port_names=(None, None), port_types=('optical', 'optical'), name='s_4e8950a1', hidden=False, simplify=None, width_function=None, offset_function=None)), components_along_path=(), radius=10.0, radius_min=5.0, bbox_layers=None, bbox_offsets=None)

cross_section_spec_string = "pin"
c = gf.components.straight(cross_section=cross_section_spec_string)
c.plot()

xs = gf.get_cross_section("pin", width=2)
wg_pin = gf.components.straight(cross_section=xs)
wg_pin.plot()

ComponentSpec¶

You can get Component from the active pdk using the cell name (string) or a dict.

c = pdk1.get_component("mmi1x2")
c.plot()

c = pdk1.get_component(dict(component="mmi1x2", settings=dict(length_mmi=10)))
c.plot()

Testing PDK cells¶

To make sure all your PDK PCells produce the components that you want, it’s important to test your PDK cells.

As you write your own cell functions you want to make sure you do not break or produced unwanted regressions later on. You should write tests for this.

Make sure you create a test_components.py file for pytest to test your PDK. See for example the tests in the ubc PDK

Pytest-regressions automatically creates the CSV and YAML files for you, as well gdsfactory.gdsdiff will store the reference GDS in ref_layouts and check for geometry differences using XOR.

gdsfactory is not backwards compatible, which means that the package will keep improving and evolving.

To make your work stable you should install a specific version and pin the version in your requirements.txt or pyproject.toml as gdsfactory==9.5.0 replacing 9.5.0 by whatever version you end up using.
Before you upgrade gdsfactory to a newer version make sure your tests pass to make sure that things behave as expected

"""This code tests all your cells in the PDK

it will test:

1. difftest: will test the GDS geometry of a new GDS compared to a reference.
2. settings test: will compare the settings in YAML with a reference YAML.

"""

try:
    dirpath = pathlib.Path(__file__).absolute().with_suffix(".gds")
except Exception:
    dirpath = pathlib.Path.cwd()


component_names = list(pdk1.cells.keys())
factory = pdk1.cells


@pytest.fixture(params=component_names, scope="function")
def component_name(request) -> str:
    return request.param


def test_gds(component_name: str) -> None:
    """Avoid regressions in GDS files. Runs XOR and computes the area."""
    component = factory[component_name]()
    test_name = f"fabc_{component_name}"
    difftest(component, test_name=test_name, dirpath=dirpath)


def test_settings(component_name: str, data_regression: DataRegressionFixture) -> None:
    """Avoid regressions in component settings and ports."""
    component = factory[component_name]()
    data_regression.check(component.to_dict())

Compare gds files¶

You can use the command line gf gds diff gds1.gds gds2.gds to overlay gds1.gds and gds2.gds files and show them in KLayout.

For example, if you changed the mmi1x2 and made it 5um longer by mistake, you could gf gds diff ref_layouts/mmi1x2.gds run_layouts/mmi1x2.gds and see the GDS differences in Klayout.

help(gf.diff)

Help on function diff in module gdsfactory.difftest:

diff(ref_file: pathlib.Path | str, run_file: pathlib.Path | str, xor: bool = True, test_name: str = '', ignore_sliver_differences: bool | None = None, ignore_cell_name_differences: bool | None = None, ignore_label_differences: bool | None = None, show: bool = False, stagger: bool = True) -> bool
    Returns True if files are different, prints differences and shows them in klayout.
    
    Args:
        ref_file: reference (old) file.
        run_file: run (new) file.
        xor: runs xor on every layer between old and run files.
        test_name: prefix for the new cell.
        ignore_sliver_differences: if True, ignores any sliver differences in the XOR result. If None (default), defers to the value set in CONF.difftest_ignore_sliver_differences
        ignore_cell_name_differences: if True, ignores any cell name differences. If None (default), defers to the value set in CONF.difftest_ignore_cell_name_differences
        ignore_label_differences: if True, ignores any label differences when run in XOR mode. If None (default) defers to the value set in CONF.difftest_ignore_label_differences
        show: shows diff in klayout.
        stagger: if True, staggers the old/new/xor views. If False, all three are overlaid.

mmi1 = gf.components.mmi1x2(length_mmi=5)
mmi2 = gf.components.mmi1x2(length_mmi=6)
gds1 = mmi1.write_gds()
gds2 = mmi2.write_gds()
gf.diff(gds1, gds2)

cell name differs mmi1x2_gdsfactorypcompo_85778fc7
Running XOR on differences...

: XOR difference on layer WG (1/0)

True