masque/masque/pattern.py

585 lines
19 KiB
Python

"""
Base object representing a lithography mask.
"""
from typing import List, Callable, Tuple, Dict, Union, Set, Sequence, Optional, Type, overload, cast
from typing import Mapping, MutableMapping, Iterable, TypeVar, Any
import copy
from itertools import chain
from collections import defaultdict
import numpy
from numpy import inf
from numpy.typing import NDArray, ArrayLike
# .visualize imports matplotlib and matplotlib.collections
from .subpattern import SubPattern
from .shapes import Shape, Polygon
from .label import Label
from .utils import rotation_matrix_2d, normalize_mirror, AutoSlots, annotations_t
from .error import PatternError
from .traits import AnnotatableImpl, Scalable, Mirrorable
from .traits import Rotatable, Positionable
P = TypeVar('P', bound='Pattern')
class Pattern(AnnotatableImpl, Mirrorable, metaclass=AutoSlots):
"""
2D layout consisting of some set of shapes, labels, and references to other Pattern objects
(via SubPattern). Shapes are assumed to inherit from masque.shapes.Shape or provide equivalent functions.
"""
__slots__ = ('shapes', 'labels', 'subpatterns')
shapes: List[Shape]
""" List of all shapes in this Pattern.
Elements in this list are assumed to inherit from Shape or provide equivalent functions.
"""
labels: List[Label]
""" List of all labels in this Pattern. """
subpatterns: List[SubPattern]
""" List of all references to other patterns (`SubPattern`s) in this `Pattern`.
Multiple objects in this list may reference the same Pattern object
(i.e. multiple instances of the same object).
"""
def __init__(
self,
*,
shapes: Sequence[Shape] = (),
labels: Sequence[Label] = (),
subpatterns: Sequence[SubPattern] = (),
annotations: Optional[annotations_t] = None,
) -> None:
"""
Basic init; arguments get assigned to member variables.
Non-list inputs for shapes and subpatterns get converted to lists.
Args:
shapes: Initial shapes in the Pattern
labels: Initial labels in the Pattern
subpatterns: Initial subpatterns in the Pattern
"""
if isinstance(shapes, list):
self.shapes = shapes
else:
self.shapes = list(shapes)
if isinstance(labels, list):
self.labels = labels
else:
self.labels = list(labels)
if isinstance(subpatterns, list):
self.subpatterns = subpatterns
else:
self.subpatterns = list(subpatterns)
self.annotations = annotations if annotations is not None else {}
def __copy__(self, memo: Dict = None) -> 'Pattern':
return Pattern(
shapes=copy.deepcopy(self.shapes),
labels=copy.deepcopy(self.labels),
subpatterns=[copy.copy(sp) for sp in self.subpatterns],
annotations=copy.deepcopy(self.annotations),
)
def __deepcopy__(self, memo: Dict = None) -> 'Pattern':
memo = {} if memo is None else memo
new = Pattern(
shapes=copy.deepcopy(self.shapes, memo),
labels=copy.deepcopy(self.labels, memo),
subpatterns=copy.deepcopy(self.subpatterns, memo),
annotations=copy.deepcopy(self.annotations, memo),
)
return new
def append(self: P, other_pattern: P) -> P:
"""
Appends all shapes, labels and subpatterns from other_pattern to self's shapes,
labels, and supbatterns.
Args:
other_pattern: The Pattern to append
Returns:
self
"""
self.subpatterns += other_pattern.subpatterns
self.shapes += other_pattern.shapes
self.labels += other_pattern.labels
return self
def subset(
self,
shapes: Callable[[Shape], bool] = None,
labels: Callable[[Label], bool] = None,
subpatterns: Callable[[SubPattern], bool] = None,
) -> 'Pattern':
"""
Returns a Pattern containing only the entities (e.g. shapes) for which the
given entity_func returns True.
Self is _not_ altered, but shapes, labels, and subpatterns are _not_ copied.
Args:
shapes: Given a shape, returns a boolean denoting whether the shape is a member
of the subset. Default always returns False.
labels: Given a label, returns a boolean denoting whether the label is a member
of the subset. Default always returns False.
subpatterns: Given a subpattern, returns a boolean denoting if it is a member
of the subset. Default always returns False.
Returns:
A Pattern containing all the shapes and subpatterns for which the parameter
functions return True
"""
pat = Pattern()
if shapes is not None:
pat.shapes = [s for s in self.shapes if shapes(s)]
if labels is not None:
pat.labels = [s for s in self.labels if labels(s)]
if subpatterns is not None:
pat.subpatterns = [s for s in self.subpatterns if subpatterns(s)]
return pat
def polygonize(
self: P,
poly_num_points: Optional[int] = None,
poly_max_arclen: Optional[float] = None,
) -> P:
"""
Calls `.to_polygons(...)` on all the shapes in this Pattern, replacing them with the returned polygons.
Arguments are passed directly to `shape.to_polygons(...)`.
Args:
poly_num_points: Number of points to use for each polygon. Can be overridden by
`poly_max_arclen` if that results in more points. Optional, defaults to shapes'
internal defaults.
poly_max_arclen: Maximum arclength which can be approximated by a single line
segment. Optional, defaults to shapes' internal defaults.
Returns:
self
"""
old_shapes = self.shapes
self.shapes = list(chain.from_iterable((
shape.to_polygons(poly_num_points, poly_max_arclen)
for shape in old_shapes)))
return self
def manhattanize(
self: P,
grid_x: ArrayLike,
grid_y: ArrayLike,
) -> P:
"""
Calls `.polygonize()` on the pattern, then calls `.manhattanize()` on all the
resulting shapes, replacing them with the returned Manhattan polygons.
Args:
grid_x: List of allowed x-coordinates for the Manhattanized polygon edges.
grid_y: List of allowed y-coordinates for the Manhattanized polygon edges.
Returns:
self
"""
self.polygonize()
old_shapes = self.shapes
self.shapes = list(chain.from_iterable(
(shape.manhattanize(grid_x, grid_y) for shape in old_shapes)))
return self
def as_polygons(self, library: Mapping[str, 'Pattern']) -> List[NDArray[numpy.float64]]:
"""
Represents the pattern as a list of polygons.
Deep-copies the pattern, then calls `.polygonize()` and `.flatten()` on the copy in order to
generate the list of polygons.
Returns:
A list of `(Ni, 2)` `numpy.ndarray`s specifying vertices of the polygons. Each ndarray
is of the form `[[x0, y0], [x1, y1],...]`.
"""
pat = self.deepcopy().polygonize().flatten(library=library)
return [shape.vertices + shape.offset for shape in pat.shapes] # type: ignore # mypy can't figure out that shapes are all Polygons now
def referenced_patterns(self) -> Set[Optional[str]]:
"""
Get all pattern namers referenced by this pattern. Non-recursive.
Returns:
A set of all pattern names referenced by this pattern.
"""
return set(sp.target for sp in self.subpatterns)
def get_bounds(
self,
library: Optional[Mapping[str, 'Pattern']] = None,
) -> Optional[NDArray[numpy.float64]]:
"""
Return a `numpy.ndarray` containing `[[x_min, y_min], [x_max, y_max]]`, corresponding to the
extent of the Pattern's contents in each dimension.
Returns `None` if the Pattern is empty.
Returns:
`[[x_min, y_min], [x_max, y_max]]` or `None`
"""
if self.is_empty():
return None
min_bounds = numpy.array((+inf, +inf))
max_bounds = numpy.array((-inf, -inf))
for entry in chain(self.shapes, self.labels):
bounds = entry.get_bounds()
if bounds is None:
continue
min_bounds = numpy.minimum(min_bounds, bounds[0, :])
max_bounds = numpy.maximum(max_bounds, bounds[1, :])
if self.subpatterns and (library is None):
raise PatternError('Must provide a library to get_bounds() to resolve subpatterns')
for entry in self.subpatterns:
bounds = entry.get_bounds(library=library)
if bounds is None:
continue
min_bounds = numpy.minimum(min_bounds, bounds[0, :])
max_bounds = numpy.maximum(max_bounds, bounds[1, :])
if (max_bounds < min_bounds).any():
return None
else:
return numpy.vstack((min_bounds, max_bounds))
def get_bounds_nonempty(
self,
library: Optional[Mapping[str, 'Pattern']] = None,
) -> NDArray[numpy.float64]:
"""
Convenience wrapper for `get_bounds()` which asserts that the Pattern as non-None bounds.
Returns:
`[[x_min, y_min], [x_max, y_max]]`
"""
bounds = self.get_bounds(library)
assert(bounds is not None)
return bounds
def translate_elements(self: P, offset: ArrayLike) -> P:
"""
Translates all shapes, label, and subpatterns by the given offset.
Args:
offset: (x, y) to translate by
Returns:
self
"""
for entry in chain(self.shapes, self.subpatterns, self.labels):
entry.translate(offset)
return self
def scale_elements(self: P, c: float) -> P:
""""
Scales all shapes and subpatterns by the given value.
Args:
c: factor to scale by
Returns:
self
"""
for entry in chain(self.shapes, self.subpatterns):
entry.scale_by(c)
return self
def scale_by(self: P, c: float) -> P:
"""
Scale this Pattern by the given value
(all shapes and subpatterns and their offsets are scaled)
Args:
c: factor to scale by
Returns:
self
"""
entry: Scalable
for entry in chain(self.shapes, self.subpatterns):
entry.offset *= c
entry.scale_by(c)
if entry.repetition:
entry.repetition.scale_by(c)
for label in self.labels:
label.offset *= c
if label.repetition:
label.repetition.scale_by(c)
return self
def rotate_around(self: P, pivot: ArrayLike, rotation: float) -> P:
"""
Rotate the Pattern around the a location.
Args:
pivot: (x, y) location to rotate around
rotation: Angle to rotate by (counter-clockwise, radians)
Returns:
self
"""
pivot = numpy.array(pivot)
self.translate_elements(-pivot)
self.rotate_elements(rotation)
self.rotate_element_centers(rotation)
self.translate_elements(+pivot)
return self
def rotate_element_centers(self: P, rotation: float) -> P:
"""
Rotate the offsets of all shapes, labels, and subpatterns around (0, 0)
Args:
rotation: Angle to rotate by (counter-clockwise, radians)
Returns:
self
"""
for entry in chain(self.shapes, self.subpatterns, self.labels):
entry.offset = numpy.dot(rotation_matrix_2d(rotation), entry.offset)
return self
def rotate_elements(self: P, rotation: float) -> P:
"""
Rotate each shape and subpattern around its center (offset)
Args:
rotation: Angle to rotate by (counter-clockwise, radians)
Returns:
self
"""
for entry in chain(self.shapes, self.subpatterns):
cast(Rotatable, entry).rotate(rotation)
return self
def mirror_element_centers(self: P, axis: int) -> P:
"""
Mirror the offsets of all shapes, labels, and subpatterns across an axis
Args:
axis: Axis to mirror across
(0: mirror across x axis, 1: mirror across y axis)
Returns:
self
"""
for entry in chain(self.shapes, self.subpatterns, self.labels):
entry.offset[axis - 1] *= -1
return self
def mirror_elements(self: P, axis: int) -> P:
"""
Mirror each shape and subpattern across an axis, relative to its
offset
Args:
axis: Axis to mirror across
(0: mirror across x axis, 1: mirror across y axis)
Returns:
self
"""
for entry in chain(self.shapes, self.subpatterns):
cast(Mirrorable, entry).mirror(axis)
return self
def mirror(self: P, axis: int) -> P:
"""
Mirror the Pattern across an axis
Args:
axis: Axis to mirror across
(0: mirror across x axis, 1: mirror across y axis)
Returns:
self
"""
self.mirror_elements(axis)
self.mirror_element_centers(axis)
return self
def scale_element_doses(self: P, c: float) -> P:
"""
Multiply all shape and subpattern doses by a factor
Args:
c: Factor to multiply doses by
Return:
self
"""
for entry in chain(self.shapes, self.subpatterns):
entry.dose *= c
return self
def copy(self: P) -> P:
"""
Return a copy of the Pattern, deep-copying shapes and copying subpattern
entries, but not deep-copying any referenced patterns.
See also: `Pattern.deepcopy()`
Returns:
A copy of the current Pattern.
"""
return copy.copy(self)
def deepcopy(self: P) -> P:
"""
Convenience method for `copy.deepcopy(pattern)`
Returns:
A deep copy of the current Pattern.
"""
return copy.deepcopy(self)
def is_empty(self) -> bool:
"""
Returns:
True if the pattern is contains no shapes, labels, or subpatterns.
"""
return (len(self.subpatterns) == 0
and len(self.shapes) == 0
and len(self.labels) == 0)
def addsp(self: P, *args: Any, **kwargs: Any) -> P:
"""
Convenience function which constructs a subpattern object and adds it
to this pattern.
Args:
*args: Passed to `SubPattern()`
**kwargs: Passed to `SubPattern()`
Returns:
self
"""
self.subpatterns.append(SubPattern(*args, **kwargs))
return self
def flatten(
self: P,
library: Mapping[str, P],
) -> 'Pattern':
"""
Removes all subpatterns (recursively) and adds equivalent shapes.
Alters the current pattern in-place
Args:
library: Source for referenced patterns.
Returns:
self
"""
flattened: Dict[Optional[str], Optional[P]] = {}
def flatten_single(name: Optional[str]) -> None:
if name is None:
pat = self
else:
pat = library[name].deepcopy()
flattened[name] = None
for subpat in pat.subpatterns:
target = subpat.target
if target is None:
continue
if target not in flattened:
flatten_single(target)
if flattened[target] is None:
raise PatternError(f'Circular reference in {name} to {target}')
p = subpat.as_pattern(pattern=flattened[target])
pat.append(p)
pat.subpatterns.clear()
flattened[name] = pat
flatten_single(None)
return self
def visualize(
self: P,
library: Optional[Mapping[str, P]] = None,
offset: ArrayLike = (0., 0.),
line_color: str = 'k',
fill_color: str = 'none',
overdraw: bool = False,
) -> None:
"""
Draw a picture of the Pattern and wait for the user to inspect it
Imports `matplotlib`.
Note that this can be slow; it is often faster to export to GDSII and use
klayout or a different GDS viewer!
Args:
offset: Coordinates to offset by before drawing
line_color: Outlines are drawn with this color (passed to `matplotlib.collections.PolyCollection`)
fill_color: Interiors are drawn with this color (passed to `matplotlib.collections.PolyCollection`)
overdraw: Whether to create a new figure or draw on a pre-existing one
"""
# TODO: add text labels to visualize()
from matplotlib import pyplot # type: ignore
import matplotlib.collections # type: ignore
if self.subpatterns and library is None:
raise PatternError('Must provide a library when visualizing a pattern with subpatterns')
offset = numpy.array(offset, dtype=float)
if not overdraw:
figure = pyplot.figure()
pyplot.axis('equal')
else:
figure = pyplot.gcf()
axes = figure.gca()
polygons = []
for shape in self.shapes:
polygons += [offset + s.offset + s.vertices for s in shape.to_polygons()]
mpl_poly_collection = matplotlib.collections.PolyCollection(
polygons,
facecolors=fill_color,
edgecolors=line_color,
)
axes.add_collection(mpl_poly_collection)
pyplot.axis('equal')
for subpat in self.subpatterns:
subpat.as_pattern(library=library).visualize(
library=library,
offset=offset,
overdraw=True,
line_color=line_color,
fill_color=fill_color,
)
if not overdraw:
pyplot.xlabel('x')
pyplot.ylabel('y')
pyplot.show()
def __repr__(self) -> str:
return (f'<Pattern: sh{len(self.shapes)} sp{len(self.subpatterns)} la{len(self.labels)}>')