The general idea is to implement something resembling the GDSII file-format, but with some vectorized element types (eg. circles, not just polygons) and the ability to output to multiple formats.

Installation

Requirements:

python >= 3.11
numpy
klamath (used for GDSII i/o)

Optional requirements:

ezdxf (DXF i/o): ezdxf
oasis (OASIS i/o): fatamorgana
svg (SVG output): svgwrite
visualization (shape plotting): matplotlib
text (Text shape): matplotlib, freetype

Install with pip:

pip install 'masque[oasis,dxf,svg,visualization,text]'

Overview

A layout consists of a hierarchy of Patterns stored in a single Library. Each Pattern can contain Refs pointing at other patterns, Shapes, and Labels.

masque departs from several "classic" GDSII paradigms:

Layer info for Shapess and Labels is not stored in the individual shape and label objects. Instead, the layer is determined by the key for the container dict (e.g. pattern.shapes[layer]).
- This simplifies many common tasks: filtering Shapes by layer, remapping layers, and checking if a layer is empty.
- Technically, this allows reusing the same shape or label object across multiple layers. This isn't part of the standard workflow since a mixture of single-use and multi-use shapes could be confusing.
- This is similar to the approach used in KLayout
Ref target names are also determined in the key of the container dict (e.g. pattern.refs[target_name]).
- This similarly simplifies filtering Refs by target name, updating to a new target, and checking if a given Pattern is referenced.
Pattern names are set by their containing Library and are not stored in the Pattern objects.
- This guarantees that there are no duplicate pattern names within any given Library.
- Likewise, enumerating all the names (and all the Patterns) in a Library is straightforward.
Each Ref, Shape, or Label can be repeated multiple times by attaching a repetition object to it.
- This is similar to how OASIS reptitions are handled, and provides extra flexibility over the GDSII approach of only allowing arrays through AREF (Ref + repetition).
Labels do not have an orientation or presentation
- This is in line with how they are used in practice, and how they are represented in OASIS.
Non-polygonal Shapes are allowed. For example, elliptical arcs are a basic shape type.
- This enables compatibility with OASIS (e.g. circles) and other formats.
- Shapes provide a .to_polygons() method for GDSII compatibility.
Most coordinate values are stored as 64-bit floats internally.
- 1 earth radii in nanometers (6e15) is still represented without approximation (53 bit mantissa -> 2^53 > 9e15)
- Operations that would otherwise clip/round on are still represented approximately.
- Memory usage is usually dominated by other Python overhead.
Pattern objects also contain Port information, which can be used to "snap" together multiple sub-components by matching up the requested port offsets and rotations.
- Port rotations are defined as counter-clockwise angles from the +x axis.
- Ports point into the interior of their associated device.
- Port rotations may be None in the case of non-oriented ports.
- Ports have a ptype string which is compared in order to catch mismatched connections at build time.
- Ports can be exported into/imported from Labels stored directly in the layout, editable from standard tools (e.g. KLayout). A default format is provided.

Glossary

Library: A collection of named cells. OASIS or GDS "library" or file.
"tree": Any Library which has only one topcell.
Pattern: A collection of geometry, text labels, and reference to other patterns. OASIS or GDS "Cell", DXF "Block".
Ref: A reference to another pattern. GDS "AREF/SREF", OASIS "Placement".
Shape: Individual geometric entity. OASIS or GDS "Geometry element", DXF "LWPolyline" or "Polyline".
repetition: Repetition operation. OASIS "repetition". GDS "AREF" is a Ref combined with a Grid repetition.
Label: Text label. Not rendered into geometry. OASIS, GDS, DXF "Text".
annotation: Additional metadata. OASIS or GDS "property".

Design choices & shorthand

Most syntax and behavior should follow normal python conventions. There are a few exceptions, either meant to catch common mistakes or to provide a shorthand for common operations:

`Library` objects don't allow overwriting already-existing patterns

library['mycell'] = pattern0
library['mycell'] = pattern1   # Error! 'mycell' already exists and can't be overwritten
del library['mycell']          # We can explicitly delete it
library['mycell'] = pattern1   # And now it's ok to assign a new value

Insert a newly-made hierarchical pattern (with children) into a layout

# Let's say we have a function which returns a new library containing one topcell (and possibly children)
tree = make_tree(...)

# To reference this cell in our layout, we have to add all its children to our `library` first:
top_name = tree.top()              # get the name of the topcell
name_mapping = library.add(tree)   # add all patterns from `tree`, renaming elgible conflicting patterns
new_name = name_mapping.get(top_name, top_name)    # get the new name for the cell (in case it was auto-renamed)
my_pattern.ref(new_name, ...)       # instantiate the cell

# This can be accomplished as follows
new_name = library << tree       # Add `tree` into `library` and return the top cell's new name
my_pattern.ref(new_name, ...)       # instantiate the cell

# In practice, you may do lots of
top_pattern.ref(lib << make_tree(...), ...)

Place a hierarchical pattern into a layout, preserving its port info

# As above, we have a function that makes a new library containing one topcell (and possibly children)
tree = make_tree(...)

# We need to go get its port info to `place()` it into our existing layout,
new_name = library << tree          # Add the tree to the library and return its name (see `<<` above)
abstract = library.abstract(tree)   # An `Abstract` stores a pattern's name and its ports (but no geometry)
my_pattern.place(abstract, ...)

# With shorthand,
abstract = library <= tree
my_pattern.place(abstract, ...)

# or
my_pattern.place(library << make_tree(...), ...)

Quickly add geometry, labels, or refs:

The long form for adding elements can be overly verbose:

my_pattern.shapes[layer].append(Polygon(vertices, ...))
my_pattern.labels[layer] += [Label('my text')]
my_pattern.refs[target_name].append(Ref(offset=..., ...))

There is shorthand for the most common elements:

my_pattern.polygon(layer=layer, vertices=vertices, ...)
my_pattern.rect(layer=layer, xctr=..., xmin=..., ymax=..., ly=...)  # rectangle; pick 4 of 6 constraints
my_pattern.rect(layer=layer, ymin=..., ymax=..., xctr=..., lx=...)
my_pattern.path(...)
my_pattern.label(layer, 'my_text')
my_pattern.ref(target_name, offset=..., ...)

Accessing ports

# Square brackets pull from the underlying `.ports` dict:
assert pattern['input'] is pattern.ports['input']

# And you can use them to read multiple ports at once:
assert pattern[('input', 'output')] == {
    'input': pattern.ports['input'],
    'output': pattern.ports['output'],
    }

# But you shouldn't use them for anything except reading
pattern['input'] = Port(...)   # Error!
has_input = ('input' in pattern)   # Error!

TODO

Better interface for polygon operations (e.g. with pyclipper)
- de-embedding
- boolean ops
Tests tests tests
check renderpather
pather and renderpather examples

README.md

Masque README