Shapes · Files
Paths in layout: centerline, width, end caps, and bends
A layout path stores a centerline and one width. How end caps and corners are built, when a path turns into a polygon, and how a wrong cap causes an open or a short.
Last edited: September 19, 2026

You want a wire of constant width that turns a corner, and you do not want to push every edge by hand when the width changes.
You click the centerline. You set a width. The editor draws a ribbon. That object is a path. The centerline stays where you put it, and the metal moves when the width changes. A neck in the middle of the route is two paths, or one polygon. One width cannot tell the truth about a narrow segment. The picture at the top of this page is one path at one width: a ribbon that runs right, turns up, and runs right again. A thin line runs down the middle and stops at the bend points. That line is the centerline the file stores. The ribbon runs past those points. The extra length is the end cap.
Lay the end cap against the pad. Too short is an open. Too long is a short. Change the width and both edges move. Change a bend point and the ribbon follows. The centerline does not have to move for the metal to reach a neighbor it did not reach before.
Ends, bends, and the polygon you meant
The second picture sets a path beside other shapes so the differences are visible.

The navy stroke is a straight path. Its ends are squared off by the end-cap rule in use here: the ribbon extends past the first and last centerline points by about half the width, or stops flush, depending on the cap you set. The line you would draw down its middle ends before the cap ends. The gold stroke bends. At the bend the editor builds a corner, mitered, beveled, or rounded, from the same width. The teal rectangle on the right is not a path. It is a box. It will not grow on both sides when you edit a width property, because it has no width property. It has a left, a right, a bottom, and a top.
Paths are a convenience inside the editor. Many file exchanges want outlines. On the way out, the path is expanded into a polygon with the corners of the ribbon. After that expansion, editing a vertex edits the outline, not a centerline. If you still needed the centerline, you expanded too early.
A path of zero width is a line some checkers ignore and some treat as a shape with no area. Do not use it as a wire. A path whose bends sit closer together than the width can fold over itself and make a lump. Look at the ribbon at the top: each leg is long compared with the width, so the corner is a corner and not a blob.
Paths and the shapes people confuse them with
Four sides of a rectangle drawn as four paths are not a box and not a filled polygon. There is no inside. A route that must be filled metal should be one path with a width, or one polygon.
On a board, a track is the same idea under another name: a centerline, a width, and a layer. On a chip, a path is handy for a bus at constant width and a nuisance when the width must neck down in the middle. A neck is either two paths that meet or a polygon you draw on purpose. Forcing one width for the whole route is how a neck never happens.
Where the cap and the neck go wrong
Changing the width and forgetting the end cap grew into a neighbor. The centerline did not move. The metal did. Put the end of the ribbon next to a pad and look at the cap before you call the net closed. If the cap stops short of the pad, the wire is open even though the centerline arrived. If the cap overruns into the next track, the wire is short even though the centerline looked clear.
The navy stroke in the second picture is the place to practice that look. The stored points are the middle. The metal is wider. The extra metal is not a separate object you can delete. Edit the width or the cap. Leave the teal rectangle as a box. The day you need that rectangle to follow a bend, stop stretching it and draw a path.
Converting to a polygon to satisfy an exporter, then converting back and losing the bend points, is a one-way door. You cannot always round-trip. Using a path on a sheet that the design-rule deck measures as an outline with a different width than the one you set is the other trap, because the deck expands paths with its own cap rule. If the deck and the editor disagree about the end, the picture and the report disagree about a short.
If a bus must neck, split the path. The gold bend in the second picture is a constant width on purpose, so the corner is the only surprise. Learn that surprise, miter versus bevel, before you add a neck. Then the ribbon at the top of the page is a route you can edit by moving three points, which is the whole reason a path exists. The cap against the pad, and the neck as its own path or its own polygon, are the two checks that keep that convenience from becoming an open or a short.
Continue exploring
-
Polygons in a layout: vertices, holes, and why a box is not enough
A layout polygon is a closed list of vertices on one layer: how GDSII and OASIS store it, vertex order and area, holes as cut lines, broken polygons, vertex limits, and how to check them in LayoutEditor.
-
Database units: how layout files store coordinates as integers
A GDSII or OASIS file stores every coordinate as a whole number of database units. How that scale becomes microns, and what breaks when tools disagree.
-
Layout grids: display grid, database unit, and manufacturing grid
A corner can sit on the display grid and still miss the manufacturing grid. How the three grids differ, and why an off-grid corner is rounded by the mask writer and can fail a space check.
-
GDSII in plain language
A 1970s suitcase that still carries almost every chip drawing on Earth.
-
OASIS vs GDSII: when GDS gets huge
OASIS (SEMI P39) can store a repeated shape as one record plus a repetition, so contact, via, and fill fields take less space than in GDSII; unique polygons do not get smaller.
-
When the computer moves an edge
Sizing moves every edge of a shape outward or inward by the same distance. Masks, solder openings, and rule checks depend on it, and corners, narrow necks, and unmerged pieces are where it surprises you.
-
Manhattan versus all-angle
Manhattan layouts use only horizontal and vertical edges; 45-degree and all-angle layouts allow slants and curves. The choice changes wire length, how gaps are measured, which corners land on the grid, which rules apply, and what the mask costs.
-
Waveguide layout: bend radius, width, and paths on a photonic chip
A waveguide is a path that light follows. Width sets the mode, a tight bend radiates, and the bend radius is part of the device, not a routing convenience.