Wafer assembly

Wafer assembly packs square dies on a round wafer, names every die and puts a chip layout on each one. The result is one wafer OAS file with the wafer outline, the die labels, the alignment marks and your chips. Use it when a set of chips goes to the fab on one wafer, or when you want to know how many dies of a given size fit.

The code is quantcore_circuits.layouts.wafers.wafer (class Wafer). The demo notebook is examples/wafer/wafer_assembly.ipynb.

A first wafer

from quantcore_circuits.layouts.wafers.wafer import Wafer

w = Wafer(diameter_mm=100, die_mm=7.5)        # 4 inch, dense, notch on the bottom edge
print(w.die_count)                            # 100
w.write("build/wafer/wafer.oas", chips="examples/designs/legacy/chip_06_test_cyro_tlq.json")
print(w.table_markdown(limit=5))              # die, size, centre, chip

write builds the wafer, writes the OAS and a .csv with the same die table next to it.

Parameters

All lengths are in mm. The layout itself is written in µm, like every other Quantcore layout. The wafer centre is the origin, x points right, y points up.

Parameter

Default

Meaning

diameter_mm

100

wafer diameter: 100 (4 in) and 150 (6 in) are the presets, any value works

die_mm

7.5

die side (square): 5, 7.5, 10 or any size

street_mm

0.1

dicing street between neighbouring dies

edge_mm

3.0

edge exclusion: no die closer than this to the rim, the notch or the flat

orientation

"notch"

"notch" (V-notch, 1 mm deep), "flat" (32.5 mm on 100 mm, 57.5 mm on 150 mm) or "none"

orientation_side

"bottom"

"bottom", "right", "top" or "left"

packing

"dense"

"dense" or "centred" (see below)

radius_mm

none

radius of the packing circle, needed for packing="centred"

offset_mm

(0, 0)

x, y shift of the packing centre from the wafer centre

align

"auto"

how the die grid sits on the packing centre: "centre" (a die centred there), "corner" (a street crossing there), "row", "column", or "auto" (the one that fits most dies)

bands

none

mixed die sizes: [(die_mm, n_rows), ...]

die_outline, labels

on

draw a thin frame and the name of every die on 501/0

The two packings

Dense fills the whole wafer. A die is placed where it lies completely inside the usable disc: the wafer radius minus edge_mm, and clear of the notch (the notch is cut out of the outline before the edge exclusion is applied).

Centred fills only a circle of radius_mm around the packing centre (the wafer centre plus offset_mm). A die counts when all four corners are inside that circle and it is also inside the usable disc. The die grid is laid out around the packing centre, so the pattern follows the offset. Use it for a small run in the middle of the wafer, or a run away from a bad wafer region.

Mixed sizes (bands): rows of one die size are stacked from top to bottom, the stack is centred on the packing centre, and each row takes as many dies as fit. The demo uses [(5, 2), (7.5, 2), (10, 3), (7.5, 2), (5, 2)].

Naming

Rows are letters from the top (A, B … Z, AA …), columns are numbers from the left (1, 2 …). A die is <row><column>, for example F7.

On a one-size wafer the column number counts from the leftmost occupied column of the whole wafer, so a short top row starts at a higher number, like a wafer map. On a mixed wafer the rows do not share a grid, so each row is counted from its own leftmost die.

Layers

Layer

PDK name

What is drawn on it

500/0

wafer_boundary

the wafer outline: a 200 µm ring just inside the rim, with the notch or flat

501/0

wafer_markers

die labels and thin die frames, and a small cross at each axis point in the edge band (alignment and orientation marks; none at the notch)

The chips keep their own layers. Earlier versions of the PDK used 500/0 and 501/0 for these two; the layer map is on the PDK page.

Placing chips

chips= takes a source for the dies:

Value

Meaning

one source

the same chip on every die

{7.5: src, 10: src2}

by die size in mm ("*" is the default)

{"B3": src}

by die name

lambda die: src or None

a function of the die

A source is a design .json (built headless), an .oas / .gds file, or a BuildResult from quantcore_circuits.assembly.build. Each distinct chip is loaded once and placed as a sub-cell. The chip is centred on its die using its 100/0 chip boundary; a chip larger than its die raises an error. Dies without a chip stay empty (label and frame only). The design JSON files are only read.

Pictures

Wafer.plot_map(path) draws the die map (outline, notch, names) with matplotlib. KLayout pictures come from quantcore_circuits.live.snapshot.snapshot(oas, png, box=...); box (µm) zooms to a region. Both are used in the notebook.

Four inch

4 in, dense, 7.5 mm

Dense, 7.5 mm dies, 100 dies, the Quantcore cryogenic test chip on every die. The notch is at the bottom.

Die map, 4 in dense

4 in, centred, radius 30 mm

Centred, radius 30 mm, offset (6, -4) mm: 37 dies.

4 in mixed

Mixed sizes: 42 dies of 5 mm, 38 of 7.5 mm, 25 of 10 mm (105 dies). The three chips are a 4.5 mm test-pad chip, the cryogenic test chip and the Metal four-qubit chip (9.0 x 6.5 mm).

Zoom on two dies

A zoom on two dies: the label is the top-left corner of the die on 501/0, the green frame is the die outline.

Zoom on dies of two sizes

Six inch

6 in, dense, 10 mm

150 mm wafer, dense, 10 mm dies: 137 dies with the Metal four-qubit chip on each.

6 in, centred, 7.5 mm

150 mm wafer, centred, 7.5 mm dies, radius 45 mm: 89 dies.

Die counts

Edge exclusion 3 mm, street 0.1 mm, notch at the bottom.

Wafer

Die

Packing

Dies

4 in

7.5 mm

dense

100

4 in

7.5 mm

centred, radius 30 mm, offset (6, -4)

37

4 in

5 mm

dense

238

4 in

10 mm

dense

52

4 in

5 / 7.5 / 10 mm

bands (2, 2, 3, 2, 2 rows)

105

6 in

10 mm

dense

137

6 in

7.5 mm

centred, radius 45 mm

89

Run the notebook

cd examples/wafer
QCC_LIVE=0 jupyter lab wafer_assembly.ipynb

It builds the chips headless, writes every wafer to build/wafer/ (OAS, CSV table, PNG pictures) and does not touch examples/designs/. The tests are pytest tests/test_wafer.py.

Limits

  • Dies are squares on a regular grid; there is no stepper-field or reticle model.

  • The notch and flat are cut into the outline and kept clear by the edge exclusion; no SEMI notch tolerances are checked.

  • The wafer OAS is written by Wafer.write, not by qcc mask.