# SKY130 Open MPW mask-layer renders **Start at [masks.html](masks.html)** - every run against every mask layer, each cell linking to that render's 40-die grid with the project, die size and shape count behind each image. There is a page per shuttle run (`mpw-001.html` ...) and one inside every render directory (`_/_.html`). They are plain HTML: no JavaScript, no server. One directory per (shuttle run, mask layer): the drawn shapes that go on that photomask, rendered from the tapeout layouts the 40 projects of that run submitted. _/__reticle.png all 40 dies, stitched at the manifest's frame positions (A1..F7), 0.5 um/pixel __reticle_overview.png the same, scaled to fit 4096 px __tile_XY.png one grid cell of the reticle, 42 of them __tiles.json where each tile belongs, and its checksum reconstruct_reticle.py rebuilds the reticle PNG from the tiles __slot-NNN_XY.png one die, 0.125 um/pixel, white = mask shapes __slot-NNN_XY.json that die's extent, shape count and source path job.json / result.json what was rendered and from where MD5SUMS / SHA1SUMS / SHA256SUMS checksums for that directory's own files ## The full-resolution reticles are 3.5 gigapixels Every `_reticle.png` is 53432 x 65056 pixels. The files are small - a few MB for a sparse mask layer - but the *dimensions* are the problem: decoding one needs around 3.5 GB of memory however small the file is, and most image viewers refuse outright. So each reticle is also published as **42 tiles**, a 7 x 6 grid of about 7576 x 10776 pixels each, named by the frame position they cover (`_tile_A1.png` ... `_tile_F7.png`). The cuts fall in the gaps between dies, never through one, so **every tile holds whole projects** - tile `D3` is exactly the die in frame D3, with the surrounding scribe. Which project that is is in `__tiles.json`, alongside each tile's position, size and SHA-256. Two of the 42 cells hold no project: those are SkyWater's process test structures, which are not in the tapeout manifest and so are not rendered. They are still published, just unattributed. ### Putting the big image back together `reconstruct_reticle.py` rebuilds the full-resolution PNG from the tiles, **byte for byte**: python3 reconstruct_reticle.py It needs nothing but Python 3 itself - no pip install, no Pillow, no numpy. Run it in a directory holding the tiles and the `_tiles.json` manifest and it finds them itself. It checks every tile's SHA-256 as it reads it, and the finished file's SHA-256 against the manifest. This works because both the tiles and the full image are written with PNG's *Up* filter, which predicts each pixel from the one above it and never looks left - so a row of tiles' scanlines laid end to end *is* the full image's scanline. Rebuilding moves bytes; it does no pixel arithmetic, which is why a pure-Python script can do it at a sensible speed. If your Python's zlib compresses differently from the one that produced these files (some distributions ship zlib-ng), the rebuilt file will have identical pixels but different bytes. The script detects exactly that case and tells you so, rather than reporting a failure: it verifies the uncompressed PNG data stream, which matches regardless. ## Checksums Every directory carries `MD5SUMS`, `SHA1SUMS` and `SHA256SUMS` listing its own files, in the format the standard tools read: sha256sum -c SHA256SUMS They cover only that directory, never its subdirectories, so they stay valid for whatever subset of the tree you actually downloaded. ## Provenance Sources are the public shuttle repositories under https://foss-eda-tools.googlesource.com/third_party/shuttle/sky130/ ; every slot's layout is named by `repo_path` in the JSON, so any image here can be traced back to the tapeout it came from. Layer numbers are SKY130 drawn layers; fab-derived masks (HVTPM, LVTNM, NCM, NTM, HVNTM) are evaluated as a boolean expression over drawn layers, recorded as `expr`. These are renders of *drawn* data, not photomask artwork: reticle pitch, 4x reduction, mirroring and the frame features the fab adds are not modelled. Empty images are real results - several masks are used by no project on a given shuttle. Generated with sky130-mask-decoder. The custody field recording who holds a physical mask set is omitted from the metadata published here.