한국어로 보기

v0.7 Release Notes

Main Focus

v0.7 makes the pya-native analysis workflow easier to run within a bounded service memory budget and easier to inspect in a layout viewer.

The v0.6 integer-DBU geometry engine remains the canonical geometry path. This release focuses on review output organization and service execution around that engine.

Combined Review GDS

The service now produces one combined review GDS that preserves the original layout and adds the analysis overlays on high-numbered review layers.

Content Layer / datatype
Width violation components 10001/0
Space violation components 10002/0
Threshold 0.20 contours 10020/1, 10020/2, …
Threshold 0.30 contours 10030/1, 10030/2, …
Threshold 0.40 contours 10040/1, 10040/2, …
Inner ROI outlines 10050/1, 10050/2, …

The datatype identifies the ROI candidate. This keeps contours generated from different marker positions and ambit windows separate, even when a small layout causes several candidate windows to overlap.

Bounded-Memory Abbe Imaging

The service now accumulates each coherent source contribution directly into the final aerial image and releases its intermediate arrays before processing the next source point.

Experiments and debugging retain an explicit option to store every source’s pupil, filtered spectrum, field, and intensity. The numerical accumulation order is unchanged between the two modes.

For a local 320 x 320 dense7 comparison:

Mode Stored source results Tracked peak allocation
Service streaming 0 16.54 MiB
Debug retention 29 164.22 MiB

The aerial-image checksum was identical in the comparison. These figures are local Python allocation measurements, not a general hosting guarantee or total process-RSS measurement.

Resource Limit Update

The pya-native path and legacy gdstk path now use separate marker limits:

  • pya-native: 12,000 raw edge pairs per request
  • legacy gdstk: 5,000 generated raw marker polygons per request

The distinction preserves the former gdstk safety boundary while allowing the optimized pya-native path to analyze the public tt04_pwm.gds example on layer/datatype 67/20.

The local validation case produced 10,814 raw pya-native edge pairs and 6,281 merged violation components. At 0.01 um with five ROI candidates, the full service analysis completed in approximately 13.2 seconds in the development environment.

Pages Result Refresh

The Home and Demo pages now use a result set generated through the current service path and presented in the same order as the Render result interface. The explanatory Demo process-flow image remains unchanged.

The published snapshot uses:

  • sky130_fd_sc_hd_dfrtp_1.gds, layer/datatype 67/20
  • strict < 0.175 um width and space screening
  • 2.56 um ROI size and 0.01 um pixels
  • five alternating width / space ROI candidates
  • KrF 248 nm, NA 0.68, sigma 0.60, dense7 imaging

The refresh run completed in 3.68 seconds on the local workstation and reported 27 width components, 34 space components, an approximate minimum printed width of 0.169 um, and an approximate minimum printed space of 0.174 um. Runtime varies by environment; the result values and images are the relevant published snapshot.

The TinyTapeout input was also rerun with the stricter screening threshold. Its first five candidates remained the same 0.170 um geometry candidates, so the smaller threshold reduced the total component count without producing a more representative weak optical result.

Local comparison ROI Raster Approx. width / space
tt04_pwm.gds, < 0.175 um 2.56 um 320 x 320 0.259 / 0.197 um
tt04_pwm.gds, < 0.175 um 5.00 um 564 x 564 0.254 / 0.194 um
sky130_fd_sc_hd_dfrtp_1.gds, < 0.175 um 2.56 um 320 x 320 0.169 / 0.174 um

The 5.00 um TinyTapeout comparison used a local raster-axis allowance above the public service limit. It provides more surrounding context, but the sky130 result remains the clearer weak-pattern example for Home and Demo.

The previous manually annotated Home image and four representative dense11 ROI images are preserved in the v0.4 visual gallery.

Validation

  • Verified candidate-specific contour datatypes with sky130_fd_sc_hd_dfrtp_1.gds.
  • Verified tt04_pwm.gds layer/datatype 67/20 at 0.01 um.
  • Regenerated the Pages visual set with the current Render-aligned service path.
  • Confirmed that streaming and retained modes produce the same aerial image.
  • Passed all 41 automated tests in the GeoSignal application repository.

Current Limits

The interpretation limits remain unchanged:

  • results remain qualitative
  • the optical model is simplified
  • wafer-data calibration, resist, and etch models are not included
  • printed metrics remain approximate contour-based indicators
  • EPE remains outside the current scope