Demo

1. Demo Overview

This page summarizes representative demo results for GeoSignal Preview.

Try the live Render demo

Upload a public, non-confidential GDS/OAS file and inspect local contour behavior using the lightweight GeoSignal workflow.

Try Live Demo

Demo limits apply. The free server may sleep when idle.

GeoSignal Preview starts from layout geometry, selects candidate ROIs for review, and generates rasterized masks, simplified Abbe-based aerial images, and multi-threshold contours for those selected regions.

The purpose of this demo is not to provide accurate process prediction or an optimized hotspot detector. It is to show how geometry-only review can be extended with optical-response and contour-behavior visualization.

The basic demo flow is:

Layout Geometry
    -> Geometry-based Candidate Filtering
    -> ROI Selection
    -> Rasterized Mask
    -> Abbe-based Aerial Image
    -> Multi-threshold Contour
    -> Hotspot-like Shape Review

GeoSignal demo pipeline

The workflow can be understood in two parts:

Layout Geometry
    -> Width / Space screening
    -> Candidate ROI selection

Candidate ROI
    -> Rasterized Mask
    -> Aerial Image
    -> Multi-threshold Contour
    -> Hotspot-like Shape Review

This demo does not run optical simulation on every layout region. Instead, it first narrows down regions that may deserve review from a geometry viewpoint, then applies optical-model-based contour review to a limited number of ROIs.

Detailed implementation changes behind this workflow are recorded in Release Notes.

2. Candidate Selection in This Demo

In this demo, candidate regions are first identified from a minimum width / space viewpoint. Five ROIs are then selected for aerial-image and contour generation, matching the current service default.

The selected candidates should be understood as representative review examples, not final hotspot judgments. The exact screening rule and ordering method are preview-stage heuristics used to make the demo concrete and computationally manageable.

The important point is the review structure:

Find potentially weak locations from layout geometry
    -> Calculate optical response at those locations
    -> Review shape changes through threshold contours

The current snapshot uses the following reproducible conditions.

Setting	Value
Input layout	`sky130_fd_sc_hd_dfrtp_1.gds`
Target layer / datatype	`67/20`
Width / space threshold	strict `< 0.175 / 0.175 µm`
Geometry engine	`pya-native`
ROI size / pixel size	`2.56 / 0.01 µm`
Imaging	KrF 248 nm, NA 0.68, sigma 0.60, dense7
Selected candidates	`WIDTH_0001`, `SPACE_0001`, `WIDTH_0002`, `SPACE_0002`, `WIDTH_0003`

The pya-native geometry engine and candidate ordering are summarized in the v0.6 Release Notes. Combined review GDS output, candidate-specific contour datatypes, and service memory changes are summarized in the v0.7 Release Notes. The optional focus/dose process-window workflow and its performance refactor are recorded in v0.8 and v0.8.1. The exact Hopkins cross-validation added in v0.9 does not change the Abbe-generated snapshot on this page; see the v0.9 milestone notes.

3. What This Demo Shows

This demo reviews the following items.

Item	Description
Geometry-based Candidate	Candidate region first selected from minimum width / space in the layout
ROI Selection	Selected review region within the current runtime budget
Rasterized Mask	Binary mask generated by rasterizing layout polygons onto a pixel grid
Aerial Image	Optical intensity map calculated using a simplified Abbe-based imaging model
Multi-threshold Contour	Contours extracted at threshold levels 0.20 / 0.30 / 0.40
Hotspot-like Shape Review	Visual review of necking, pinch, corner rounding, line-end pullback, and bridge-like behavior

The key comparison is:

geometry-based candidate
    vs
aerial-image-based optical response
    vs
threshold-contour-based printed-shape-like behavior

The threshold contours used in this demo are not calibrated resist contours. They are qualitative visualization references for observing how the aerial image appears as contour behavior under different threshold levels.

4. Representative Candidate Results

The result below follows the same presentation order as the current Render interface. It was generated locally through the same service path because the public hosted instance is intended for lightweight interactive use.

Input sky130_fd_sc_hd_dfrtp_1.gds Layer / datatype 67 / 20 Backend pya-native Runtime 3.68 s

Analysis Summary

Input shapes32

Raw edge pairs116

Width components27

Space components34

Approximate Printed Metrics and Imaging Condition

Minimum printed width0.169 µmthreshold 0.30

Minimum printed space0.174 µmthreshold 0.30

Optical condition248 nm · NA 0.68sigma 0.60 · dense7 · 29 source points

Main Hotspot Overlay

Selected WIDTH_0002 hotspot overlay

The main overlay centers on WIDTH_0002, the selected ROI with the lowest valid direct-intensity width metric. It combines the aerial image, three threshold contours, the inner ROI, and the worst-width location.

Worst Printed Width and Space

Worst approximate printed width zoom at 0.169 micrometers — Worst approximate printed width: 0.169 µm near `WIDTH_0002`.

Worst approximate printed space zoom at 0.174 micrometers — Worst approximate printed space: 0.174 µm near `SPACE_0001`.

Geometry Candidate Overview

pya-native geometry candidate overview

The overview shows all merged width and space violation regions and the selected candidate anchors. A dense marker field is expected here: it is a geometry screening map, while the images above are the optical review results for the selected ROIs.

5. Common Interpretation Points

The summary, selected overlay, measurement zooms, and geometry overview should be reviewed together.

The main points to check are:

whether edge blur or intensity spreading appears in the aerial image
whether line-end pullback-like behavior appears near line ends
whether corner rounding-like behavior appears around corners
whether necking or pinch-like behavior appears around narrow-width regions
whether bridge-like behavior may appear around narrow-space regions
how much the 0.20 / 0.30 / 0.40 threshold contours move
whether large contour movement aligns with the geometry-based candidate region

The philosophy of this demo is not to make a final hotspot judgment. It is to help reviewers quickly narrow down locations that deserve attention.

6. Current Scope and Limitations

The current demo is a qualitative visualization result for public preview.

It has the following limitations.

The width / space screening criterion is a preview heuristic, not a calibrated process rule.
The candidate selection method is not an optimized hotspot-ranking logic.
Only a small number of representative ROIs are shown due to runtime considerations.
A simplified Abbe-based imaging model is used.
Wafer-data-based calibration is not included.
Resist and etch models are not included.
Threshold contours are used for qualitative comparison and visualization.
CD prediction accuracy is not the goal.
Public or synthetic examples are used.
Use only public, non-confidential GDS/OAS files with the live demo.
Core implementation code is not included in this public preview repository.

Therefore, the current results should be interpreted as:

qualitative visual indicators

not as:

production specifications

7. Feedback Points

Feedback is especially helpful for the following points.

whether multi-threshold contour helps understand contour sensitivity
whether hotspot-like shape observation is intuitive
whether observing necking, corner rounding, line-end pullback, and bridge-like behavior is useful
what additional pattern cases would make the preview clearer

Share feedback through the GeoSignal Preview form