ebeamtime 0.1.0

Analytical electron-beam lithography write-time estimation from GDSII layouts

ebeamtime

ebeamtime estimates analytical electron-beam lithography write time from GDSII polygon area, exposure dose, beam current, and optional stage/write-field metadata. It preserves raw polygon-instance area semantics: overlapping source polygons are counted separately because each is written separately.

python -m pip install ebeamtime
ebeamtime layout.gds --exposure 1:0:100:1 --backend cpu

For reproducible local development on WSL, synchronize the committed production-index lock into the platform-specific environment:

UV_PROJECT_ENVIRONMENT=.venv_wsl UV_LINK_MODE=copy uv sync --locked --extra test
TMPDIR=/tmp .venv_wsl/bin/python -m pytest

When qualifying coordinated unpublished changes, a sibling gdsdiff checkout may be installed editable for that local test, but it must not be recorded in package metadata or uv.lock.

Python usage:

from ebeamtime import EbeamLayerExposure, EstimateConfig, LayerSpec
from ebeamtime import estimate_gds_write_time

config = EstimateConfig(
    "layout.gds",
    (EbeamLayerExposure("metal", LayerSpec(1, 0), 100, 1),),
    backend="cpu",
)
report = estimate_gds_write_time(config).report
print(report.total_s)

CUDA and Apple Metal are runtime options. They require compatible hardware and the platform toolchain (nvcc or Xcode command-line tools); CPU requires neither. See ebeamtime-diagnostics and docs/semantics.md. On-demand builds default to NVIDIA's -arch=native, which detects visible GPUs and emits native SASS without PTX. For cross-machine builds, set EBEAMTIME_CUDA_ARCHITECTURES to comma-separated compute capabilities without decimals (for example, 89,120); targets and standard NVCC_*_FLAGS become part of the content-addressed cache key.

Installation and import never compile GPU code. Inspect the shared CUDA toolchain through ebeamtime.inspect_cuda_toolchain(), then build and verify only the estimator kernel with ebeamtime.prepare_cuda() or ebeamtime-prepare-cuda --json. Explicit backend="cuda" use retains lazy preparation. backend="auto" does not probe or compile a GPU backend for workloads below gpu_min_polygons. Metal is experimental until validated on Apple Silicon hardware.

--project-config loads and executes Python. Use it only with trusted project configuration files.

The project is early-stage software and is licensed under GPL-3.0-only.