ebeamtime 0.1.2

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

ebeamtime

ebeamtime estimates electron-beam lithography write time from the polygon area, exposure dose, and beam current in a GDSII file.

Install

python -m pip install ebeamtime

Example

An absolute GDS path is preferred. A relative path such as Path("layouts/device.gds") starts at the directory where you run Python. Add one exposure per layer/datatype pair:

from pathlib import Path

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

# Preferred: an absolute path to the input GDS file.
gds_path = Path("/absolute/path/to/device.gds")

exposures = (
    EbeamLayerExposure(
        config_name="junction",   # Label shown in the report.
        layer=LayerSpec(1, 0),    # Junction GDS layer and datatype.
        dose_uC_cm2=700,          # Dose (µC/cm²).
        beam_current_nA=1,        # Beam current (nA).
    ),
    EbeamLayerExposure(
        config_name="undercut",   # Label shown in the report.
        layer=LayerSpec(2, 0),    # Undercut GDS layer and datatype.
        dose_uC_cm2=100,          # Dose (µC/cm²).
        beam_current_nA=1,        # Beam current (nA).
    ),
)

config = EstimateConfig(
    gds_path=gds_path,            # GDSII file to analyse.
    exposures=exposures,          # Every layer to include in the estimate.
)

report = estimate_gds_write_time(config).report
for layer in report.layers:
    hours = layer.beam_on_s / 3600
    print(f"{layer.config_name}: {layer.beam_on_s:.3f} s ({hours:.6f} h)")
total_hours = report.total_s / 3600
print(f"Total: {report.total_s:.3f} s ({total_hours:.6f} h)")

Names are report labels; LayerSpec(layer, datatype) selects the GDS polygons. Overlapping polygon instances count separately because each is written.

GPU acceleration (optional)

CPU needs no GPU or compiler. NVIDIA users need a compatible GPU and driver, the CUDA Toolkit with nvcc, and a host C++ compiler such as g++.

On WSL, install the NVIDIA display driver on Windows and install only the CUDA Toolkit inside WSL. Do not install a Linux NVIDIA display driver in WSL. See the NVIDIA CUDA on WSL guide.

First-time CUDA setup

ebeamtime-diagnostics
ebeamtime-prepare-cuda --json
ebeamtime-diagnostics

The preparation command detects the toolchain and GPU, compiles the packaged kernel, verifies a known geometry, and stores the result in the user cache. Run it once per Python, toolkit, GPU architecture, or package-source change.

Force CUDA from Python

config = EstimateConfig(
    gds_path=gds_path,            # GDSII file to analyse.
    exposures=exposures,          # Exposure layers to include.
    backend="cuda",               # Force the NVIDIA CUDA backend.
    require_gpu=True,             # Fail clearly instead of using CPU.
)

Force CUDA from the command line

ebeamtime /absolute/path/to/device.gds \
  --exposure 1:0:700:1 \
  --backend cuda \
  --require-gpu

Allow automatic CPU/GPU selection

config = EstimateConfig(
    gds_path=gds_path,            # GDSII file to analyse.
    exposures=exposures,          # Exposure layers to include.
    backend="auto",               # Use a prepared GPU when worthwhile.
)

auto does not perform first-time compilation. It uses a prepared GPU backend for at least 4,096 polygons by default and otherwise uses CPU. Prepared CUDA libraries are normally cached under ~/.cache/ebeamtime/native/.

Apple Metal is available experimentally on Apple Silicon with the Xcode command-line tools.

Licensed under GPL-3.0-only. Development instructions are in CONTRIBUTING.md.