E08 · Electron microscopy / specimen physics

Electron Dose, Noise & Damage

Set current, pixel dwell, frame count, pixel size and probe width. Compare an ideal periodic structure with a counted image, its dose budget and retained structure. Reach useful signal-to-noise without destroying the specimen.

Incident electrons per pixel $0$

Expected detected counts $0$

Mean raster dose $0\,\mathrm{e^{-}\,\mathring{A}^{-2}}$

Probe peak pulse dose $0\,\mathrm{e^{-}\,\mathring{A}^{-2}}$

Useful-detail SNR $0$

Retained structure $0\,\mathrm{\%}$

Relative counting noise $0\,\mathrm{\%}$

Temperature rise proxy $0\,\mathrm{K}$

Contamination proxy $0\,\mathrm{nm}$

Carbon threshold comparison $\text{—}$

512-square raster exposure $0\,\mathrm{s}$

Experiment task $\text{—}$

Exact electron counting and uniform-raster areal dose; Poisson image sampling uses a Gaussian approximation above 40 counts. An assumed exponential contrast decay models radiolysis, a carbon-12 threshold gates an illustrative knock-on term, and heating uses an assumed thermal conductance. Not material-specific damage kinetics. Probe peak pulse dose differs from mean raster dose. The scan animation is a slowed presentation of the selected exposure plan.

Start with the essentials

Focus question: When does more exposure stop revealing more detail?
One-sentence intuition: Counting noise falls with exposure, but useful high-frequency contrast can fall faster as structure is damaged. Averaging does not erase accumulated dose.

Core mathematical model

Counting and areal dose

N_e=\frac{I\tau N_f}{e},\quad D=\frac{N_e}{100a^2}\quad(a\text{ in nm})

Current, dwell and frame count enter the same total electron budget. Dose is in electrons per square angstrom.

Counting noise with damage

\sigma_N=\sqrt{N},\quad \mathrm{SNR}_{\mathrm{detail}}=0.3H\,S(D)\sqrt{\eta N_e}

Transfer accounts for Gaussian probe blur and pixel integration of a 5 nm mode; specimen detail survival is modeled separately.

Radiolysis proxy

S_{\mathrm{rad}}(D)=\exp(-D/D_c)

The critical dose is an adjustable assumption. The optimum is a feature- and material-dependent constraint.

Common difficulties

Model boundary

Typical misconceptionA schematic image is a calibrated material prediction.

Better mental modelExact electron counting and uniform-raster areal dose; Poisson image sampling uses a Gaussian approximation above 40 counts. An assumed exponential contrast decay models radiolysis, a carbon-12 threshold gates an illustrative knock-on term, and heating uses an assumed thermal conductance. Not material-specific damage kinetics. Probe peak pulse dose differs from mean raster dose. The scan animation is a slowed presentation of the selected exposure plan.

Run the experiment

01
Predict
Predict whether ten times more frames always improves useful detail. Compare too-few-counts, low-dose and overexposed presets.
What to observe: Counting noise falls with exposure, but useful high-frequency contrast can fall faster as structure is damaged. Averaging does not erase accumulated dose.
02
Tune and check
Reach SNR at least 5, retain at least 75 percent detail, keep dose below 30 percent of the critical dose and the heating proxy below 10 K.
What to observe: The task checks quantitative readouts rather than visual brightness.
03
Explain the limit
Trade current against dwell while holding their product fixed, then increase frame count. Explain which quantities remain fixed and which accumulate.
What to observe: Exact electron counting and uniform-raster areal dose; Poisson image sampling uses a Gaussian approximation above 40 counts. An assumed exponential contrast decay models radiolysis, a carbon-12 threshold gates an illustrative knock-on term, and heating uses an assumed thermal conductance. Not material-specific damage kinetics. Probe peak pulse dose differs from mean raster dose. The scan animation is a slowed presentation of the selected exposure plan.