Photon Shot Noise in EUV Lithography
Photon shot noise in extreme-ultraviolet (EUV) lithography is the random variation in how many EUV photons reach each tiny part of the wafer. Because each printed feature may be formed from a limited number of photons, one location can receive enough exposure to print correctly while a nearby, nominally identical location does not. The result is stochastic defects: random missing, bridged, rough, or otherwise malformed features.
The problem EUV creates
EUV lithography uses light with a wavelength of 13.5 nanometres to project a mask pattern onto a photoresist-coated wafer. A photoresist is a light-sensitive film that is chemically changed by exposure and then selectively removed during development, leaving the pattern that later processing transfers into the wafer.
The difficulty is that EUV photons are energetic but relatively scarce at practical exposure doses. A photon at 13.5 nm carries about 92 electronvolts of energy, yet much of the light is lost in the optical system, the mask, and the resist. A very small printed contact or line therefore depends on a limited population of absorbed photons.
Without this randomness, engineers could treat dose as a smooth, continuous quantity: increase the dose slightly and every feature would change in nearly the same way. With a small photon population, dose has statistical variation from place to place. Two identical features can receive different photon counts simply by chance.
Why the variation is unavoidable
Photon arrivals follow approximately Poisson statistics, a probability model for independent counting events. If a region receives an average of N photons, its typical count variation is about √N. The relative variation is therefore approximately:
relative variation ≈
1 / √N
More photons reduce the relative fluctuation, but EUV dose, feature size, and tool throughput limit how many photons can be used. Smaller features are especially vulnerable because they occupy less area and consequently collect fewer photons.
The photons do not directly draw a perfectly sharp binary image in the resist. An absorbed EUV photon releases energetic electrons, which produce additional lower-energy electrons. These secondary electrons spread the chemical effect beyond the original absorption point. The resist's chemical reactions, diffusion during baking, and development then convert this irregular energy distribution into a boundary.
In a chemically amplified resist, exposure creates an acid or another reaction trigger, and that trigger causes a larger chemical change during a later bake. Amplification makes the resist more sensitive, meaning less EUV dose is needed, but it also introduces more random variables: where the initial reaction occurs, how much trigger is produced, and how far it moves before development.
The final pattern is effectively a threshold decision. If enough of the resist changes at a location, development removes it; if not, it remains. A small random exposure difference near that threshold can flip the result.
What a stochastic defect looks like
For isolated contact holes or vias, the most conspicuous failure is a missing contact: the hole does not open during development, so the later interconnect cannot reach the layer below. A partially opened hole may be too small or irregular to meet its electrical resistance target.
For dense lines and spaces, the same mechanism can create:
- a broken line, where a narrow section fails to print;
- a bridge, where adjacent lines remain connected by unwanted resist;
- local line-edge roughness, meaning short-scale variation along a line boundary; and
- line-width variation, where nominally equal lines print at different widths.
These defects are called stochastic because their exact locations are not predictable from the mask layout alone. They differ from a systematic focus error or dose calibration error, which tends to affect many features in a repeatable pattern. Mask roughness, resist composition, and development fluctuations can add to the randomness, but photon shot noise is the starting fluctuation that makes the limited-photon problem fundamental.
How fabs find the defects
A fab usually detects the problem through several levels of inspection rather than one instrument.
Critical-dimension scanning electron microscopy (CD-SEM) uses an electron beam to measure printed line widths, hole sizes, and edge positions at selected sites. It reveals increased variation and can show malformed features, but it examines sampled locations and is too slow to inspect every feature on a production wafer.
Wafer inspection tools scan large areas optically or with electrons and compare one die, or one repeated pattern, with another. A difference can indicate a missing hole, a bridge, or a broken line. Optical inspection is fast and covers much of the wafer, while electron-beam inspection provides finer detail at lower throughput. Suspect locations are sent to a higher-resolution review tool for classification.
Engineers analyse both the number and the distribution of defects. A wafer map with isolated random failures is evidence of stochastic behaviour; repeating stripes, field-edge effects, or a consistent shift in hole size point more strongly to focus, dose, overlay, or tool-patterning problems. They also measure critical-dimension uniformity, the spread of feature sizes, and line-edge roughness, the spread of boundary positions along a line.
Electrical tests provide a later confirmation. A missing or undersized contact can appear as high resistance or an open circuit, while a bridge can cause a short. These tests are essential for product impact, although they identify consequences after several process steps rather than exposing every physical defect immediately.
How the process is improved
The most direct response is to increase EUV dose. More photons reduce relative shot noise, but higher dose slows the scanner and may change the resist profile. Production therefore balances defect probability against wafer throughput.
Other approaches reduce the effect without simply spending more photons:
- Improve the resist: Formulations can increase EUV absorption, improve the efficiency of the chemical reaction, or reduce unwanted reaction spread. Molecular resists and carefully controlled chemically amplified systems aim for a sharper, more uniform response.
- Control acid and electron movement: The post-exposure bake, quencher concentration, resist thickness, and underlayer are tuned so the reaction spreads enough to form a continuous feature but not enough to blur neighbouring features.
- Optimise the pattern and dose: Computational lithography can adjust the mask pattern and local exposure to compensate for predictable optical effects. This does not remove random photon arrivals, but it keeps more features away from the failure threshold.
- Improve development and process control: Uniform coating, bake temperature, developer concentration, and rinse conditions prevent process variation from adding to the photon-driven variation.
- Use inspection feedback: Defect maps and CD-SEM measurements reveal whether a change reduced missing contacts, roughness, or bridges, allowing the fab to tune dose and materials against measured wafer results.
The central trade-off remains: sensitivity needs fewer photons, while stochastic reliability favours more photons. EUV process development is largely the work of managing that trade-off without sacrificing resolution, throughput, or electrical performance.