Anti-Reflection Coating Design: The Quarter-Wave Condition

How single-layer anti-reflection coatings work and how to design one: the quarter-wave thickness condition, the ideal-index rule, worked examples on glass and silicon, and when a single layer isn't enough.

Published July 26, 2026Updated October 5, 20268 min read

Scope

This article covers the design of single-layer anti-reflection (AR) coatings at normal incidence: where the reflection comes from, the two conditions a quarter-wave layer must satisfy, the exact formulas, and two worked examples (a visible-band coating on glass and a 1550 nm coating on silicon). It ends with the limits of a single layer and a map of what lies beyond (V-coats, broadband stacks, laser-facet coatings). A companion AR coating thickness calculator computes the numbers and plots the reflectance spectrum for any substrate/film pair.

Fresnel reflection at an uncoated surface

Whenever light crosses a boundary between two refractive indices, part of it reflects. At normal incidence the reflected fraction follows from the Fresnel equations:

R = {(\frac{n_{2} - n_{1}}{n_{2} + n_{1}})}^{⁣ 2}

For an air–glass interface ( $n = 1.52$ ) that is 4.3% per surface: annoying in a camera lens, fatal in a 20-element optical system. For semiconductors the numbers turn brutal:

Interface (from air)	Index	Uncoated $R$
BK7 glass (visible)	1.52	4.3%
Fused silica (1550 nm)	1.44	3.3%
InP (1550 nm)	3.17	27%
GaAs (1550 nm)	3.37	29%
Silicon (1550 nm)	3.48	31%

A bare silicon photonic chip facet throws away nearly a third of the light at every crossing, before any mode-mismatch loss is counted.

How a quarter-wave layer cancels reflection

Add a thin transparent film between the two media and there are now two reflected waves: one from the air–film interface, one from the film–substrate interface. Anti-reflection coating is interference engineering: arrange for those two reflections to cancel.

Cancellation needs two things simultaneously:

The phase condition. The second reflection travels an extra round trip through the film. If the film's optical thickness is a quarter wavelength,

n_{1} t = \frac{λ_{0}}{4} ⟹ t = \frac{λ_{0}}{4 n_{1}},

the round trip adds half a wavelength of path: the two reflected waves emerge 180° out of phase and interfere destructively. (Both reflections here occur going from lower to higher index, so both pick up the same π phase flip on reflection, and only the propagation phase separates them.)

The amplitude condition. Destructive interference only nulls if the two reflections have equal amplitude. Working through the Fresnel coefficients, equal amplitudes require

n_{1} = \sqrt{n_{0} n_{s}},

the geometric mean of the surrounding indices. With both conditions met, reflectance at $λ_{0}$ is exactly zero. With the phase condition met but the index imperfect (the usual situation, since films come in the indices nature provides), the residual reflectance of a quarter-wave layer is

R_{\min} = {(\frac{n_{0} n_{s} - n_{1}^{2}}{n_{0} n_{s} + n_{1}^{2}})}^{⁣ 2} .

These three formulas are the entire single-layer design method.

Worked example 1: MgF₂ on glass

Design target: BK7 camera-lens surface, centered at $λ_{0} = 550$ nm.

The ideal film index would be $\sqrt{1.0 \times 1.52} = 1.233$ . No robust, depositable dielectric has an index that low. That is why the classic answer is magnesium fluoride at $n_{1} = 1.38$ , the lowest-index hard coating in common use.

Thickness: $t = 550 / (4 \times 1.38) = 99.6$ nm. Residual reflectance:

R_{\min} = {(\frac{1.52 - {1.38}^{2}}{1.52 + {1.38}^{2}})}^{⁣ 2} \approx 1.3 % .

One 100 nm layer cuts 4.3% to 1.3%. This is the ubiquitous single-layer MgF₂ coating on every mid-grade lens, and the reason their reflections look magenta (the null is centered in the green; red and blue leak slightly more).

Worked example 2: Si₃N₄ on silicon

Design target: silicon facet or photodiode window at $λ_{0} = 1550$ nm.

Ideal index: $\sqrt{3.48} = 1.87$ . Here the materials cooperate: silicon nitride sits at $n_{1} \approx 1.99$ , close to ideal. Thickness: $t = 1550 / (4 \times 1.99) = 195$ nm. Residual:

R_{\min} = {(\frac{3.48 - 3.96}{3.48 + 3.96})}^{⁣ 2} \approx 0.4 % .

From 31% to 0.4% with one PECVD layer: a reduction by a factor of about 74, and why nitride AR layers appear on photodiodes, solar cells (at their design wavelengths), and chip facets as the default first move. SiO₂ at $n = 1.45$ , for comparison, only reaches ~6% on silicon: too far below the ideal index. The rule of thumb falls out of the math. High-index substrates are easy to AR-coat (plenty of films sit near their geometric mean), while low-index glasses are hard.

Bandwidth, angle of incidence and real films

A quarter-wave null is exactly that: a null at one wavelength. Reflectance grows quadratically as $λ$ moves off design; the calculator plots the V-shaped spectrum. A single layer typically holds "good" performance over ±10–15% of $λ_{0}$ , comfortably covering a telecom band but not the whole visible range.

Off-normal incidence shifts the null blue (the phase thickness scales with the cosine of the internal angle) and splits s- and p-polarization performance. Coatings for 45° fold mirrors and steep lens surfaces are designed at angle, not adapted afterward.

And the model here assumes lossless, dispersion-free indices. Real films have both, plus density and stoichiometry that depend on the deposition process. Production coatings are optimized numerically against measured film data. The quarter-wave design remains the honest starting point and the sanity check on any vendor curve.

When one layer isn't enough

Two layers (V-coat). A high/low pair (e.g., Ta₂O₅ + MgF₂) adds enough degrees of freedom to reach a true zero at one wavelength on common substrates. It is the usual design for laser-line AR coatings.

Broadband stacks (BBAR). Four to eight alternating layers flatten reflectance below ~0.5% across an octave: camera optics, eyeglasses, display glass.

Laser-facet coatings. External-cavity lasers, SOAs, and gain chips need facet reflectivities of 10⁻⁴ or lower, down to 10⁻⁵ for high-gain SOAs, beyond what thickness control alone delivers. These combine multilayer designs with angled facets and mode-matched measurement, a specialty unto itself.

Textured / gradient surfaces. Moth-eye structures and index-graded layers sidestep interference entirely by removing the abrupt boundary. They are broadband and angle-tolerant, at the price of fabrication complexity.

AR coating thickness for common films

The quarter-wave rule $t = λ / 4 n$ evaluated for common cases:

Film	n	Design λ	Thickness
MgF₂ on glass	1.38	550 nm	99.6 nm
MgF₂ on glass	1.38	1064 nm	192.8 nm
SiO₂ on silicon	1.45	1550 nm	267.2 nm
Si₃N₄ on silicon	1.99	1550 nm	194.7 nm

The AR coating thickness calculator evaluates the rule, the ideal index and the residual reflectance for any film, substrate, and wavelength, and the Thin-Film Reflectance Calculator extends it to two-layer and multilayer stacks at any angle of incidence.

AR versus HR coatings

An anti-reflection coating and a high-reflectivity mirror coating are built from the same quarter-wave layers; they differ only in the phase relationship chosen between the partial reflections. A single low-index quarter-wave layer places its two interface reflections half a wave apart so they cancel. Stack quarter-wave pairs of alternating high and low index and every interface reflection instead arrives in phase, so the reflections add: a distributed Bragg reflector. Ten to twelve pairs of Ta₂O₅ and SiO₂ reflect more than 99.9%; semiconductor mirrors, whose layers differ less in index, need 25 or more pairs of GaAs and AlAs for the same figure. Laser mirrors, DBR structures and fiber Bragg gratings all work this way, the last with an index step of only about 10⁻⁴ and many thousands of periods. AR and HR coatings are therefore opposite devices built from the same parts: one suppresses reflection at a surface meant to transmit, the other maximizes it at a surface meant to reflect.

The AR coating glossary entry has the one-paragraph version; the Fresnel equations entry covers the reflection math this builds on; catastrophic optical damage explains one reason laser facets get coated for survival, not just transmission. And the AR coating thickness calculator runs every formula in this article interactively.

Common questions

What is an anti-reflection coating?

An anti-reflection (AR) coating is a thin transparent film, or a stack of films, on an optical surface, chosen so that the reflections from its interfaces interfere destructively and remove most of the surface's Fresnel reflection. A single MgF₂ layer lowers the 4.3% reflectance of an uncoated BK7 surface to about 1.3% at its design wavelength.

How does an anti-reflection coating work?

The film produces two reflected waves, one at its outer surface and one at its boundary with the substrate. A film with a quarter-wave optical thickness returns the second wave half a wavelength behind the first, and when its index is the geometric mean of the indices on either side the two waves also have equal amplitude and cancel completely.

How thick should an AR coating be?

A single-layer coating is a quarter wavelength thick inside the film: $t = λ_{0} / (4 n_{1})$ for design wavelength $λ_{0}$ and film index $n_{1}$ . MgF₂ ( $n = 1.38$ ) needs 99.6 nm at 550 nm and 192.8 nm at 1064 nm; silicon nitride ( $n = 1.99$ ) on silicon needs 194.7 nm at 1550 nm.

What refractive index should an AR coating have?

The ideal single-layer index is the geometric mean of the indices on either side, $\sqrt{n_{0} n_{s}}$ : 1.233 for glass in air and 1.87 for silicon in air. No robust coating material goes as low as 1.233, so the classic glass coating is MgF₂ at 1.38, which leaves about 1.3%, while silicon nitride at 1.99 lowers the reflectance of silicon from 31% to about 0.4%.

When is a single-layer AR coating not enough?

A single layer has its reflectance minimum at one wavelength, performs well over roughly ±10–15% of it, and on glass cannot reach zero with available materials. Two-layer V-coats reach zero at one wavelength on common substrates, broadband stacks of four to eight layers hold reflectance below about 0.5% across an octave, and laser facets that need 10⁻⁴ or less combine multilayer designs with angled facets.