A hot mirror is a filter that reflects infrared and transmits visible light. It exists because most light sources produce more heat than useful illumination, and because the components downstream — colour wheels, LCD panels, DMD chips, lenses — are considerably more sensitive to heat than to light.

The applications have changed, and the specifications have changed with them.
From lamp houses to LED engines
In lamp-based projection, the hot mirror sat in a converging beam behind a reflector, handling a large angular spread and a substantial IR load. The conventional design used a multilayer dielectric stack on borosilicate, with the edge placed around 700–750 nm and transmission optimised across the visible band.
LED and laser-phosphor engines changed the geometry. The source is smaller, the etendue is tighter, and the IR content is different — a phosphor wheel emits a broad spectrum that includes a near-IR tail rather than the concentrated IR of a xenon arc. The practical result is that hot mirrors are now often specified together with a colour-separation function, which pushes designs toward steeper edges and tighter tolerances on the blue and red ends.
The wafer inspection case is harder
Semiconductor inspection and metrology tools use shortpass filters to separate illumination from detection, and to keep pump light out of sensitive cameras. The constraints are unusual: high laser power at the illumination wavelength, a need for very deep blocking at the pump line, and transmitted wavefront quality that affects the measurement itself.
Here the thermal question is not "how much heat reaches the sample" but "how much heat the filter absorbs". A coating with even modest absorption at several watts of illumination will develop a thermal gradient, which produces a shift in the edge position. For a filter with a 10 nm transition, that shift is measurable in the data.
Substrate choice carries most of the answer
Borosilicate is cheap, stable and adequate for projection. Fused silica costs more but has lower absorption and much better thermal shock behaviour, which matters when the filter is close to a high-power source and cycling. For large apertures — inspection objectives can use 100 mm or larger filters — the coating stress also becomes a flatness issue, and stress compensation in the design becomes part of the specification.
Where designs are going
Two directions are visible. First, cold mirrors are increasingly used as a complement rather than an alternative: reflect the visible, transmit the IR to a beam dump, and keep the heat out of the imaging path entirely. Second, shortpass designs are being combined with band-blocking layers so that a single component handles edge separation, pump rejection and colour trimming, which reduces the number of surfaces and the associated loss.
Both directions increase layer count, which increases deposition time and stress. That is the trade: fewer components, more demanding coatings.
Specification points that matter
- Illumination wavelength range, power density, and beam geometry at the filter.
- Required transmission across the visible, measured, with the AR coating on the exit surface included.
- Blocking range and depth at the pump or IR band.
- Angle of incidence range, and whether the beam is converging.
- Substrate, thickness, transmitted wavefront, and flatness after coating.
- Thermal environment: ambient, cycling, and proximity to the source.
A hot mirror is a thermal component that happens to be optical. Specifying it as though it were only a filter is how heat ends up in the wrong place.