Digital micromirror device
Routes a configurable binary micromirror pattern into ON and optional OFF orders.
Open in the canvas →In the real world
A DMD is an array of hundreds of thousands of aluminium mirrors, each a few micrometres across, sitting on a CMOS memory cell. Every mirror has exactly two stable positions — tilted one way or the other about its diagonal, typically by ±12° — and is held there electrostatically against mechanical landing posts. Writing a bit to the cell underneath flips it.
That makes the device fundamentally different from a liquid-crystal SLM. An SLM is analogue and works on phase: it retards light and can therefore redirect it. A DMD is binary and works on amplitude: each mirror either sends its light toward the target or throws it away. There is no in-between position.
Because a mirror tilted by θ deflects a beam by 2θ, the two states send light in directions separated by four times the tilt angle:
Grey levels out of a binary device
If each mirror is only ever fully on or fully off, brightness has to come from somewhere else — and it comes from time. The mirror is switched on and off thousands of times per frame, and the fraction of the frame it spends in the ON state sets the perceived brightness. The eye, or any detector slower than the switching, integrates the result. Pulse-width modulation in space's place.
This is why DMDs are fast. A micromirror flips in microseconds, giving binary frame rates in the tens of kilohertz — three or four orders of magnitude quicker than liquid crystal, which has to physically reorient. It also explains the colour-fringing "rainbow effect" some people see in single-chip DLP projectors, where red, green and blue are displayed sequentially rather than together.
What it is good and bad at
Being a mirror rather than a birefringent layer, a DMD is polarization-insensitive and broadband — aluminium reflects from the ultraviolet well into the infrared, so one device works at any wavelength. It has a high fill factor, around 92%, and tolerates far more optical power than liquid crystal. Those properties took it well beyond projectors: maskless photolithography, structured-illumination microscopy, hyperspectral imaging, single-pixel and compressive-sensing cameras, and patterned optogenetic stimulation.
The cost is efficiency. Because it works by discarding light rather than redirecting it, everything in the OFF state is simply thrown away — at 50% duty you lose half the beam, and that light has to be caught by a beam dump, which at high power needs to be a real cooled one. A phase SLM steering the same light into the pattern wastes almost none of it. The periodicity is the other complication: with coherent illumination the array behaves as a blazed grating and splits the beam into diffraction orders, so a laser-illuminated DMD needs its geometry chosen so that the wanted order and the blaze direction coincide.
In OpticalSetup
The device is traced as a binary mirror array. Rays are sorted into ON and OFF by where they land, and each state reflects into its own direction — the ON beam deflected one way from the specular direction, the OFF beam the other:
- Micromirror tilt sets the deflection. The ON and OFF beams emerge separated by exactly four times this angle, so the default 12° puts 48° between them, matching a real device.
- Pattern pitch and ON fraction define the pattern itself as periodic stripes across the aperture — the fraction of each period whose mirrors are ON.
- Show OFF order decides whether the discarded beam is drawn.
That last toggle is worth understanding. It is off by default, so the rejected light simply stops at the device — which is what a setup with a properly dumped OFF path looks like, and keeps a teaching figure uncluttered. Turn it on and the OFF beam is traced to wherever it actually goes, which is the honest picture while you are designing: you can see the 48° separation, confirm nothing downstream is sitting in that path, and put a beam dump there to terminate it.
Sweeping the tilt is the quickest way to see the geometry that makes DLP work. At 6° the two beams leave 24° apart and are awkward to separate; at 20° they are 80° apart and trivially separable, but the device would be harder to build. The real ±12° is the compromise.
Diffraction is not modelled, and for a DMD that is the significant omission: a real array is periodic at roughly 7.6 µm and acts as a blazed grating, so coherent illumination produces a set of diffraction orders that a laser-based design has to be built around. Here reflection is purely geometric and a single beam produces a single ON beam. The pattern is also periodic stripes measured in millimetres of canvas rather than an addressable array of micromirrors, so it cannot display an image, and there is no time dimension — no pulse-width modulation, no grey levels, no switching time, and no colour sequencing. Fill-factor loss, aluminium reflectivity, absorption, and the damage threshold are all absent, so the ON and OFF beams together carry the full incident power.