A Digital Micromirror Device, or DMD is an optical semiconductor that is the core of DLP projection technology, and was invented by Dr. Larry Hornbeck and Dr. William E. "Ed" Nelson of Texas Instruments (TI) in 1987.
The DMD project began as the Deformable Mirror Device in 1977, using
micromechanical, analog light modulators. The first analog DMD product
was the TI DMD2000 airline ticket printer that used a DMD instead of a
laser scanner.
A DMD chip has on its surface several hundred thousand microscopic mirrors arranged in a rectangular array which correspond to the pixels
in the image to be displayed. The mirrors can be individually rotated
±10-12°, to an on or off state. In the on state, light from the
projector bulb is reflected into the lens making the pixel appear
bright on the screen. In the off state, the light is directed elsewhere
(usually onto a heatsink), making the pixel appear dark.
To produce greyscales, the mirror is toggled on and off very quickly, and the ratio of on time to off time determines the shade produced (binary pulse-width modulation). Contemporary DMD chips can produce up to 1024 shades of gray. See DLP for discussion of how color images are produced in DMD-based systems.
The mirrors themselves are made out of aluminium and are around 16 micrometres across. Each one is mounted on a yoke which in turn is connected to two support posts by compliant torsion hinges. In this type of hinge, the axle is fixed at both ends and literally twists in the middle. Because of the small scale, hinge fatigue is not a problem and tests have shown that even 1 trillion (1012)
operations do not cause noticeable damage. Tests have also shown that
the hinges cannot be damaged by normal shock and vibration, since it is
absorbed by the DMD superstructure.
Two pairs of electrodes control the position of the mirror by electrostatic
attraction. Each pair has one electrode on each side of the hinge, with
one of the pairs positioned to act on the yoke and the other acting
directly on the mirror. The majority of the time, equal bias charges
are applied to both sides simultaneously. Instead of flipping to a
central position as one might expect, this actually holds the mirror in
its current position. This is because attraction force on the side the
mirror is already tilted towards is greater, since that side is closer
to the electrodes.
To move the mirrors, the required state is first loaded into an SRAM
cell located beneath each pixel, which is also connected to the
electrodes. Once all the SRAM cells have been loaded, the bias voltage
is removed, allowing the charges from the SRAM cell to prevail, moving
the mirror. When the bias is restored, the mirror is once again held in
position, and the next required movement can be loaded into the memory
cell.
The bias system is used because it reduces the voltage levels
required to address the pixels such that they can be driven directly
from the SRAM cell, and also because the bias voltage can be removed at
the same time for the whole chip, so every mirror moves at the same
instant. The advantages of the latter are more accurate timing and a
more filmic moving image.