Why is the Texas Instruments

Digital Micromirror Device™ (DMD™)

so reliable?

CVPINFO\DOUGLASS\PAPER\
myths8.doc
The remarkable Digital Micromirror Device down. Not any more. The development of new
materials, new manufacturing processes and new
There are two remarkable things about the Digital testing programs have enabled us to engineer
Micromirror Device (DMD) from Texas Instru- mechanical products whose performance and
ments. The first is that it works at all: the second reliability vastly surpasses that of their
is that it works so reliably. predecessors.

The Digital Micromirror Device is at the heart of

TI’s Digital Light Processing technology. It
allows images to be projected and displayed
which are brighter, sharper and more realistic than
has previously been possible with alternative
technologies.

Figure 2: A close-up of the DMD array

The DMD: changing the rules

And how many of the mechanical devices with

which we’re familiar (even today) have been the
Figure 1: Simplified schematic of the underlying subject of a development program lasting close to
principle of the Digital Micromirror Device twenty years - like the DMD? How many of them
have been precision-engineered to tolerances of a
Put simply, the DMD is an array of microscopi- micron or less, as the DMD is? (The average
cally small, square mirrors - some half a million or human hair is between sixty and eighty microns
more in a space no larger than a finger nail - each wide). How many of them have been manufactured
of which can be turned on and off thousands of using the same manufacturing techniques that are
times per second. Each mirror corresponds to a used to make the world’s most powerful micropro-
single pixel in the projected or displayed image. cessors and computer memories, as is the case with
the DMD?
Given that the DMD has half a million rapidly
moving parts, it’s easy to believe that it shouldn’t But still, we carry with us preconceptions about
be reliable. After all, it’s a mechanical device as what should and what should not work - and these
much as it’s an electronic one - and don’t preconceptions have given rise to a number of
mechanical things wear out and break down? myths. Like many myths, we tend to believe
them, because they sound plausible - they accord
Yes - and no. Our experience of the mechanical with what we think we know. The DMD changes
world hasn’t always been a positive one. On the the rules about what we think we know.
other hand, mechanical devices today are typically
far more reliable than they were ten - or even five DMD: the questions
- years ago. It wasn’t that long ago that we took it
for granted that a new car engine had to be ‘run So what do we think we know? What are the
in’: we were happy to accept that a mechanical questions we ask ourselves about the DMD?
device was bound to have rough edges, and
needed time for the component parts to be bedded We wonder whether the hinge - which allows the
micromirror to tilt back and forth - will eventually The mirror hinge is manufactured using ‘thin-
break under the strain of constant twisting? film’ technology. Thin films have distinctly
different properties from the general macro-
We ask ourselves whether the micromirrors are scopic concept of bending metal. A thin-film
likely to stick in one position, causing defects in material is said to be more ‘compliant’; in other
the projected image? words, it has less stiffness. Stiffness is the
property of a material that causes the material
We ask whether such microscopically small com- to resist bending. The more the material resists,
ponents aren’t so fragile that they’ll break if the greater the likelihood of its breaking.
handled roughly?
Consider, for example, a bar of aluminum which
We wonder if the enormous speed at which the is repeatedly bent against itself: it will ulti-
micromirrors must switch will, over time, cause mately snap. Few of us have not at some time
them to become damaged and inoperative? bent and re-bent a soft drink can until the metal
has torn. But how many of us have taken a
And we ask ourselves about the effect of dust: sheet of aluminum foil and attempted to cause it
after all, if a human hair could cover five to break by bending it back and forth?
micromirrors, wouldn’t a piece of dust in the DMD
surely cause one or more mirrors to become The important relationship is between the size
inoperative? of the crystals - or ‘grains’ - which comprise the
material of the item and the size of the item in
DMD: the answers question. When a material breaks, it is because
of dislocations caused to the crystal structure
Each of these questions seems entirely reasonable, within it. These dislocations migrate to, and
based as it is on our preconceptions. What follows accumulate at, joints between the grains. This
is an explanation of what’s different about the has the effect of concentrating mechanical
DMD that helps us to understand why the answers stress until the yield point of the material is
to the above questions aren’t what we might exceeded - at which point, breakage occurs.
expect.
In something as microscopically small as a
Question 1: won’t the hinges bend or break? DMD hinge, there is, in effect, no internal
crystal structure - all crystals are at the surface
Each micromirror is hinged, allowing it to rotate of the material. What this means is that the
on its diagonal axis. Given that each mirror will be stresses caused by crystal dislocations are
switched through twenty degrees thousands of relieved immediately on the hinge surface -
times per second, it seems to many people that it before the hinge’s crystalline structure can be
must, sooner or later, break. In fact, “hinge fa- damaged.
tigue” has never been a problem for the Digital
Micromirror Device. To demonstrate this concept, sets of devices
were tested through 1 trillion cycles, well in
excess of the requirement for a ‘normal’ com-
mercial lifetime. No broken hinges were ob-
served.

In this case, the DMD ‘changes the rules’

because of the microscopic physical size of its
mechanical parts: the laws that apply to the
everyday objects in our lives work differently at
this level of miniaturization
Question 2: are the mirrors likely to stick?

When the micromirrors are switched between their

‘on’ and ‘off’ positions, they are held in place by
electrostatic forces. It is true that during early
DMD prototype development, some mirrors
tended to stick to the underlying surface due to
large (in sub-micron technology terms) adhesive
forces. This, in turn, caused the mirrors to fail to
switch.
What would be the cause of such an adhesive
force? There are two phenomena at work. The
first phenomenon is the relatively straightforward Figure 4: The addition of the ‘spring tip’ has further
one in which capillary water condensation will improved the reliability of the Digital Micromirror Device
cause the landing tip and the landing surface to
become ‘stuck’. And third, to minimize the effect of capillary
condensation of water, the DMD is sealed in a dry
To understand better how this happens, consider environment using special hermetic packaging
the case of two glass plates stuck together with designed to ensure that it stays dry throughout its
water - an extreme example of the adhesive force lifetime.
of capillary condensation. By wetting the common
interface of the two plates, a partial vacuum is Thus the forces which would otherwise cause the
produced at this interface due to the surface DMD to stick were eliminated - and to improve
tension of the water on the glass. As a result, the DMD’s operating margin, its ability to “bounce
enormous forces are required to pull the plates back” was dramatically improved.
apart.
Question 3: will the mirrors fall off if the DMD
The second phenomenon, which is probably famil- is treated roughly?
iar to those with an advanced education in physics,
is the phenomenon known as ‘van der Waals Most people find it hard - if not impossible - to
forces’. Van der Waals forces are short range imagine a device in which over half a million
forces which cause materials to become attracted individual moving parts are arrayed in an area
at the molecular level. measuring less than 1.5 sq. cms. What they can
imagine, however, is how fragile such a device
In order to prevent the mirrors from sticking, three must be. Surely it must be true that only the
actions are taken in the design and manufacture of smallest shock or vibration will be enough to cause
the DMD. First, the effect of van der Waals severe dislocation of the micromirrors when
forces is significantly mitigated by the application, subjected to, for example, the robust treatment a
during the fabrication process, of a thin anti-stick projector will typically receive during its working
layer which lowers the surface energy of the life?
contacting parts.
Again, our thinking is conditioned by the phenom-
Second, the mirror was redesigned in 1995 to ena we observe during everyday life. However, the
improve the ability of the mirror to overcome the objects with which we are familiar, and which
remaining forces. The redesign added miniature exhibit such characteristics, have very different
springs to the mirror landing tips, as can be seen in profiles in terms of their modes of vibration.
Figure 4. These springs store energy upon landing
and push the mirror away from the surface upon Why are ‘modes of vibration’ important? Simply,
release. because it’s vibration that causes fracture and
breakage. A simple illustration of this is the cm/second. By comparison, the terminal velocity
classic example of a wine glass shattered by of an oak leaf falling from a tree is of the order of
an opera singer’s voice. The sound waves - at 100 cm/second - which makes the landing of the
the appropriate frequency - cause the glass to DMD mirror a relatively gentle event!
vibrate, and it is the vibration which causes the
glass to break. Knowing this, it’s easy to understand why the
speed with which the mirrors move will not cause
The DMD superstructure has modes of vibra- them to break.
tion with frequencies which are at least two
orders of magnitude above the frequency of Question 5: surely, tiny particles of dust could
vibration generated during normal handling cause mirrors to fail?
and operation (the lowest frequency mode of
the device is about 100kHz: all of the other Yes, they could. So small are the dimensions of the
resonant modes are measured in MHz). As DMD that even the smallest particle of contami-
such, there is virtually no vibration coupling nant could cause one or more micromirrors to
from the environment to the DMD array: the become non-functional. The single most signifi-
DMD has a much higher vibration frequency cant potential problem in fabricating a DMD is the
than can be generated with conventional shock existence of particles in the manufacturing envi-
and vibration sources. ronment (just as it is in the manufacture of micro-
processors and other sophisticated electronic
This is the theory: the fact is that laboratory devices).
testing has shown that the mirrors do not fall
off as a result of shock or vibration. Much to
many peoples’ surprise, dropping a DMD on The DMD is manufactured in exactly the same
the floor does not cause thousands of tiny way, using exactly the same facilities, as the
mirrors to start rattling around behind the world’s most sophisticated semiconductor devices
transparent enclosure! (Texas Instruments is a world leader in this field).
As such, the DMD manufacturing process - which
Question 4: as the mirrors move so rapidly, occurs in a world-class wafer fabrication clean
surely they must eventually become dam- room facility - can take advantage of the huge
aged? amount that has been learned about contaminant
particle elimination.
To achieve the high quality image for which
Digital Light Processing is renowned, the In the manufacture of any sophisticated semicon-
individual micromirrors - each of which corre- ductor device, initial yields are expected to be low.
sponds to a single pixel in the projected image Process improvements are identified and imple-
- must be switched on and off thousands of mented, and, over time, yields increase substan-
times per second. As such, each mirror is tially. It is this “yield curve” which gives rise to
subjected to (relatively) enormous forces. the significant reductions in price that we have
become accustomed to during the lifetime of, for
To understand why the speed at which the example, a memory chip or a microprocessor. We
mirrors switch does not cause them to fail, it are seeing the DMD making its expected progress
must be understood that the size of the mirrors up the yield curve since the effort to reduce the
and the air gap are fantastically small (the impact of particles began in the autumn of 1994.
weight of an individual micromirror is mea- Naturally, the effort is continuing.
sured in millionths of a gram!) and that the
mirrors rotate through only twenty degrees. In Today, Texas Instruments is able to meet the
fact, in 5 microseconds, the mirror’s tip moves demand for DMDs which are 100% defect free at
through just 2 micrometers: as such, the the time of manufacture, thus surpassing a specifi-
average velocity of the micromirror is just 40 cation which is already extraordinarily rigorous
(the specification allows a maximum of one defec- tory testing has thus far shown that it is capable of
tive micromirror/pixel per 100,000 - significantly 100% reliable operation for periods of time mea-
more demanding than is often found elsewhere). sured in decades (and, in fact, testing has not yet
Figure 5: The architecture of the DMD has continued to revealed the point at which failure might be ex-
evolve over time, to the point where its reliability for pected under normal operating conditions).
commercial applications is assured.

The DMD is, of course, hermetically sealed so that

once the manufacturing process is complete, no
contaminants can enter the enclosure.

Testing has put each micromirror through hun-

dreds of billions of cycles - of being switched ‘on’
and ‘off’ - yet there has been no sign of sticking or
of fracture. Moreover, the DMD has been put
through comprehensive shock and vibration tests -
tests which exert far more significant forces on it
than could be expected in normal commercial
operation - and, despite its apparent fragility,
By early 1996, exhaustive tests (120,000 operating
micromirrors haven’t broken off or become mis-
hours, which is approximately equivalent to 120
aligned as a result.
years in the life of a typical portable business
projector) on 150 DMDs revealed only nine - 6% -
There are three key reasons for the DMD’s ability
which exhibited an increase in the number of
to confound our preconceptions.
nonfunctional pixels/mirrors.
The first relates to the microscopic size of the
Detailed examination of each particle which had
mechanical parts. Many of our preconceptions are
caused a failure allowed its source to be identified
based on our experience with objects that are
and eliminated. This effort has seen particulate
readily observed by the naked eye. The micro-
contamination improved by over 80%.
scopic size of the DMDs mechanical parts is
therefore a part of the explanation for the way in
The very small possibility that a microscopic
which the DMD apparently contradicts those
particle of dust was sealed into the DMD package
truths which we hold to be self-evident.
during manufacture is now the only potential factor
which might affect the operation of the DMD
The second reason is that the DMD is the first
during its lifetime - all others having been progres-
commercially produced
sively eliminated during its development. The
mechanical technology based upon conventional
Digital Micromirror Device has now reached a
semiconductor techniques. As such, DMD tech-
level of reliability which is the envy of competing
nology derives significant benefit from building on
technologies.
the past experience and the current knowledge
base of the semiconductor industry.
The DMD: confounding our preconceptions
Finally, while Digital Light Processing products
So why does the reliability of the DMD appear to
featuring the Digital Micromirror Device only
defy and contradict everything which we hold to be
came to market during 1996, the DMD itself had
true about mechanical devices? Intuitively, it
been under development for twenty years in TI’s
should not work: the fact is, though, that labora-
 laboratories. Much of that development effort was
spent in progressively refining the design, architec-
ture and manufacture of the DMD such that its
reliability to commercial standards could be as-
sured.

The Digital Micromirror Device is an astonishing

phenomenon. That it works, and works reliably, is
almost beyond belief. It challenges our preconcep-
tions about what should and should not be achiev-
able in much the same way that seeing moving
pictures did a hundred years ago, that seeing
television pictures did fifty years ago and that
seeing television pictures of a man walking on the
moon did twenty-five years ago. The DMD defies
- and redefines - our understanding of what is
possible.

For more information about Digital Light Processing

by Texas Instruments, visit our World Wide Web site
at http://www.ti.com/dlp. Alternatively, in the USA,
call 1-888-DLP-BY-TI xtn 2930. In Europe, call the
TI DLP Hotline on +44-1604-663066, email
[email protected] or fax +44-1604-663099.