|
MEMS (Micro Electro-Mechanical
Systems) Technology

In less than
20 years, MEMS (micro electro-mechanical systems) technology
has gone from an interesting academic exercise to an integral
part of many common products. But as with most new technologies,
the practical implementation of MEMS technology has taken a while
to happen. The design challenges involved in designing a successful
MEMS product (the ADXL2O2E) are described in this article by
Harvey Weinberg from Analog Devices.
In early MEMS
systems a multi-chip approach with the sensing element (MEMS
structure) on one chip, and the signal conditioning electronics
on another chip was used. While this approach is simpler from
a process standpoint, it has many disadvantages:

* The overall
silicon area is generally larger.
* Multi
chip modules require additional assembly steps.
* Yield
is generally lower for multi chip modules.
* Larger
signals from the sensor are required to overcome the stray capacitance
of the chip to chip interconnections, and stray fields necessitating
a larger sensor structure.
* Larger
packages are generally required to house the two-chip structure.
Of course,
history teaches us that integration is the most cost effective
and high performance solution. So Analog Devices pursued an integrated
approach to MEMS where the sensor and signal conditioning electronics
are on one chip.

Figure 1
The latest
generation ADXL2O2E is the result of almost a decades worth of
experience building integrated MEMS accelerometers. It is the
world's smallest mass-produced, low g, low cost, integrated MEMS
dual axis accelerometer.

The mechanical
structure of the ADXL2O2E is shown in Figure 1 along with some
key dimensions in Figure 2.

Figure 2
Polysilicon
springs suspend the MEMS structure above the substrate such that
the body of the sensor (also known as the proof mass) can move
in the X and Y axes. Acceleration causes deflection of the proof
mass from its centre position. Around the four sides of the square
proof mass are 32 sets of radial fingers.
These fingers are positioned between plates that are fixed to
the substrate. Each finger and pair of fixed plates make up a
differential capacitor, and the deflection of the proof mass
is determined by measuring the differential capacitance.

This sensing
method has the ability of sensing both dynamic acceleration (i.e.
shock or vibration) and static acceleration (i.e. inclination
or gravity).
The differential
capacitance is measured using synchronous modulation/demodulation
techniques. After amplification, the X and Y axis acceleration
signals each go through a 32KOhm resistor to an output pin (Cx
and Cy) and a duty cycle modulator (the overall architecture
can be seen in the block diagram in Figure 3). The user may limit
the bandwidth, and thereby lower the noise floor, by adding a
capacitor at the Cx and Cy pin.
The output
signals are voltage proportional to acceleration and pulse-width-modulation
(PWM) proportional to acceleration.
Using the PWM outputs, the user can interface the ADXL2O2 directly
to the digital inputs of a microcontroller using a counter to
decode the PWM.

Figure 3
Challenges
in MEMS Design
The mechanical
design of microscopic mechanical systems, even simple systems,
first requires an understanding of the mechanical behaviour of
the various elements used. While the basic rules of mechanical
dynamics are still followed in the miniaturised world, many of
the materials used in these structures are not well mechanically
characterised. For example, most MEMS systems use polysilicon
to build mechanical structures. Polysilicon is a familiar material
in the IC world, and is compatible with IC manufacturing processes.

Until recently,
little work has been done to fully understand polysilicon's mechanical
properties. In addition, many materials mechanical properties
change in the microscopic world. Again, polysilicon is a good
example. In the macro world it is rarely used as a mechanical
element. It is too brittle and fragile to withstand all but small
mechanical deflections. But in the
extremely small movements of MEMS structures (less than a few
pm), it turns out to be an almost ideal material.
The electronic
design of MEMS sensors is very challenging. Most MEMS sensors
(the ADXL2O2E included) mechanical systems are designed to realise
a variable capacitor. Electronics are used to convert the variable
capacitance to a variable voltage or current, amplify, linearise,
and in some cases, temperature compensate the signal. This is
a challenging task as the signals involved are very minute.
In the case of the ADXL2O2E for example, the smallest resolvable
signal is approximately 2OzF and this is on top of a common mode
signal several orders of magnitude greater than that! Of course,
for cost reasons the
electronics must be made as compact as possible at the same time.

The integrated
approach presented further challenges.
Many standard production steps that improve the mechanical structure
degrade the electronics and vice versa. For example, the usual
method for flattening out the Polysilicon mechanical structure
is annealing (where the structure is exposed to controlled high
temperatures). While the annealing process is beneficial to the
mechanical structure, it can degrade or destroy the BiMOS transistors
used in the signal conditioning electronics. So compatible mechanical
and electronic process methods had to be devised.
Another roadblock
for the MEMS designer has been the unavailability of standard
design software. Modern integrated circuits are rarely designed
by hand. Complex CAD and simulation software is used to help
design and optimise the designers concepts.
MEMS design
software is still in its infancy, and most MEMS manufacturers
develop part or all of their CAD and simulation software to suit
their particular needs.
The fabrication
process design challenge is perhaps the greatest one. Techniques
for building three-dimensional MEMS structures had to be devised.
Chemical and trench etching can be used to "cut out"
structures from solid polysilicon, but additional process steps
must be used to remove the material underneath the patterned
polysilicon to allow it to move freely.

Standard plastic
injection molded IC packaging cannot be used because of the moving
parts of the MEMS structure. A cavity of some type must be maintained
around the mobile MEMS structure. So alternative low-cost cavity
packaging was developed.
In addition, this package must also be mechanically stable as
external mechanical stress could result in output changes.
Even mundane
tasks, such as cutting the wafer up into single die, becomes
complicated. In a standard IC the particle residue created by
the sawing process does not effect the IC. In a moving MEMS structure
these particles can ruin a device.
The Users
Challenge
MEMS
sensors, like almost all electronic devices, do not exhibit ideal
behaviour. While most designers have learned how to handle the
non-ideal behaviour of op-amps and transistors, few have learned
the design techniques used to compensate for non-ideal MEMS behaviour.
In most cases, this type of information is not available in textbooks
or courses, as the technology is quite new. So generally designers
must get this type of information from the MEMS manufacturer.
Analog Devices,
for example, maintains a web site with design tools, reference
designs, and dozens of application notes specific to its MEMS
accelerometers to ease the users work.

Conclusion
As
with all new technologies both designers and users of MEMS devices
have a learning curve to overcome. The effort is worthwhile,
as the latest generation MEMS devices high performance and low
cost have enabled innovative new products in dozens of markets.
____________________________________________________
Harvey Weinberg
is an Applications Engineer for Analog Devices Inc.
Micromachined Products Division in Cambridge, Massachusetts.
For more information,
please contact :-
Analog Devices - Tel: +44(0) 1932 266013
|