Surface micromachining is a technique for building electromechanical structures in silicon. Combined with onboard signal conditioning circuits, complete electromechanical systems can be economically built on a single piece of silicon. The first commercially successful applications for surface micromachined sensors were accelerometers for automotive airbags. Since then, work has diverged into a number of different directions. New development work is focused on three areas: improved accelerometer performance, greater integration, and new functions. Two low g accelerometers have been developed, one with the capability to resolve 5 milli-g signals and a second which implements a digital output using a sigma delta loop. Much work is focused on new functions and recent developments including surface micromachined rate gyros.

What is surface micromachining?

Structures are made out of polysilicon and use a sacrificial oxide of SiO₂ as a 'mold'. Standard integrated circuit photolithographic techniques are used to deposit a patterned layer of SiO₂. This is followed by a patterned layer of structural polysilicon. The structure is then etched to remove the sacrificial layer leaving a free standing polysilicon structure. The structure is typically 1.6uM above the surface of the wafer, with lateral feature sizes on the same order. Because standard integrated circuit techniques are used, the process integrates well with a standard wafer fabrication process. This allows the consistent and repeatable production of large quantities of devices at low cost.

What makes surface micromachining so versatile?

The small feature size of micromachined structures makes it economical to place circuits and sensors on the same die. This contrasts to other micromachined technologies, such as bulk micromachining, which are fundamentally size limited. Bulk micromachining is the process of choice for building pressure sensors. Diaphragms may be built easily and the physical pressure is applied to the backside of the silicon, easing the problem of connecting the sensor to the outside world. However, in motion sensing and resonating structures, more complicated structures are desired and bulk micromachining runs into limits. Bulk micromachining uses an anisotropic etch to cut through a silicon or quartz wafer to create structures. The feature size of the structures are set by the thickness of the wafer and the angle of the etch. Processes such as wafer bonding can be used to reduce wafer thickness, but even with this advancement, accelerometer sensor dimensions typically reach 10mm₂, while typical surface micromachined structures are 10 times smaller (<1mm₂). To achieve integration, a bulk sensor will typically need to be combined with a separate integrated circuit for signal conditioning and then assembled as a hybrid. Surface micromachining can implement a sensor, complete with all of the necessary signal conditioning in the same size as a bulk sensor alone. The development of a 1µm CMOS process known as iMEMS^® will allow even greater feats of integration that will include microprocessors and EPROM on a single chip with a micromachined sensor.

The shapes and sizes of the structures are limited only by the imagination. The position of the mechanical structure above the surface of the wafer enables sensors that move in all three axes, X, Y, and Z. Typical building blocks of sensors are tensile and non-tensile springs, differential capacitance sensing cells, and electrostatic drive plates. Motion sensing is usually implemented with a differential capacitance structure, which has the advantage that the change in capacitance for a given displacement is 1st order linear for small deflections, simplifying the signal conditioning task. Differential capacitive plates may work as an actuator by using the plates to impart an electrostatic force, causing a flexible structure to move back and forth, thus motion of a structure can be both sensed and iduced. The ability to sense the sometimes tiny changes in capacitance that result from moving silicon is simplified by the presence of the on board signal conditioning circuitry, which reduces parasitic capacitance and interference from noise.

Using the fundamental building blocks of springs, motion detectors and frce inducers, many electromechanical systems can be implemented in minature. In the first commercially successful device, a ±50g electromechanical servo accelerometer, normally built with metal springs, hinges, and electromagnets was implemented in silicon. Since then, a multitude of demonstration projects have been spawned that include X, Y, and Z accelerometers, rate gyros, flow meters, electromechanical filters, resonant accelerometers and pressure sensors, and electromechanical relays. One of the interesting things about these sensors is that they are not just laboratory curiosities. These have been fabricated on a proven wafer fabrication process and could be built in high volume at reasonable prices.

Demonstration of a ±5 g Force-Balance Accelerometer with 500mg/Hz Noise Floor

Several new innovations in micromachining have been combined to make a commercially available accelerometer for low g applications. Several design challenges presented themselves in this device. The largest problems were to decrease susceptibility to packaging stress and, second, to improve signal to noise ratios to allow the resolution of much smaller accelerations. The key to both problems lies in the design of the sensor beam. Figure 1 shows the sensor beam of a ±50 g airbag sensor compared to a new beam for the ±5 g sensor. The ±50 g springs are in tension and are pulling on the four anchor points. While this sensor is stiff and robust, it is also sensitive to mechanical stress imparted to the die from the package and die mount. A new sensor beam used on a ±5 g part uses a 'folded beam' structure. This structure is anchored at only two points; the wrap around structure relieves the tension of the polysilicon, making the beam much less sensitive to package stress. The compliance of the beam, i.e. the deflection of the beam for a given g force applied, is increased and delivers a greater change in capacitance per g and thus a higher signal to noise ratio. This makes possible an accelerometer with the capability to measure milli-g level signals, useful for tilt measurement and some inertial sensing applications. The compliance of the beam, while solving some problems, creates others. Shock survival is reduced from the 50 g version, from 2000 to 1000 g. Performance has however been enhanced by the addition of beam stops that limits the travel of the beam to a safe distance.

In a completely different approach, a beam was constructed to be sensitive to deflections in the Z axis (perpendicular to the plane of the wafer). One of the difficulties of this design is the presence of only 2 layers of polysilicon for force and sense plates preventing the use of a differential capacitance structure. This design compensates for the missing upper plate by using electrostatic levitation to provide a pull up force of 6g/V. The beam is rebalanced to the center position by the electronics, allowing the chip to sense bipolar acceleration signals. A sigma delta force balance loop is used to read the chip deflection and provide a force balance signal to return the beam to the center position, and provides a pulse density output proportional to acceleration.

This design demonstrated the ability to make a Z axis, and hence the ability to measure 3 axes of acceleration on one chip. The use of a digital output scheme simplifies the post processing of the signal enabling A/D conversion with the use of a simple counter.

ARPA Project Chip

Analog Devices has made its surface micromachining technology available for use by outside designers. Project chips are run on a periodic basis. The first such project chip, done in conjunction with Berkeley University, had 20 different chips implementing a variety of different applications, including X, Y, and Z accelerometers, rate sensors, flow sensors, filters, resonators and clock chips.

In the near future, we will see many of these new designs commercialized. The availability of proven functional building blocks will eventually reduce the cost and time of development, and allow complicated integrated functions to be designed in short periods of time. The future will see acceleration, pressure, temperature, and magnetic sensing all integrated onto a single chip.